1

PROGRAMMATIC
HEADER
BIDDING
IN CONTEXT

David Bauckmann
A practical textbook for publishers on maximizing revenue
2

© David Bauckmann, 2022

Cover and design © Programmatic Media s.r.o., 2022
www.HeaderBiddingBook.com
ISBN 978-80-11-01389-9
 3

Acknowledgements
This book could not have been written I have also had great help from peo-
without the support of my colleagues ple in the programming field who have
and friends in my professional commu- read this book in various stages of
nity. completion and provided invaluable
comments and feedback. Thanks to
First and foremost, I must thank my
David Voráček and Lukáš Šmol (both
colleagues in the publishing and mon-
cpex.cz), Matěj Kolarovky (GroupM,
etization team at Impression Media,
IAM Slovakia), Petr Kolář and Tereza
who have worked tirelessly with me
Tůmová (both Spir.cz), Michal Marek
for several years to document new
(Adform) and Eugen Cirak.
trends and practices in the field. It
can be very challenging at times, and And a big thanks to Martin Picek
I appreciate that they have the en- (GroupM) for his support and great
ergy to constantly innovate and seek marketing advice without which this
out new practices. Thanks to Veronika book would be much more boring.
Buriánková, Petronella Ördögh, Marcela
And mainly thanks to my family for
Hofmanová, Mark Jeníček and Martina
their support.
Ševčíková. You are a great team with
great know-how that complements one
another perfectly.
Furthermore, I would like to thank
Roman Strolejda, CEO at Impression
Media, who constantly supports us and
gives the whole teamspace to find new
solutions. Thanks also to other col-
leagues who have read this book and
provided feedback, especially Martin
Holeček and Marcela Pecková.
Special thanks to Beatric Trojan
(beartcz.art) for the comprehensive
illustrations and graphic design. I know
it is not easy for creatives to do tech-
nical illustrations and endure several
rounds of comments and corrections.
4

Contents

Acknowledgements 3
Foreword 10
For whom the book is intended 11
What you can and can’t find here 12
Part One
The history and technology that header bidding builds on 13
Chapter 0: Programmatic buying 14
RTB 14
Technology 15
Demand partners 17
Conclusion of the chapter: Programmatic buying 17
Chapter 1: History of header bidding 18
Mass expansion of header bidding 20
Financial benefits of header bidding for publishers 21
Header Bidding as an answer to Google
Dynamic Allocation 23
Conclusion of the chapter: History of header bidding 23
Chapter 2: Waterfall 24
History and reason for the creation of the waterfall 24
Functioning of the waterfall 26
Advantages and disadvantages of waterfall 27
Waterfall integration options 28
Passback 29
Header bidding versus waterfall 32
The future of the waterfall 32
Conclusion of the chapter: Waterfall 33
Chapter 3: Types of auctions by first vs second price 34
Floor price 34
Second price auction 35
First price auction 36
The difference between First price
and Second price auctions 38
Problems of the second price model 38
Problems of the first price model 39
Hard and soft floor price 40
Bid shading 41
 5

Floor price setting strategy 43

Conclusion of the chapter:
Types of auctions by first vs second price 43
Part Two
Header bidding solutions 45
Chapter 4: Header bidding 46
Principle 46
Client-side vs Server-side header bidding 48
Client-side header bidding (C2S) 48
Server-side header bidding (S2S) 51
Difference between C2S and S2S –
advantages and disadvantages 53
The rise of S2S 55
Hybrid header bidding 57
Google’s solution 57
Open Bidding 57
Advantages and disadvantages of Open Bidding 58
Dynamic Allocation 59
Amazon’s solution 59
Unified Ad Marketplace (UAM) 60
Transparent Ad Marketplace (TAM) 60
The difference between UAM a TAM 61
Video header bidding 61
Types of video ads 62
How the auction works 63
Solution for AMP sites and Real-time configuration (RTC) 65
In-app bidding 65
Setting the auction type by price 69
Floor price settings 71
Floor price at the level of the demand partner 71
Floor price at header bidding level 72
Floor price at ad server level 72
Rule Priorities 72
Post-bid 73
The post-bid principle 73
6

Advantages and disadvantages of post-bid 74

Comparison of post-bid and header bidding 75
Footer bidding 76
Benefits of header bidding for publishers 78
Benefits of header bidding for advertisers 80
What else the advertiser should know 80
Chapter conclusion: Header bidding 81
Chapter 5: Wrapper 82
Types of wrappers 82
Most used wrappers 83
Main functions of the wrapper 85
Common timeout 86
Asynchronous functionality 86
Connection to the ad server 86
Easy management of demand partners 86
Transparency and independent monitoring 87
Price adjustment of bids 87
Connection of other functions 87
Floor price definition 87
Special rules for displaying ads 88
The main disadvantages of wrappers 88
Installation and setup 88
Upgrades 88
Barriers between technologies 89
Response time control 89
Involvement of demand partners 90
Conclusion of the chapter: Wrapper 95
Chapter 6: Statistics related to header bidding 96
Bids data 97
Bid request 97
Bid response 97
Timeout rate 98
Bid rate 99
Impression 99
Win rate 100
 7

Fill rate 100

Revenue 100
eCPM 100
Visible impressions 101
Viewability rate 101
Clicks 101
Discrepancy report 102
Bid Latency 102
Evaluating statistics 104
Conclusion of the chapter: statistics related to header bidding 105
Chapter 7: Ad server 106
Sales strategy 106
Ad server 107
eCPM 107
First look options 107
How the ad server works 108
Ad server settings related to header bidding 109
Setting price levels 109
Setting other parameters in the ad server 110
Setting up multiple header bidding solutions 111
How to connect an ad server to header bidding 112
Ad server takes priority 112
Header bidding takes priority 112
Hybrid engagement 112
The problem of set priorities 112
Ad server as an auction participant 114
Why it is good to have a direct sales and ad server 116
Conclusion of the chapter: Ad server 117
Part Three
Areas with a strong relationship to header bidding
Chapter 8: Cookies and other user ID solutions 120
What are Cookies? 122
How cookies are classified 123
Why third-party cookies will disappear 123
Impact of cookies on publishers’ revenues 124
8

Cookie Matching (Cookie Syncing) 125

Alternative solution to Cookies – User ID 127
Deterministic and probabilistic matching 128
SharedID (PubCommonID) 128
Unified ID 2.0 from The TradeDesk (UID 2.0) 129
Universal ID from ID5 129
Interoperability 129
Effect of Cookie and user ID on header bidding 130
Google Privacy Sandbox 131
Relationship of user identity to header bidding 132
Chapter conclusion: Cookies and other solutions for user ID 133
Chapter 9: Protecting user rights 134
General Data Protection Regulation (GDPR) 136
Content of the GDPR 136
What preceded the GDPR 137
Opt-in versus Opt-out 137
Relationship of GDPR to header bidding 139
Transparence and Consent Framework (TCF) 140
Consent Management Platform (CMP) 142
Checking the settings 144
CMP and the world beyond header bidding 146
Chapter conclusion: Protecting user rights 147
Chapter 10: Supply Path Optimization and Demand
Path Optimization 148
The principle of SPO 149
The SPO procedures 153
Supply chain 154
Transparency projects 155
Ads.txt 155
Sellers.json and supply chain 159
Ads.cert 161
Tools for supply chain control 161
The Demand Path Optimization (DPO) 161
Conclusion of the chapter: Supply Path Optimization and
Demand Path Optimization 163
 9

Chapter 11: Page load speed 164

Latency 164
Studies on speed 165
How to increase the speed of ads 167
How to increase the speed of the whole page 169
Tools for analyzing ad loading speed 170
Google SpeedPage Insight 170
ChromeDevTools 170
Headerbid Expert 172
Conclusion of the chapter: Page load speed 173
Chapter 12: Viewability 174
Viewability is important 174
Above-the-fold and Below-the-fold 176
How to increase ad visibility 177
Lazy load 178
Advantages of lazy load 179
Possible complications of lazy load 179
General uses of lazy load 180
Sticky ads 181
Refreshing ads 181
Bid caching 182
Combination of functions in header bidding 183
Conclusion of the chapter: Visibility 183
Conclusion 184
Used terms 186
Sources 194
10

Foreword
Header bidding is a method of conduct- ed and its new version 3 will no longer
ing online auctions in a programmatic be backward compatible with previous
environment that provides all bidders versions.
with the same conditions and time for The integration of new technologies
the auction. such as CMP tools (full discussion in
This approach has brought order and Chapter 9, page 142) and user identity
transparency to the previously grey recognition tools (also in preparation
area of programmatic buying and has for the end of 3rd party cookies) is also
strongly contributed to the massive important to understand (full discus-
expansion of programmatic buying of sion in Chapter 8, page 120).
advertising space. The complexity of the header bidding
Header bidding is now massively implementation issue is interestingly
widespread and is essentially an in- presented in the research of Admon-
dispensable tool for publishers, giving ster.com1, where the publishers them-
them flexibility and better monetiza- selves define the main challenges of
tion options. However, at the time of header bidding and the issue of page
this book’s publication (2022), the rate latency (full discussion in Chapter 11,
of adoption of header bidding in the page 164).
Western world is already declining. This And of course, we also see a differ-
is due to the fact that many publishers ence of up to a few years between
have already integrated header bidding programmatically developed countries
into their marketing practices, but that compared to other regions where there
should not be the end of the process. are still many publishers without any
The question is not the binary if it is in- header bidding solution.
tegrated, but the qualitative how. Many
publishers use header bidding in a Header bidding has become a key
completely inappropriate way and have technology for publishers, but it is still
no idea how much more they could clouded in the shadows. The market
gain by utilizing it more effectively. lacks enough experts with comprehen-
sive knowledge of the field and relat-
The issue of the path between adver- ed spheres and many publishers are
tiser and publisher (supply chain) is integrating header bidding into their ad
evolving intensively (full discussion in server inappropriately (full discussion
Chapter 10, page 148), to the point that in Chapter 7, page 106).
the RTB protocol has been reconstruct-

https://www.admonsters.com/playbook/admonsters-playbook-header-evolution/
1
 11

For whom the book is and especially in communicating with

people in the publishing industry. This
intended communication is important to get the
This book is primarily aimed at pub- maximum effect for your advertisers.
lishing professionals looking after sales In particular, the chapters in the third
systems. It is not purely for yield man- part of the book are of explicit interest
agers, but also for ad server adminis- to the advertiser side.
trators, developers, back office people– In this book, we assume that the read-
anyone involved in the technical setup er already has a basic understanding of
and interconnection of all the systems programmatic advertising and RTB and
the publisher uses. is no stranger to terms like SSP, DSP,
However, the third part of the book ad server, yield, and others (yet we
is devoted to related disciplines that describe them at least in basic terms
people working on the advertiser side on page 14).
should also be familiar with. I believe
this third part will therefore find a wid-
er audience.
In some ways I hope it will also be
useful to business people, and not only
to people taking care of programmatic
deals, but also to their managers. They
may not know the technical details, but
understanding the whole environment
and the interconnected topics will give
them useful insight.
Simply put: if you are entering a pub-
lishing environment and will have any-
thing to do with selling advertising, you
will find a wealth of useful information
in this publication.
But even if you’re on the advertiser
side, you’ll find this book helpful. Know-
ing the principles of header bidding will
help you in setting up your campaigns
12

What you can and can’t Each chapter, and the book as a whole,
are structured similarly, starting with
find here the basics then moving on to advanced
The aim of this book is to summarize points and tips for experts. Occasion-
important know-how for online publish- ally, you may not understand some of
ers and to point them in the direction the important points of a chapter until
for further education. Therefore, the the end, so I recommend reading it
book is interspersed with many refer- multiple times as needed. Similarly,
ences and resources that can be found the chapters are interrelated and it
and clicked on in the footer of the is advisable to skip occasionally from
online text. At the end of the book, we chapter to chapter to understand the
present them in more detail including interconnections. In this sense, it is
a QR code to make it easier to load the more of a textbook than a book that
links. should be read linearly.
We try to present in one place basic Header bidding cannot be thought of as
information about the whole topic of a stand-alone technology, but as part
header bidding, which will be useful to of a larger whole which is also con-
everyone, regardless of what technolo- stantly evolving. If you already have a
gy they decide to use. Often examples basic understanding of header bidding,
of specific technologies such as Google, then the first two parts of the book will
Amazon, Prebid and others are given probably not be of much interest to
to show how specific things work. The you. But the third part, devoted to re-
goal is not to highlight any particular lated technologies and practices, might
technology, but to show different ap- give you new insights and ideas.
proaches.
You won’t find any pieces of code for
developers here, nor specific advice on
how to set up an ad server or header
bidding. We won’t show you the best
possible setup, because there simply
isn’t one. The goal is to present differ-
ent options and explain their advantag-
es and disadvantages so that you can
decide for yourself what solution may
be appropriate in your particular case.
 13

PART
1

The history and

technology that
header bidding
builds on
14

Chapter 0:
PROGRAMMATIC BUYING

Programmatic buying is a very com-

plex, extensive topic and is not dealt RTB
with in its entirety here; it would be RTB (Real-Time Bidding) is a form of
worth a separate publication. For the programmatic buying based on the
purposes of this publication, we will auction method. That is, the highest
stick to topics that touch on header bidder auctions off the advertising
bidding in some way. However, let’s space and can place their ad on it.
at least introduce the basic points of
The buyer can attach special restric-
programmatic buying, because header
tive requirements to his request to buy
bidding is a part of it.
online advertising. Regarding technical
Programmatic is an automated way of targeting of specific Internet users, re-
buying online advertising. It replaces quirements could include what location
the original buying model where people they must be in, what device they must
emailed orders and banner ads to each be using, at what time, etc., and also
other. With programmatic buying, it specific qualitative restrictions such as
is possible to buy and sell ad space in age group, user’s interest in a specific
an automated way without having to industry, and many others.
involve human labor for each order.
In addition to RTB, there are other
Here, the human only sets the rules
programmatic buying methods, such as
within the framework of which the com-
“programmatic direct” or “programmat-
puters can then carry out the transac-
ic guaranteed”.
tions between themselves.
But human labor has not disappeared,
it has moved to another level. Pro-
grammatic buying works efficiently,
quickly, and cheaply (compared to
human labor), but is very complex and
requires more technical than business
knowledge. Instead of handling specif-
ic orders, people here keep the whole
system running as well as possible and
constantly try to optimize it.
 15

Technology ters about the implemented advertise-

ment, thanks to which we can evaluate
Both programmatic buying and RTB it in great detail and adjust the input
are general practices and protocols. settings for further buying according to
They require technical tools that work these results.
according to their principles. SSP – Supply Side Platform is a seller’s
The tools listed below are able to ex- (publisher’s) tool. In it, the publisher
change information between each other sets what advertising units are avail-
and close trades in the order of tens able and under what conditions he is
of milliseconds according to the set- willing to sell them, for example, floor
tings of their administrators. They also price (see explanation on page 34).
display the final sold ad immediately
on the sold space (e.g. website, mobile
app, smart TV, etc.).
There are many of such tools, here we
will just introduce three basic ones:
DSP – Demand Side Platform is a
buyer’s (advertiser’s) tool. In it, the
advertiser sets limiting parameters on
the ad he wants to buy, for example,
the period or location in which the ad is
to be displayed, the total budget he is
willing to spend or even the maximum
price per ad impression, or qualitative
targeting of the type of audience he
wants to reach (age, gender, interests,
etc.). The advertiser also sets his ads
in the DSP so that when a programmat-
ic purchase is made, the system itself
sends the ad to the target location. In
addition, the DSP shows many parame-
16

However, it may also have qualitative Ad Exchange – Ad Exchange is a

requirements for the type and content notional marketplace where DSPs and
of ads that can be displayed and may SSPs communicate with each other and
even allow for the publisher to manu- arrange auction-based (RTB) trades. Ad
ally approve the ads. However, in most Exchange connects multiple DSPs and
cases, the publisher is more likely to SSPs and organizes auctions between
set restrictive rules that govern the ad them.
approval process automatically.

RTB principle

AD
SPACE
100 ms
$$ $$ $$$

PUBLISHER/SUPPLIER AD EXCHANGE BRAND/ADVENTIZER

SSP DSP
(Supply-Side Platform) (Demand-Side Platform)

Figure 0.1: Principle of RTB

 17

Demand partners
Very often we use the term “demand Therefore, for the purposes of this
partners”. By this we mean any tech- publication, we do not refer to de-
nology capable of participating in an mand-side platforms (DSPs) only, but
auction run by a publisher or, more to all systems that are able to partici-
generally, technology capable of selling pate in programmatic purchasing from
real-time bidding advertising. a publisher.
Their goal is to find out enough infor-
mation about each bid request, pass
the information on to other connected
systems and, if necessary, submit
a bid response. Typically, this is a
Supply-Side Platform (SSP), but oth-
er technologies such as ad exchanges
may also be involved in the auction.

Conclusion of the chapter:

Programmatic buying
In this chapter we: The information in this chapter is gen-
● explained the basic concepts of eral in nature and only explains in very
programmatic buying, general terms the basic concepts that
are necessary to understand the entire
● explained what innovation it
has brought to the online sales content of this book.
environment,
● explained the basic technologies with
which programmatic buying works,
● explained the important concept of
“demand partners”.
18

Chapter 1:
HISTORY OF HEADER BIDDING

In December 20102, the OpenRTB Con- because in addition to the header bid-
sortium was announced with the goal ding auction, there are SSP-level and
of defining a standard for real-time ad exchanges.
bidding. The standard was created in For a better overview of the develop-
2012 and was adopted as an IAB stand- ment of header bidding, we have pre-
ard, and today we know it as the RTB pared a timeline describing the major
protocol. milestones:
This gradually led to the emergence of
a header bidding solution that enables
real-time online auctioning between
individual demand partners. Initially
this was client-side header bidding, but
later on, server-side header bidding
and hybrid solutions were developed
(see detailed description on page 48).
The standard was in place, but mass
use was still far from being achieved.
The next significant step forward was
the creation of the open source header
bidding solution, Prebid, in 20163 by the
team around AppNexus (later AT&T’s
Xandr). The broader community around
GitHub subsequently contributed and
continues to contribute to its develop-
ment, and in 2017 Prebid.js 1.0 was
released4.
The designation “header bidding” has
established itself gradually. Initially,
the technique was also referred to by
the names “Tagless”, because it does
away with the classic ad tags common
in waterfall, or “auction auctioning”,

2
https://www.iab.com/guidelines/openrtb/
3
https://prebid.org/blog/happy-birthday-prebid-js/
4
https://prebid.org/blog/announcing-prebid-1-0/
 19

Historical developments related to header bidding

1994
Cookies were
developed
2007
Google Dynamic
Allocation launched
2010
OpenRTB Consortium
founded
2015
Prebid introduced
2017
IAB introduced Cookie IAB announced intention
ads.txt functionality in to upgrade RTB protocol
2018 Apple Safari to version 3.0
GDPR went into effect browsers
restricted
2019
IAB introduced Google Google announced
Sellers.json and announced preparation of
2020 supply chain move to First Privacy Sandox
Project Rearc Price auction
founded model
2021
IAB defined Buyers.json
2022

2023
Planned end of cookie functionality
within Google Chrome

Figure 1.1: Historical developments related to header bidding

20

Mass expansion of header bidding

After the launch of Prebid, header bid-
ding started to spread among publish-
ers and was adopted incredibly quickly.
In 2019, about 80% of all publishers in
the US used header bidding5.

US Header Bidding Adoption, Oct 2017 - March 2019

79,2%

77,8%
78,5%
75,7% 75,5%
75,1%
76,4%
73,7% 75,6% 75,8%
73,3% 75,2%

73,1%

Jan Feb Apr May Jun Aug Sep Oct Nov Dec Jan Feb Mar
2018 2019

Figure 1.2: US Header Bidding Adoption. Source: AdZerk, AdTech Insights;

eMarketer.com

Due to its key position, in the following Most of these topics are covered in
years header bidding became associat- separate chapters later in this book. If
ed with several different major spheres publisher wants to wield header bidding
reflecting market developments – to your advantage, needs to be familiar
GDPR, supply chain, and User ID to with these side projects as well. Their
name a few. impact on online auctions is tremen-
dous due to their large influence on bid
price (CPM).

5
https://www.emarketer.com/chart/227616/us-header-bidding-adoption-jan-2018-march-2019-of-total-hbix-
sites
 21

Financial benefits of header

bidding for publishers
There is a general consensus in the ferent in the US versus Western Europe
market that header bidding brings versus India. But there is a dependency
more revenue to publishers, which has between the potential benefit and the
encouraged mass usage. But by what development of the programmatic mar-
percent can it increase revenue exactly? ket within a given location. One of the
This is a difficult question and the best header bidding situations with the
answer will always be individual to greatest benefit is in the United States,
each publisher. It’s also important to where the programmatic market is
note that it will be completely different developing at one of the fastest rates
depending on your region and the state in the world6.
of the market. The situation will be dif-

US Programmatic Digital Display Ad Spending 2016-2021

$81,00
84,9% 86,3%
79,3% 83,0%

73,0%
$59,45 $70,23 87,5%
$49,23
48,2%
$35,82 37,5%
$25,48
40,6%

20,7% 18,1% 15,3%

2016 2017 2018 2019 2020 2021

Programmatic digital display ad spending

% change % of total digital dispaly ad spending

Figure 1.3: US Programmatic Digital Display Ad Spending. Source: eMarketer.com

6
https://www.emarketer.com/content/five-charts-the-state-of-header-bidding
22

Here is some published information on revenue increases after

implementing header bidding. Please refer to the sources for
more information on the conditions under which the studies were
produced.

According to research published on eMarketer,7 73% of the top 1,000 US

websites used header bidding in February 2018, and 48% of them confirmed
that they gained higher CPM revenue.
UK publisher The Telegraph increased revenue by 70% after introducing
header bidding8.
StudyBreak Media increased its revenue by 25-30% after introducing header
bidding9.
Fifty publishers in Australia and New Zealand describe a 40% increase in
CPM after introducing Exchange Bidding (Publift10).
DAZN Media increased programmatic video revenue by 150% year-over-year
following the introduction of Exchange Bidding11.
SpaceStation showed12 a 100% increase in CPM following the introduction of
Exchange Bidding.
Publisher Axel Springer switched from waterfall to header bidding in 2017. In
2018, they presented a 10% increase in CPM13.
The term “Exchange Bidding” refers to header bidding from Google, which
we will discuss in more detail in the second part of the book (page 57).
There are plenty of other studies on this topic to be found, however most
of them are general and do not share precise information about the exact
system setup before or after the test

The general opinion in the market is

that header bidding brings higher re-
turns to publishers. To serve you well
though, this technology requires con-
sistent care and must be approached
comprehensively. It‘s constantly evolv-
ing, changing, and permeating other
areas of programmatic buying.

7
https://www.emarketer.com/content/more-publishers-are-using-a-hybrid-header-bidding-approach
8
https://digiday.com/media/telegraph-used-header-bidding-increase-programmatic-revenue-70-percent/
9
https://digiday.com/media/header-bidding-publishers-boosting-cpms-much-50-percent/
10
https://admanager.google.com/home/resources/publift-uses-exchange-bidding-to-deliver-higher-cpms/
11
https://admanager.google.com/home/resources/dazn-programmatic-video-exchange-bidding/
12
https://admanager.google.com/home/resources/video-spacestation-24-exchange-bidding/
13
https://www.jisem-journal.com/download/header-bidding-as-smart-service-for-selling-ads-in-the-digital-
era-8483.pdf
 23

Header Bidding as an answer

to Google Dynamic Allocation
Back in 2007 (long before the Open- is even facing lawsuits over its market
RTB Consortium and the idea of header domination16 17.
bidding), Google launched “Dynamic Header bidding is one of the responses
Allocation” on its ad server. to Dynamic Allocation. When set up
The idea behind Dynamic Allocation is correctly, it gives all ad demand part-
to give Google a chance to intervene ners the ability to bid on ad impres-
in the ad server’s decision-making sions at the same time and under the
process to give Google’s ad sources same conditions.
a chance to outbid the winning bid at
the last minute in a waterfall model.
For more information, see the section However, if you have header bidding in-
on Google solutions (full discussion on tegrated into Google’s ad server, there
page 57). is still the chance that Dynamic Allo-
cation could override the final auction
Opinions on this feature vary. Some winner, just like another direct bid in
sources say that this feature brings the ad server. We’ll cover this more in
a significant increase in revenue for a separate chapter in the second part
publishers14. Other sources15 criticize of the book (see page 106).
Google’s dominance in the ad server
field and the fact that this option is
only given to Google sources. Google

Conclusion of the chapter:

History of header bidding
In this chapter we: ● defined dynamic allocation.
● described the history of the This knowledge will help you understand
development of header bidding and why header bidding came into being and
related events, what its purpose is.
● explained what led to the mass
spread of header bidding,

14
https://www.adexchanger.com/the-sell-sider/as-google-enters-heading-bidding-game-whats-the-impact-
on-client-side-integrations/
15
https://law.stanford.edu/publications/why-google-dominates-advertising-markets/
16
https://www.bbc.com/news/technology-57567799
17
https://adage.com/article/digital/googles-digital-ad-dominance-harming-marketers-and-publish-
ers-says-new-study/2257576
24

Chapter 2:
WATERFALL

Waterfall is a way to engage multiple

demand partners and pass the adver-
tising offer between them.
Waterfall is also sometimes referred
to as “Daisy chain” or “Waterfall tags”.
It is a method of buying online adver-
tising that predates header bidding. In
many places where waterfall was used
before, header bidding has taken over.
Waterfall is used only to a limited ex-
tent today, but it still has relevance and
is used in certain cases.

History and reason for

the creation of the waterfall
In the early days of Internet advertis- advertisers was pooled. This infrastruc-
ing, orders and advertising materials ture made it possible for publishers to
were emailed to publishers and ads get enquiries from multiple advertisers
were manually deployed directly on at once without communicating with all
the publisher’s side usually using an of them personally.
ad server (for more on ad servers, see However, these demand partners were
Chapter 7, page 106), or even included often burdened with negative side
directly into the code of a web page. effects, with demand partners buying
Ad servers made it possible to process just above the floor price (see page
ads from one place, control the ad in 34 for floor price explanation). It was
real time, and track statistics online. a black box providing minimal infor-
This was a major step forward in Inter- mation. Gradually, the waterfall model
net advertising. Subsequently, demand became established and persisted for
partners (ad exchanges and SSPs) several years (until the rise of header
emerged, where the supply of multiple bidding).
 25

As a direct consequence of the water- Therefore, if a sale was about to hap-

fall model, the publisher’s advertising pen within the waterfall, Google could
space was divided into premium and “beat” the transaction at the last min-
remnant space. In practice, publishers ute with its own bid at a price of only
sold their premium space to direct cli- $0.01 higher according to the principles
ents first, and once they satisfied these of the “Second Price” auction, which
clients, they released the remaining was common practice at the time.
space to demand partners via waterfall.
The waterfall process had one more
critical point if you were using Google’s
ad server. Google favored its own ad
networks over other demand partners
by introducing Dynamic Allocation.
26

Functioning of the waterfall

The principle of the waterfall is to call In the case that no demand partner
individual demand partners in turn, delivers a bid higher than the defined
informing each one at the beginning price, the impression is returned to the
of the minimum acceptable price. If a publisher (to the ad server) and the
demand partner offers an advertise- publisher has the option to use it in
ment at a price higher than the defined another way. For example, displaying
minimum, a trade will occur and the nothing or promoting their own prod-
advertisement will be displayed. Other- ucts.
wise, the bid is passed on to the next So, this is a simple idea where by
demand partner in line. The whole pro- adding another demand partner to the
cess repeats until one of the demand waterfall process, the publisher in-
partner offers more than the minimum creases the chances of selling out their
amount, or the waterfall reaches the ad space.
end of the set path without selling the
space.

The principle of the waterfall

$10
eCPM

$9
eCPM

$7
eCPM

$4
eCPM

Figure 2.1: The principle of the waterfall

 27

Advantages and disadvantages

of waterfall
In short, the main advantages and dis-
advantages of waterfall are:

Advantages Disadvantages
• Does not guarantee selection
• Simple installation
of the highest bid

• Higher fill rate • Very high maintenance

• Extended page load time

• Lack of information

• Lack of information

Let’s elaborate on that.

Advantages of
waterfall
● Simple to use - these are usually
simple ad scripts that are easy to
deploy to a website or ad server.
Definitely easier than engaging
complex header bidding.
● Higher fill rate - by reaching out to
multiple demand partners over time,
the chances of selling your ad are
relatively high.
28

Disadvantages ● Increased page load time - the

more demand partners you call
of waterfall in succession and wait for their
● It does not guarantee the highest response, the longer it takes for the
winning bid - on the contrary, a large ad, or even the whole website on
percentage of the highest bidders will which the waterfall is hosted, to load.
not get their turn at all.
● Lack of information - if a sale is
● Very difficult to maintain - if you have made, the publisher will never know
multiple ad sources in a row, then it if any demand partner (next in line)
is difficult to keep track of them all would be willing to offer a better
and there is even a risk of the whole price.
system failing.

Waterfall integration options

Integrating demand partners in the wa- that as a seller you can maximize your
terfall can be simple, or you can take profit by reaching different consumers
a sophisticated approach to maximize at different parts of the demand curve.
revenue. In other words, there is a customer
This method is based on the economic who is willing to pay more than another
demand curve18. Economic theory says customer. But the number of custom-

Figure 2.2: Demand curve

18
https://en.wikipedia.org/wiki/Demand_curve
 29

ers willing to pay a higher price is lower options. Eventually, we should sell out
than the number of customers willing to our space at the highest possible price.
pay less. The specific price levels for individual
So, according to this theory, we put the demand partners is usually based on
different demand partners in a queue historical data.
and give them different minimum prices
at which we allow them to buy, decreas-
ing the minimum as we exhaust our

Demand curve adapted to programmatic waterfall

First ad exchange
PRICE

Second ad exchange

Third ad aexchange

QUANTITY

Figure 2.3: Demand curve adapted to programmatic waterfall

Passback
Calling the next demand partner is
referred to as a “passback”. In most
cases, an advertising code (JavaScript
tag) is set within the administration,
which will be activated (called) when
the previous demand partner is unable
to deliver any offer.
This passback usually falls into one of
the following categories:
30

1. The promotional code of the next/a problem and maintaining a long water-
different demand partner within the wa- fall can be very time consuming. At the
terfall. This creates a chain of demand end of this chain is usually the return
partners that are called in succession of the impression to the ad server if no
until one of them meets the price condi- demand partner is interested.
tion, thus selling the ad.
Calling the next demand partner is set
at the previous demand partner’s bid.
There is a possibility that a technical
problem occurs and the next demand
partner is not called. Finding such a

Direct passback

AD SERVER

demand
partner A

demand
partner B

demand
partner C

AD SERVER

Figure 2.4: Direct passback

 31

2. Ad server code that returns the un- alter the order of partners. The disad-
sold impression to the ad server before vantage is that it takes longer to load
moving on to the next demand partner. the web page on which the waterfall
With this method, we have more options is hosted, because every step we take
for unsold impressions. Most of the within the waterfall takes time, and this
time, we call the ad server to the next method almost doubles the number of
demand partner in the chain. The ad- steps.
vantage of this method is that you have
much more control over all demand
partners and can easily make changes
in case you want to add, remove, or

Passback combined with ad server call

AD SERVER

demand
parter A

AD SERVER

demand
parter B

AD SERVER

demand
parter C

AD SERVER
Figure 2.5: Passback combined with ad server call
32

Header bidding versus waterfall

Let’s take a look at the comparison of
waterfall and header bidding in a clear
table.

Functions Waterfall Header bidding

Activation method Ad tag Header bidding scripts

First one to meet the price

Auction winner Highest bid
condition
All demand partners at once
Activation process Sequential
(real time)
Length of the process of
Longer Shorter
calling demand partners
Optimal matching the cu-
Yield (revenue) Less than it could be
rrent bid

The future of the waterfall

According to research published on eMarketer,19 a total of 73% of the top
1,000 US websites used header bidding in February 2018, and 48% of them
confirmed that they gained higher CPM revenue by doing so. In other words,
their main automated sales method was no longer the waterfall

Header bidding has significantly more

benefits for publishers than waterfall as
we listed above. Does this mean that
the waterfall will gradually disappear
completely?
 33

There will probably be reasons for pub-

lishers to use waterfall in some cases
in the future. For example, it is possible
that there will be ad sources that do
not allow integration via header bidding
and their benefit to publishers will be
so significant that they will choose to
utilize the waterfall method.

Conclusion of the chapter: Waterfall

In this chapter we: This knowledge will help you to un-
● explained the concept of passback, derstand the second part of the book,
which discusses in detail how header
● described what waterfall is and its
bidding works.
history,
● discussed the advantages and
disadvantages of waterfall, including
those relative to header bidding.

19
https://www.emarketer.com/content/more-publishers-are-using-a-hybrid-header-bidding-approach
34

Chapter 3:
TYPES OF AUCTIONS BY FIRST
VS SECOND PRICE
Just after the birth of the RTB protocol,
the commonly used auction price was
the “second” price. Below we explain
the different types of auctions and
why there was an eventual shift to the
“first” price auction type.

Various demand partners have

gradually moved to first price
auctions, and in 2019, Google also
announced a move to the first
price model20.

Floor price
We will use the term “floor price” a lot
in the following pages, let’s explain it
first.
The floor price is the minimum price
that a publisher asks for its space to
place an advertisement. This price is
set by publisher in the demand part-
ners’ systems and bids below this price
are not included in the auction at all.
The floor price is important in both the
second price and first price models and
has a special meaning in header bid-
ding, which will be shown with specific
examples.

20
https://www.blog.google/products/admanager/simplifying-programmatic-first-price-auctions-goo-
gle-ad-manager/
 35

Floor Price
CPM

Floor price
Ads not considered

Bid A Bid B Bid C Bid D Bid E

Figure 3.1: Floor price

Second price auction

In a second price auction when the one cent more than the second highest
publisher does not set a floor price, the bid). Thus, the advertiser saves $0.49
winner of the auction pays not the price on their bid.
they bid, but the price of the second However, if the publisher set the floor
highest bid plus one cent. price at, say, $4.2, then the winning bid
Example: would be charged $4.21. That is, one
cent above the floor price.
Three bidders participate in an auction:
Floor price is set by publishers at lower
● Advertiser X with a bid of $4
price levels, usually when there are few
● Advertiser Z with a bid of $4.5 bidders or low bids
● Advertiser Y with a bid of $3.5
Advertiser Z wins the auction, but does
not pay $4.5, but only $4.01 (which is
36

This model seems to be more attractive

to advertisers because it saves them
money when the second highest bid in
an auction is significantly cheaper than
the winning bid.

Second price auction

BIDDER X BIDDER Y BIDDER Z

$4,00 $3,50 $4,50

WINNER
BIDDER Z
$4,01

Figure 3.2: Second price auction

First price auction

In this type of auction, the winner pays Advertiser Z wins the auction, and pays
exactly the price they bid in the auction. $4.5 (the exact price they bid).
Example: Floor price is not relevant here, unless
Three bidders participate in an auction: it is higher than the winning bid. Then
the auction wouldn‘t even take place,
● Advertiser X with a bid of $4 because no advertiser would bid higher
● Advertiser Z with a bid of $4.5 than the publisher‘s asking price. The
impression would therefore not have
● Advertiser Y with a bid of $3.5
been sold at all.
 37

The first price model seems to be

more attractive to publishers as it
increases the price that advertisers
pay. However, Google said in a
statement21 that its tests show that
the change from second price to
first price has a neutral to slightly
positive impact on publishers’
revenues.

First price auction

BIDDER X BIDDER Y BIDDER Z

$4,00 $3,50 $4,50

WINNER
BIDDER Z
$4,50

Figure 3.3: First price auction

21
https://www.blog.google/products/admanager/rolling-out-first-price-auctions-google-ad-manager-partners/
38

The difference between First

price and Second price auctions
price auction
Basically, the main difference is the
price the winning advertiser pays for the
impression they win.

Auction type First price auction Second price

auction
Winner of the auction The highest bid The highest bid

The second highest price +

The winner pays The price they bid
$0.01

Problems of the second price

model
As mentioned in the second price mod- model. Although it had been used by
el, the winning advertiser pays a price other players in the market previously,
$0.01 higher than the second highest it was Google that was responsible for
bid. its resurgence. Google defends their
choice by saying that the first price
However, the winning bidder has no
model provides greater transparency,
information about the second highest
particularly around the final bid price,
bid. They must rely blindly on the infor-
and more room for technology outside
mation they receive. Thus, in practice,
of the Google ecosystem.
price manipulation can occur where the
winning advertiser is reported a higher According to a study published on
price than their competitors actually bid. eMarket22 comparing one quarter of
For example, there may be ghost bids 2017 and 2018, the percentage of SSPs
which inflate the advertiser‘s price. using the first price model increased
from around 6% to around 43%.
Over time, this lack of transparency
has led to the promotion of first price
bidding. The most popular auction type
was second price until 2019, when
Google started applying the first price

22
https://www.emarketer.com/api/RedirectAsset?a=chart&r=242771
 39

Digital Ad Impression Share Among US Supply-Side Platform

(SSPs), by Auction Type, Dec 2017 & march 2018

19,1% 23,9%

32,8%
75,1%

43,3%

5,8%
Dec 2017 March 2018
First-price auction Second-price auction
Fair second-price auction with anomalies

Figure 3.4: US Programmatic Digital Display Ad Spending. Source: eMarketer.com23

Problems of the first price

model
In contrast to second price, the first
price model appears to be transparent
to all participants. This is its indisputa-
ble advantage. However, it also has the
disadvantage of aggressively pushing
the price in favour of the publisher.
Obviously, advertisers will be very cau-
tious in such a model and will carefully
consider what price to offer.

23
https://www.emarketer.com/chart/218560/digital-ad-impression-share-among-us-supply-side-platforms-
ssps-by-auction-type-dec-2017-march-2018-of-total-impressions-analyzed-by-getintent
40

Hard and soft floor price

The floor price definition has been ex- In practice, this model is used as a deci-
panded over time to include two vari- sion boundary in either a second or first
ants that broaden publishers‘ options: price auction depending on the bids. Let
- Hard floor price – the price below us explain this with a few examples:
which advertisers‘ offers are never ● If the highest bid in an auction is
accepted. Advertising space cannot be higher than the soft floor, then the
purchased below this price. second price auction model is applied.
- Soft floor price – a price that is rela- ● If the highest bid in the auction is
tively low for the publisher, but below lower than the soft floor but higher
which he is still willing to accept an than the hard floor, then the first
offer. price auction model is applied.
The soft price is of course always higher ● If the highest bid in the auction is
than the hard price and can be set by lower than the hard floor, then the
the publisher for cases where he does bid is eliminated and the auction ends
not want to lose low, but still acceptable unsuccessfully.
offers.

Soft a hard floor price

$1,00 2nd price auction

soft floor
price
$1,00 2nd price auction
soft floor 1st price auction
price no auction
$0,10
hard floor
price no auction

Figure 3.5: Soft and hard floor price

 41

This idea seems interesting at first, but

unfortunately, in detail it complicates
the transparency of the whole system,
because the advertiser has no informa-
tion in advance about what type of auc-
tion will be applied to his bid. Additional-
ly, the long-term benefits for publishers
are usually small in practice, as shown
in Robert Zeithammer‘s comprehensive
study24.

Bid shading
Bid shading is the name for a special $0.01 more than his competitor, no
algorithm that tries to estimate the matter how much he offers. But in the
optimal bid price from the advertiser‘s first price model, he pays exactly what
view in order to win as many auctions he offers. Thus, keeping an eye on the
as possible while paying as little as pos- optimal price is necessary. Bid shading
sible. The advertiser does not set the is therefore closely linked to the first
final bid amount, they leave that to the price model.
algorithm. The algorithm learns from With technology, we can track a number
historical data and tries to optimize the of parameters that come into play in
bid price. each auction–for example, time, de-
In the second price model, the winning vice type, user interests, domain, page
advertiser could partially rely on simply content–and compare them to historical
paying $0.01 more than competing bids. results under the same or similar con-
In the first price model, the advertiser ditions. Involving machine learning in
pays the price they bid. Thus, they must the calculation of the optimal bid price
carefully weigh what price is optimal, seems to be an ideal option for bidders.
because there is a real risk of paying
unnecessarily more. In the second price
model, the advertiser always pays only

24
https://www.anderson.ucla.edu/faculty_pages/robert.zeithammer/SoftFloors_0719.pdf
42

A detailed design of the functionality of Bid shading may appear negative from
the bid shading algorithm is presented, the publishers‘ point of view. However,
for example, in the document Bid Shad- we have to remember that publishers
ing in The Brave New World of First- are not concerned with the short-term
Price Auctions25. squeezing of advertisers, but with the
long-term relationship. And only adver-
tisers who find their investment profita-
According to some technology
ble will return and continue to advertise.
players26, bid shading saves the
advertiser about 20% on cost. As
the first price method becomes
more popular, the functionality of
bid shading and similar methods
will certainly improve.

Both DSP and SSP systems work with

bid shading. While DSP is an adver-
tisers‘ tool for the most part, SSP bid
shading functionality raises many ques-
tions27.
Since SSP refers to the publisher‘s side
(they work on the publisher‘s behalf and
are paid by them), they have a different
type of data that media owners may not
always be willing to share with advertis-
ers.

25
https://arxiv.org/pdf/2009.01360.pdf
26
https://www.adexchanger.com/online-advertising/rubicon-grows-revenue-but-bid-shading-is-driving-
down-publisher-cpms/
27
https://www.adexchanger.com/the-sell-sider/shining-a-light-on-the-darker-side-of-bid-shading/
 43

Floor price setting strategy

Publishers should consider very care- In header bidding, on the other hand,
fully whether and at what level they will the floor price is applied directly in the
set the floor price, whether they use a auction settings to all demand partners
first price or second price model, and at once. Thus, partners with a too-low
they need to know the principles of the price are excluded from the auction. We
auction. will discuss the floor price, first price
With waterfall, the usual technique was and second price in relation to header
to gradually reduce the floor price as bidding in the second part of the book
the offer got further down the waterfall (page 69).
in order to increase the fill rate. Thus,
the last demand partner in the path had
the lowest floor price, but also had the
lowest chance of the bid reaching them
at all.

Conclusion of the chapter:

Types of auctions by first vs second price
In this chapter we: This knowledge will be necessary to
● explained the concepts of floor price understand the second part of the book,
and bid shading, where we will discuss which type of
auctions header bidding works with and
● discussed the principles of first price
what floor price can be set in it.
and second price auctions.
44
 45

PART
2

Header bidding
solutions
46

Chapter 4:
HEADER BIDDING

Header bidding is also sometimes matic environment that ensures that all
referred to by the names “pre-bid” or bidders have the same conditions and
“auction of auctions”. time for the auction.
Almost all sources dealing with header Header bidding is a process that starts
bidding emphasize its difference from when the header of a web page is
waterfall (see page 24 for an explana- loaded. Here a piece of JavaScript code
tion of waterfall), and essentially refer is placed that activates the whole next
to it as a successor technology. process. Thus, by definition, header
Header bidding is a market response bidding is a technology that can only be
to Google’s ad server dominance and used in a web environment that sup-
Dynamic Allocation favoring Google’s ports JavaScript.
demand sources. It also serves as an The following chapter often refers to
alternative to waterfall solutions that the “wrapper”. A wrapper is, in short,
eliminates much of its drawbacks. an administration system for controlling
It originated as a kind of “hack” for header bidding. (For details and a more
Google, but gradually evolved into a in-depth explanation of wrappers,
massively adopted technology by the please refer to Chapter 5.)
market, which was eventually adopt-
ed by Google in its Open Bidding (see
page 57).
So what is header bidding? It is a meth-
od of online auctioning in a program-

Principle
The basic philosophy of header bidding bidding addresses them all at once and
is very simple and can be well described therefore theoretically gives them an
in contrast to waterfall bidding. equal chance in the auction. The pub-
While waterfall addresses individual lisher, by comparing all bids in the auc-
demand partners one by one, header tion at once, gets the best price at that
 47

moment, and the advertiser (buyer), is forward the technology is improving

assured that he will not lose just be- every day, and overall it is definitely an
cause he is too far down the waterfall. improvement over waterfall.
This is the theory, which might some-
times differ from reality, as we will
discuss later, but as technology moves

Difference between waterfall and header bidding

WATERFALL HEADER BIDDING

auction auction
$ 3,00 floor price $ 3,00 floor price
REQUEST PASSBACK
BID BID
PARTNER 1 $ 2,80 PARTNER 1 $ 2,80
BELOW FLOOR BELOW FLOOR

BID BID
PARTNER 2 $ 3,50 REQUEST
PARTNER 2 $ 3,50
WINNING BID LOSING BID

BID BID
PARTNER 3 $ 4,30 PARTNER 3 $ 4,30
NOT CALLED WINNING BID

BID BID
PARTNER 4 $ 3,70 PARTNER 4 $ 3,70
NOT CALLED LOSING BID

Figure 4.1: Difference between waterfall and header bidding

The figure above compares waterfall While in waterfall the first partner in the
and header bidding. The advertisers and order who bids more than the set floor
the prices they bid are the same in both price wins, in header bidding the part-
cases, but the winners and the prices ner who bids the highest price wins.
they pay are different.
48

Header bidding is also sometimes Client-side header

referred to as „auction of auctions“
because there is an auction not only at bidding (C2S)
the header bidding level but also at the
It is also sometimes referred to by the
participating demand partners. They
abbreviations CSHB (Client-Side Header
execute an internal auction within their
Bidding) or C2S (Client to Server)
solution and send the winning bid to the
publisher‘s header bidding to compete We will explain the functionality by
with their other competitors. looking at the steps of the process:
Participating demand partners are any
technology that is technically capable
of participating in the auction (most
commonly SSPs and ad exchanges). In
practice, they run a suitable API or con-
nector that can connect to the auction.
They are also demand partners that
have been selected by the publisher and
are allowed to participate in the auction.

Client-side vs Server
-side header bidding
Header bidding can be divided according
to the principle of operation into:
● Client-side
● Server-side
● Hybrid

We will discuss all three in detail below.

 49

Client-side header bidding

SSP 100 Floor price: $4,30 CPM

#1 ms

SSP 150 Bid 1:

#2 ms $5,20 CPM

SSP 200
Bid 2:
#3 ms $5,00 CPM
<JavaScript/>
Ad Bid 3:
WINNER exchange 250
#1 ms $5,90 CPM

Ad Bid 5:
exchange 350
#2 ms $5,70 CPM
Publisher’s
ad server
Ad
exchange 450 Timed out
#3 ms

Figure 4.2: Client-side header bidding

1. When the user visits the web page, a) They may also fail to respond to
the JavaScript code is executed. the offer within a defined timeout. In
2. JavaScript sends information about this case, they are excluded from the
the new auction to all the defined de- auction. The timeout has been set
mand partners. Among other things, it to 400ms for the scheme, so a bid
sends, for example, the URL where the arriving as late as 450ms is excluded.
ad will be displayed, the definition of the b) They are also excluded if their bid
ad unit, and visitor information. is lower than the floor price. In our
3. The individual demand partners eval- example, the floor price was set at
uate the demand internally and respond $4.5, so a bid of $4.3 is excluded.
with a bid price they are willing to pay,
or may not respond (they are not inter-
ested in the current auction).
50

4. We discussed both timeout (page 86) 6. Further steps are consistent with
and floor price (page 34) in more detail any other ad display and end with the
in separate sections. JavaScript selects display of the winning ad.
the highest bid and sends it to the ad
server.
5. The ad server will decide, based
on the other campaigns it has set up,
whether to use the bid or discard it and
use another campaign it has available.
So the ad server is no longer deciding
based on price (more about how the ad
server is deciding in Chapter 7, page
106) and publishers need to be aware
of this and work with other ad server
setups.
a) In ad servers that have integrat-
ed work with programmatic deals,
there is still a comparison with these
offers.
b) In the Google ad server, it is pos-
sible to enable Dynamic Allocation to
compare bids with Google bids.
 51

Server-side header bidding (S2S)

It is also sometimes referred to by the
abbreviations SSHB (Server-side header
bidding) or S2S (Server to Server).
Steps that happen:

Server-side header bidding

Publisher’s
Ad 100 Floor price: $4,30 CPM
Server ms

SSP 150 Bid 1:

#1 ms $5,20 CPM
SSHB
vendor
SSP 200
Bid 2:
#2 ms $5,00 CPM

SSP Bid 3:
WINNER 250
#3 ms $5,90 CPM

Ad Bid 5:
Network 350
#1 ms $5,70 CPM
Publisher’s
Ad server
Ad
Network 450 Timed out
#2 ms

Figure 4.3: Server-side header bidding

1. When the user visits the web other things, it sends, for example, the
page, the JavaScript code is exe- URL where the ad will be displayed,
cuted and it sends a request to the the definition of the ad unit, and visitor
header bidding server. information.
2. The header bidding server sends
information about the new auction to
all defined demand partners. Among
52

3. The individual demand partners eval- Chapter 7, page 106) and publishers
uate the demand internally and respond need to be aware of this and work with
with a bid price they are willing to pay, other ad server setups.
or may not respond (they are not inter- a) In ad servers that have integrated
ested in the current auction). work with programmatic deals, there
a) They may also fail to respond to is still a comparison with the follow-
the offer within a defined timeout. In ing offers.
this case, they are excluded from the b) In the Google ad server, it is still
auction. The timeout has been set possible to enable Dynamic Alloca-
to 400ms for the scheme, so a bid tion to compare bids with Google
arriving as late as 450ms is excluded. bids.
b) They are also excluded if their bid 6. The next steps are already classic for
is lower than the floor price. In our any other ad display and end with the
example, the floor price was set at display of the winning ad.
$4.5, so a bid of $4.3 is excluded.
We discussed both timeout (page 86)
and floor price (page 34) in more detail The points where the S2S solution dif-
in separate sections. fers from C2S are marked in bold.
As you can see, these are minimal dif-
4. The header bidding server selects ferences.
the highest bid and sends it to the pub-
lisher‘s ad server.
5. The ad server decides according
to the other campaigns it has set up
whether to use the submitted bid or
discard it and use some other campaign
it has available. So the ad server is no
longer deciding based on price (more
about how the ad server is deciding in
 53

Difference between C2S and S2S

– advantages and disadvantages
The only difference between C2S and JavaScript in the user‘s browser. In S2S,
S2S is minimal at first sight and relates the requests are evaluated by the serv-
to the second stage of the process. In er solution.
C2S, the requests are evaluated by

Difference between C2S and S2S header bidding

SSP SSP

SSP SSP
SSP SSP

SSP SSP AD SERVER

CLIENT-SIDE SERVER-SIDE
HEADER HEADER
BIDDING BIDDING

Figure 4.4: Difference between C2S and S2S header bidding

However, this difference may seem nature of the technology, faster on av-
small, but it is very technically signifi- erage.
cant and has many consequences:
2) Number of ad sources in-
1) Speed – S2S solves possible la- volved – Web browsers have a limited
tency problems with the C2S solution. number of requests while servers have
This problem is no longer as pressing as no such limitation, making S2S the
it was in the pioneering days of head- better solution if you would like to do
er bidding. Today, even C2S solutions an auction between a large number of
can be very fast with the right settings. partners (dozens). With a smaller num-
However, the S2S solution is, by the ber, this plays a minimal role. Chapter 5,
54

addresses the usual number of demand 5) Number of demand partners

partners in an auction and shows that involved – S2S solutions came later
involving a large number of partners in than C2S and the number of demand
C2S does not make sense. partners that support C2S is clearly
higher at the time of this book‘s publica-
3) Transparency – While on the
tion. However, there are already start-
browser side everything is public, trans-
ing to profile those demand partners
parent and auditable, on the server-side
that are strictly following the S2S path.
independent control is difficult. So if you
More on this area in the separate chap-
want to use S2S, then only involve de-
ter on demand partners and wrappers
mand partners that you trust and pos-
(Chapter 5, page 82).
sibly have control resolved. Trust and
transparency can also be an issue if the 6) Technical implementation –
operator of your header bidding solution While the integration of a C2S solution
is also a participant in the auction. For with, for example, the open source
more on this issue, see Chapters 5 on Prebid can be handled quite easily by
wrappers and demand partners (page an experienced team, an S2S solution is
82). more challenging.
4) Cookie Syncing – Cookie syncing
is absolutely critical in programmatic
advertising and has a huge impact on
the unit cost of CPM bids a publisher
receives. Simply put, advertisers are
much more interested in targeting ads
to users about whom they have more
information. While this is not an issue in
the C2S web environment, it is some-
thing that needs to be addressed on the
server side. On the other hand, cookies
are slowly disappearing and will not be
used at all after 2023. This point will no
longer be a major advantage when com-
paring C2S and S2S. For more informa-
tion on this issue, see Chapter 8, page
120 on cookie and User ID synchroniza-
tion.
 55

A simplified comparison:

Client-side header bidding Server-side header bidding

Fully transparent Less transparent

Slower Faster
Fewer demand partners yet, but more
A greater number of demand partners
important
No technical limit on the number of demand
Technical limits on the number of partners
partners
Easier implementation More challenging to implement
Easier in-house development of custom More challenging in-house development
solutions of custom solutions

The rise of S2S

At the end of 2020, according to
Statista.com28 the usage of individual
header bidding solutions was as follows:

Leading header-bidding wrappers

80%
71%
70%
62%
60%
50%
40% 33% 31%
30%
20%
10%
0%
Google Open Bidding Server Amazon TAM server-Side PubMatic Index Exchange
Side vIntegration Wrapper Integration Wrapper OpenWrap Wrapper

Figure 4.5: Leading header-bidding wrappers according to publishers in the United

States as of November 2020. Source: Statista.com

28
https://www.statista.com/statistics/1227654/wrappers-usa/
56

Google and Amazon clearly dominate

with their S2S solutions. Yet just three
years earlier, eMarketer reported29 a
completely different situation in the US
market:

US Websites that Use Header Bidding, by Type,

Sep 2017-Feb 2018
Sep Oct Nov Dec Jan Feb
2017 2017 2017 2017 2018 2018

Client-side wrapper 42,6% 43,2% 40,4% 41,6% 40,1% 39,0%

Server-side wrapper 5,8% 6,2% 5,8% 6,1% 5,8% 6,3%

Hybrid wrapper 13,6% 14,8% 17,7% 18,1% 20,7% 21,1%

No wrapper 7,7% 6,7% 7,7% 6,2% 6,7% 6,7%

Total 70,0% 71,0% 71,6% 72,0% 73,3% 73,1%

Figure 4.6: US Websites that Use Header Bidding. Source: eMarketer.com

While S2S has been gaining more sup-

port, its massive growth is mostly owed
to the dominance of Google and Amazon
in the online advertising market. S2S‘s
leap in growth is therefore more likely
due to the fact that these two demand
partners can only be integrated by
publishers using S2S tools from these
companies and are not available in the
C2S space.

29
https://www.emarketer.com/content/more-publishers-are-using-a-hybrid-header-bidding-approach
 57

Hybrid header bidding DoubleClick for Publishers (DfP)), which

is the most widely used ad server solu-
C2S and S2S solutions each have their tion globally. It integrates the Google
advantages and disadvantages, but it Adx network, which includes, among
is possible to combine the two, into a other things, all of the Google Ads (pre-
Hybrid solution, the operation of both viously AdWords) advertiser offerings,
solutions at the same time. The publish- as well as the larger DV360 (previously
er can then choose which demand part- DoubleClick) advertisers.
ners are involved in C2S and which in We’ve already discussed how the whole
S2S. He is free to involve some of them idea of header bidding originated as
in both, but this puts a burden on the a hack to give publishers better con-
whole system and carries possible risks. trol over advertising from non-Google
But in some cases, you have no other demand partners. These partners were
choice or the benefits clearly outweigh previously waterfall partners in the ad
the potential issues. server and did not have access to fa-
It is also not uncommon to have a com- vorable positions in the auction process.
bination of several different solutions to Ironically, this hack caught on so much
gain as high profit as possible; publish- that it forced Google to come up with its
ers use Google and Amazon with their own solution30 called Exchange Bidding
S2S solutions and then other Hybrid in Dynamic Allocation (EBDA). The prod-
solutions in which they plug in other de- uct was later renamed Open Bidding.
mand partners like AppNexus, Pubmat-
ic, OpenX, Criteo, etc. whether it‘s using
C2S or S2S solutions with some kind of
wrapper. Open Bidding
For more on the possibilities of combin-
ing multiple header bidding solutions, This is a server-side header bidding
see Chapter 7 on ad server (page 111). solution where external demand part-
ners compete with Google AdX. The
solution is entirely in Google‘s hands
and, as mentioned earlier, S2S solutions
Google‘s solution are closed and non-transparent.
In addition, access to other demand
Google has the advantage thanks to its partners is limited by Google and the
Google Ad Manager product (previously

30
https://www.blog.google/products/admanager/announcing-exchange-bidding-open-beta/
58

publisher does not have as wide a range Advantages and

of partner options as it has with Prebid,
for example. disadvantages of Open
The publisher just selects demand part- Bidding
ners31 and agrees on their participation
A common question from publishers is
in Open Bidding. Further engagement
whether to engage Open Bidding32 or
and invoicing is done through Google.
other header bidding solutions and what
However, if a publisher wants to include the differences are between them. The
Google‘s demands in their header bid- answer is: engage both if you can. Just
ding system, then they have no choice as you should consider engaging other
but to use the Dynamic Allocation func- solutions such as Amazon.
tion in Google Ad Manager (ad server),
Let‘s take a look at a comparison of
which is not a classic header bidding
Open Bidding and traditional header
(see details on page 57).
bidding:
It‘s also worth noting that regular de-
mand partners send money to publish-
ers for ads sold after a longer period of
time than Google does, where payment
within 30 days is common.

Open Bidding Header bidding

Client-side and Serv-
Integration type Server-side
er-side
Many options for
Partner Google wrappers and custom
development
Complexity of integration Simple Challenging
For Google Ad Manager
Availability For anyone
360 clients only
With each demand
Payments United through Google
partner separately

31
https://support.google.com/admanager/answer/7128453
32
https://support.google.com/admanager/answer/7128958
 59

As we have already mentioned, Open Amazon’s solution

Bidding, is only an S2S solution. We
could find even more differences be- Amazon is the second pure server-side
tween the two depending on whether header bidding operator next to Google.
we compare C2S or S2S. And, like Google, it has a huge number
of distinctive advertiser offerings due
to its unique market position. It also
has a large amount of data about its
Dynamic Allocation users that it can use for ad targeting,
and therefore bring higher eCPMs to the
publisher.
The Dynamic Allocation feature Amazon regularly ranks among the top
was launched by DoubleClick long
demand partners in surveys of this type
before the creation of header
and should not be left out of your mone-
bidding33. Its essence is to give the
Google ad server (Ad Manager) the tization solution. On the other hand, its
last chance to get an impression reach is not global and its offerings are
for a bid from a Google AdX source not available in some regions.
if it has already decided to assign it Again like Google, Amazon won’t give
to another campaign. you a connector to another header
bidding solution–it only operates on its
own platform and allows other partners
Dynamic Allocation allows Google to to plug into its solution. In doing so, it
evaluate the final offer and decide does not interfere in any way with your
whether it wants to outperform it. So current header bidding solution, if you
technically, Google AdX bids can out- are already using one.
bid the final winning bid from header
Amazon offers two versions of its solu-
bidding. Sao it competes directly with
tion.
header bidding. The problem is that the
decision mechanism is completely hid-
den and entirely in the hands of Google
Ad Manager.

33
https://www.accc.gov.au/system/files/Google%20-%20Report%20from%20Daniel%20Bitton%20and%20
Stephen%20Lewis%20%285%20May%202020%29.pdf
60

Unified Ad Marketplace the publisher gets a user interface with

statistics where they can see individual
(UAM) demand partners and relevant data.
Auctions take place on an Amazon serv-
er that is connected to other demand
partners. Unlike UAM, the relationship with indi-
vidual demand partners is in the hands
Like Google Open Bidding, Amazon UAM of the publisher, including the auction
is a marketplace that acts as a single setup. Thus, it is a similar principle to
entity towards the publisher, working the one we know from Prebid, for exam-
with other demand partners internally ple.
and directly paying the publisher ad
revenue without letting them in on the
details of their internal decision-making.
In fact, the whole system is very simple
and the publisher does not have to con-
figure much, as most things are preset
by Amazon.
UAM is a closed product, so it does
not provide any data on its operations,
which can be quite a complication for
publishers if they want to have all their
data in one place.

Transparent Ad
Marketplace (TAM)
This solution is suitable for larger
publishers and Amazon offers data
transparency to reveal how its S2S
solution works. For the TAM solution,
 61

The difference
between UAM a TAM
Amazon itself describes the difference
between TAM and UAM as follows34:

Unified Ad Transparent Ad
Marketplace Marketplace
Publisher signs one click-
Publisher signs contracts
through contract with
Agreement with SSP’s directly with each SSP
Amazon and Amazon sign
and also with Amazon
contracts with SSP’s
Amazon combines earn- Each SSP pays the Pub-
ings from Amazon and lisher directly, Amazon
Payments
SSPs, sends one monthly pays the Publisher for
payment Amazon demand
Header bidding method-
Server-to-server Server-to-server
ology

Video header bidding counted for about 40% of total spending

in 2020 and at this rate will become the
Video is one of the fastest growing strongest advertising channel in a few
types of advertising. According to the years.
IAB Europe AdEx Benchmark 2020 Re-
port35, video ad spending grew by over
10% year-on-year in 2019/2020, while
traditional display declined by 1.6% (the
Covid 19 pandemic had a hand in the
data). Meanwhile, video already ac-

34
https://aps.amazon.com/aps/2018/03/05/unified-ad-marketplace-uam-vs-transparent-ad-marketplace-tam/
35
https://iabeurope.eu/knowledge-hub/iab-europe-adex-benchmark-2020-report/
62

Europe: Display Ad Spend excl. Social (€bn)

€15,6bn (+2,5%)
-1,6%
12
10,3 10,1
10
8 +10,1%
6 4,7 5,2

4
+16,7%
2
0,4 0,5
0
Traditional Video Audio

Figure 4.7: Display Ad Spend excl.Social. Source: IAB

Header bidding for video differs only possible to use video advertising in a
minimally from header bidding for banner without the publisher having
display formats. The principle of the video content.
auction is basically the same, but the
technical implementation of the vid-
eo differs. The increase in revenue
from video advertising may also differ Types of video ads
substantially. This is mainly due to the
following reasons: Video advertising can be divided into
the following categories:
● The unit price of video advertising
(CPM) is usually significantly higher ● Instream – The ad is played with
than for display advertising. the publisher’s video content in a
special video player. Depending on
● When using a combination of display the placement, the formats are: at
and video advertising, it is possible the beginning of the video content
to significantly increase the CPM of a (preroll), during (midroll) or at the
given advertising unit. It is therefore end (postroll).
 63

● Outstream – A video ad is placed ly spaces, often hosted directly by the

where there is no video content. video player (i.e., the player includes a
Thus, the video player is only used to header bidding wrapper), instead of a
play the video ad. single auction. This is easier for them
to integrate, as the video player takes
care of everything needed and no inte-
gration with the header bidding hosting
How the auction works the display area auctions is required.

The video auction does not take place However, it is two auctions, i.e. two
in the video player, but in the header of scripts, thus doubling the administration
the page (in the <HEAD> tag) where and data intensity of the scripts used
the video is placed, unless the video which leads to longer page load times.
player includes a wrapper to control So, for larger publishers, handling all
header bidding (for more information on auctions in header bidding separate-
what a wrapper is, see Chapter 5, page ly, including video units, and add the
82). auction result to an independent player
Header bidding can be used in the form should be more appealing.
of C2S or S2S with the same benefits
and limitations as previously described.
Just be aware that video advertising is
usually significantly more data-intensive
than display advertising. Thus, upload-
ing it may reduce the loading speed of a
web page, but this is a video issue and
not a header bidding issue.
Video header bidding lags slightly
behind publisher adoption relative to
display ad space36 mainly due to the
greater technical complexity of inte-
grating with a video ad server or video
player.
Many publishers without their own video
ad server have taken the route of pro-
viding a dedicated auction for video-on-

36
https://pubmatic.com/wp-content/uploads/2020/01/Video_Advertising_Requires_New_Monetization_
Strategies_TLP.pdf
64

Some practical tips for video

header bidding
● Do not start the video auction after the user clicks on the PLAY button. By
that time, the ad should already be auctioned. It depends on the place-
ment of the video on the page, whether you run the auction directly in the
header of the page (with a display ad) or follow up with an action (such as
a user scrolling to the video player).

● Make sure you know which of your demand partners can handle video and
in what format. Not all demand partners from display areas are worth
engaging in video auctions. Conversely, there may be other interesting
demand partners for video that won’t make sense to include in display
auctions.

● Video advertising, like display advertising, can be renewed after a cer-

tain period of time. For more on the renewal feature, see Chapter 12 on
Viewability (page 174). When using it, keep in mind the increased volume
of data that video involves.

● Limit the length of the video ad for demand partners, or allow the user to
skip the ad in the player. Otherwise, your users might be forced to watch
long ad videos.

● Consider whether you want to show video ads to mobile users. In some
locations, video data usage can be a major issue.
 65

Solution for AMP sites and

Real-time configuration (RTC)
The purpose of Accelerated Mobile evaluated in an auction. This bidding
Pages (AMP) technology is essentially also has the same evolutionary de-
to make web pages load faster. This velopment and succeeds the waterfall
is done by eliminating a large number principle.
of elements that regular HTML pages Even this in-app bidding is based on
contain, including JavaScript. the header bidding principle. One tool
In 2018, the creators of AMP came for mobile app developers is the Prebid
up with Real-time configuration (RTC) SDK39 built as a classic S2S header
technology37 designed specifically to bidding solution (see Chapter 4, page
address the header bidding of AMP 51 for more on S2S solutions).
pages. With it, you can engage a max- Because it is not “header” bidding, sev-
imum of five demand partners with a eral other labels are used for this type
maximum response time of 1 second. of auction in the mobile environment.
RTC technology is developed specifi- For example: in-app bidding, unified
cally with speed in mind, but there is bidding, parallel bidding, etc.
also the option to use a classic header
bidding wrapper that has built-in AMP
support. By definition, it is necessary
to use S2S header bidding. AMP is also
supported by the most widely used
open source, Prebid38.

In-app bidding
Technology similar to header bidding
can also be found in mobile applica-
tions. It is not “header” bidding, be-
cause apps do not have any header, but
the principle is basically the same. Bids
are sent to all the demand partners
at the same time and their bids are

37
https://github.com/ampproject/amphtml/blob/main/extensions/amp-a4a/rtc-documentation.md
38
https://docs.prebid.org/formats/amp.html
39
https://prebid.org/product-suite/prebid-mobile/
66

Within the context of mobile apps, it

is vital to understand their importance
in terms of the time users spend with
them. The Mobile Trends 2020 study40
shows, for example, that since 2019
people in the US have spent more time
with mobile devices than in front of
a TV screen:

TV and Mobile Devices: Average Time Spent in the US,

2014-2021
4:20
4:10 4:05
3:56 3:54
3:44 3:49
3:43

3:35 3:35 3:29

3:25 3:22
3:08
2:49
2:32
TV Mobile devices

2014 2015 2016 2017 2018 2019 2020 2021

Figure 4.8: TV and mobile devices: average time spent in the US. Source:
eMarketer

And at the
hrs:mins persame time,
day among they overwhelm-
population
ingly spend their time on apps rather
than the traditional internet browser:

40
https://on.emarketer.com/rs/867-SLG-901/images/Branch-Mobile%20Trends%202020%20Report%20
Sponsorship.pdf
 67

Mobile App vs. Mobile Browser: Average Time Spent

in the US, 2017-2021
3:06 3:12
2:57
2:47
2:33

0:27 0:26 0:26 0:25 0:25

2017 2018 2019 2020 2021

Mobile app Mobile browser

hrs:mins per day among population

Figure 4.9: Mobile app vs. mobile browser: average time spent in the US. Source:
eMarketer

Of course, advertising spending predic-

tions correlate to these findings:

US Digital Ad Spending on Select Channels, 2019-2023

2019 2020 2021 2017 2022 2023

Mobile $99,21 $120,21 $138,43 $152,93 $152,93 $166,67

Desktop/laptop $23,20 $22,03 $23,04 $22,34 $22,34 $21,04

Connected TV $6,94 $8,88 $10,81 $12,49 $12,49 $14,12

billions

Figure 4.10: US digital ad spending on selected channels. Source: eMarketer

68

It can be seen that mobile advertis- As we spend much more time in apps
ing, and therefore in-app bidding, is than in browsers, there are many more
a bigger financial topic than header ad impressions. In addition to the
bidding in the classic web browser classic ad performance units such as
environment. It is interesting in this impression, click, revenue, CPM, CTR,
context to look at Facebook as the etc. mobile apps bring other important
demand partner. It makes its Facebook metrics. Probably the two most import-
Audience Network primarily available to ant ones:
the mobile app environment. After all, ● ARPDAU – Average Revenue Per Daily
Facebook users primarily engage with Active User
the site on mobile apps41.
● CPI – Cost Per Installation
However, the mobile app environment
is not as open as the browser environ- Ad refreshes also play a much more
ment and the adoption of bidding tech- important role here, which we describe
nology here has taken a little longer. in Chapter 12, page 174.
In-app bidding is basically similar to
header bidding, but adapted to the
It’s important to note that while
advertising is a major source of mobile app environment and different
revenue for web publishers, in user behavior. As with traditional head-
apps it is much more important to er bidding, developers have several
generate revenue from installs or options when using bidding in applica-
registrations, whether in the form tions:
of a one-off payment or a regular
● Custom development over an open
fee.
source solution. For example, the
Prebid wrapper.
● A ready-made solution from a special
It’s important to note that while adver-
partner built on top of open source.
tising is a major source of revenue for
web publishers, in apps it is much more ● A ready-made solution from a
important to generate revenue from special partner built on a custom
installs or registrations, whether in the (proprietary).
form of a one-off payment or a regular
fee.

41
https://www.statista.com/statistics/377808/distribution-of-facebook-users-by-device/
 69

Setting the auction type by price

In the first part of this book, we dis- If other types of auctions happen, it
cussed the difference between first may be that even in header bidding the
price and second price models in RTB highest bid is not guaranteed to win.
auctions (Chapter 3, page 34). Imagine a simple situation – in header
Header bidding complicates this issue. bidding you have only two SSP demand
As mentioned, header bidding is es- partners and each has only two con-
sentially an auction with bids that have nected DSP systems.
already been through other auctions
(within the demand partners’ system).

Auction using second and first prices

DSP #1

$11
Publisher SSP #1
header bidder wrapper
1st price auction DSP #2
clearing price: $6,01

$3
2nd price auction
clearing price $3,01

DSP #3
SSP #2
$7,5

DSP #4
2nd price auction
clearing price $6,01 $6

Figure 4.11: Auction using second and first prices

70

Let’s say there is a second price auc-

tion inside each of these (Figure 4.11). Therefore, distortions may occur
The winning bid from SSP1 is $3.01 and during repeated auctions. But if you
from SSP2 is $6.01. also consider the length of the path
that each bid has to travel between
You then execute a first price auction systems, it is inevitable that they will
in header bidding and mark the winning go through repeated auctions. For
bid from SSP2 at $6.01, which would be more on the issue of bid supply path
partner DSP3. length, see Chapter 10, page 148
But if you examine Figure 4.11, you will
see that: If we change the auctions in the SSP
● The winning DSP is not selected systems to second price and the header
correctly, the highest bid was from bidding to first price, we get a winning
DSP1. bid of $7.51.

● The winning price is not correct

according to either the first price or
the second price model.

Auction using first and second prices

DSP #1
SSP #1
$11
Publisher
header bidder wrapper
DSP #2
2nd price auction
clearing price: $7,51
1st price auction $3
clearing price $11
DSP #3
SSP #2
$7,5

DSP #4

1st price auction $6

clearing price $7,5

Figure 4.12: Auction using first and second prices

 71

Understanding what type of auctions your partners are working with can make a
difference to publishers and help them set up their auctions correctly. Pubmatic 42
or OpenX43 are transparent in this regard and allow you to work with auction type
settings

Floor price settings

The issue of floor price is not compli- but if you have, for example, seven
cated (see Chapter 3, page 34 for full demand partners, then you have to do
discussion). Simply put, the publisher the same operation seven times. This is
sets a minimum price and will not ac- time consuming.
cept bids below it. Additionally, you have to consider
What complicates matters is that the whether you set the same floor price
floor price can be set at several levels. for all of them, or if there are cases
We will describe the three most import- where you deliberately set them differ-
ant ones: ently.

It often happens that over time,

the floor prices of the individual
Floor price at the level demand partners start to be
of the demand partner different. Maintaining them over the
long term is time-consuming and
Most demand partners allow you to organizationally demanding.
set the floor price. The problem is that
the publisher uses multiple demand
partners and has to set the price for
each demand partner separately. Each
demand partner has a different admin-
istration interface and the publisher
needs to be able to manage each one
separately. It’s not very complicated,

42
https://pubmatic.com/wp-content/uploads/2017/08/PubMatic-UnderstandingAuctionDynamics.pdf
43
https://docs.openx.com/#understanding-first-price
72

Floor price at header

bidding level Rule Priorities
The header bidding is the central place When setting up floor price rules, be
where the auction takes place between aware of the order in which they are
the demand partners. It is therefore applied. This is a tree structure:
possible to set the floor price in the
● The ad server has the highest priority
header bidding administration. Most
and makes the final decision on all
wrappers allow this (see Chapter 5,
demand partners, regardless of the
page 82).
method of engagement
If you set a central floor price at the
● The header bidding wrapper decides
header bidding level, then consider
only on all demand partners involved
whether it makes sense to continue to
in header bidding
administer the floor price in the sys-
tems of individual demand partners. ● Demand partner decides only on what
it sends to header bidding

It is possible to set floor price rules

Floor price at ad server at the ad server level to cover
level everything you need and save you
time by managing multiple sites.
If you have header bidding connected However, this may not always meet
to the ad server (see Chapter 7, page all the requirements you have. Think
106 for more information), then you carefully about your overall setup
can still set the rules for floor price in before you start arranging floor
the ad server. pricing. Careful consideration will
save you a lot of time in the future
The ad server controls not only the bids management of the whole system.
received from header bidding, but also
other bids. The floor price settings can
therefore be set more generally at this
level. Each ad server handles this issue
differently, but we’ll highlight Google’s
“unified pricing rules”44, which can
handle floor price rules for all demand
partners from a single central point,
both at the level of header bidding and
Open bidding (see page 57), Google
AdX, and other sources.

44
https://support.google.com/admanager/answer/9298008
 73

Post-bid
Header bidding is an excellent solution See detailed technical information on
with many advantages, but as we have the Prebid website45.
shown, it also has its disadvantages.
The fundamental question that every-
one who wants to implement header
bidding asks is the technical complexity The post-bid principle
of integration and operation, including
the issue of page latency (see Chapter Post-bid performs auctions between
11, page 164). individual demand partners, but it does
not do so when the page is loaded (in
However, there is also a type of header the header), rather after the impres-
bidding that is significantly simpler and sion is processed by the ad server.
behaves more like a classic ad tag in Thus, the ad server is the trigger of the
terms of implementation. That is, you whole process and no special scripts
just deploy it to the ad server and wait are needed in the page header
to see what it returns.
Let’s look at this principle in a diagram:
This is post-bid.

Post-bid: The principle of setting

Page Ad Server Line Items Creatives

Class1 Line items

Adx

Post-bid

Figure 4.13: Post-bid: The principle of setting. Source: Prebid

45
https://docs.prebid.org/overview/what-is-post-bid.html
74

A post-bid is therefore one ad line item time, but all at once in an equivalent
among others in the ad server. It is auction.
therefore the ad server that decides See Chapter 7, page 106 for more
whether and under what conditions it information on the ad server function-
decides to activate the campaign. ality of header bidding.
Header bidding always takes place in
the header of the page, then passes
the information about the winner to the
ad server to decide whether to use it Advantages and dis-
or not. The ad server therefore already
knows the winner and the price of the advantages of post-bid
winning bid at the time of the header Benefits:
bidding decision.
● Post-bid is faster than header bidding
In the case of a post-bid, it is up to because it does not execute any
the ad server to activate the campaign scripts in the header of the page.
when it sees fit and without knowing Technically, it is a much simpler
in advance what price it will bid. So, as solution.
a rule, the administrators set the ad
● It is a relatively simple JavaScript -
server to the average eCPM price that
ad tag.
the post-bid has historically achieved,
so that they have a price for the ad ● It reduces page loading speed issues.
server to decide on. ● It performs a full auction between the
The post-bid behaves in the ad serv- participating demand partners.
er like any other third-party ad tag. ● We can use the passback setting.
That is, it is JavaScript that is called to
perform the action and return the ad.
As with other third-party tags, you can Disadvantages:
work with things like passback (Chapter ● Ad server does not know the winner’s
2, page 29, which is activated when a price in advance and therefore cannot
post-bid fails to deliver any result. compare it to other campaigns it
If the post-bid call method reminds manages. When it decides to run
you of waterfall, you’re not wrong. But a post-bid, it has already decided
unlike waterfall, post-bid doesn’t call on the winner. Header bidding is
individual demand partners one at a therefore a better solution in this
 75

respect, as it allows the ad server to auction were to be done for multiple

make better decisions about which ad units on one page, then it would
campaign to run. be technically less demanding to do
● In header bidding, it is possible to get only one joint action in the classic
a lot of data about the progress of header bidding.
the auctions and the competing bids.
This is done in bulk in one place. For
more on this area, see Chapter 6,
page 96. With post-bid, nothing like Comparison of post-bid
this is possible. The only statistics
and header bidding
available to you are those provided
by demand partners. A comparison of post-bid and header
bidding can be seen in the following
● While header bidding performs
diagram:
auctions for all ad units at once in
the header of the page, post-bid
performs a separate auction for each
ad unit for which it is called. If the

Difference between post-bid and header bidding

Header Bidding Post-bid

Website Website

Prebid.js Ad Server

SSP1 SSP2 SSP3

Direct sale AdX line Post-bid
line item item line item
Ad Server

Direct sale AdX line Prebid line

SSP1 SSP2 SSP3
line item item item

Figure 4.14: Difference between post-bid and header bidding. Source: adpushup.
com
76

A comparison in a table might also be useful:

Header bidding
Post-bid
(Pre-bid)
The auction is being JavaScript in the page
Ad server
launched header
Calling ad server After header bidding Before post-bid
It is not known in ad-
It is known in advance vance and the ad server
Price of the winning bid
and delivered ad server triggers post-bid based
on predefined settings
The ability to compete
with the price against
Yes No
other campaigns in the
ad server
A separate auction is
One auction in the head- performed for each post-
Number of auctions
er for all ad units. bid call, i.e. for each ad
unit separately
Complexity of technical
More complex Simpler
implementation
Loading speed Depends on many factors Fast loading of one ad tag

Footer bidding
If you’ve ever analyzed the loading closely related to header bidding and
speed of a web page, you’ve probably affects the final return for publishers.
found out that ads are one of the most What if we don’t load ads simultane-
common causes of slowdown. This is ously with the page content, but after?
obviously a pressing topic and the solu- We may lose some revenue, but we
tion is not simple. We cover it in Chap- will significantly speed up page load-
ter 11, page 164 because speed is very
 77

ing, which has a positive effect on the

user’s experience and will be appreciat- If you think about it more, you will find
ed by search engines. In the long run it that the idea of footer bidding opens
can bring increased traffic and there- up new possibilities. For example,
fore an increase in advertising volume. you might consider increasing the
number of your demand partners
Activating ads after the rest of the con- if you were previously limiting
tent has loaded therefore makes sense them because of page load speed
from a speed optimization perspective. concerns. This issue is also closely
This is called footer bidding. related to lazy loading, viewability,
and ad refreshes, which we cover
Footer bidding is a term for a specific
again in Chapter 12, page 174.
type of header bidding which is activat-
ed after the actual page content is up-
loaded. The JavaScript itself (for exam- How much of a loss can occur? The an-
ple, the Prebid solution) is still placed swer is not simple and always depends
in the header of the page, but it is only on the specific site and, more impor-
activated after the entire page content tantly, the loading speed of the other
has been loaded. This activation must elements on the page. Interesting
be provided by your developer. data presented by Admonsters.com46.
Thanks to proper implementation, the
In order for footer bidding to make
drop in revenue was minimal.
sense for you, you need to solve all
other page speed issues first. Other-
wise, footer bidding will only ensure
that your ads start to activate after the
slow page has taken too long to load
and you will lose a significant amount
of revenue.
But as we’ve already mentioned, footer
bidding has one major drawback–it’s
likely to decrease your ad revenue. So
this is more of a long-term strategic
move that should pay off in the future
if you do it right.

46
https://www.admonsters.com/what-is-footer-bidding/
78

Benefits of header bidding

for publishers
The list below could be called “bene- to solve is the SSP sending a bid at a
fits”, but as is often the case in prac- certain amount, but then reducing it
tice, even the best technology, if used because of various fees in the billing.
poorly, can become a barrier, so it 4. Page load speed – Save time
depends very heavily on your setup by simply calling all partners at once -
experience. asynchronously - and setting a max-
imum response time. It is obviously
1. Better control – With wrappers, much faster than waterfall, but here we
we have all demand partners in one ad- must be wary of the technical function-
ministration and we can work with them ality of the wrapper.
at once and partially manage them from
this environment. For example, we can 5. Increase fill rate – This benefit
set a common floor price for all adver- is often stated in the literature, but the
tisers or adjust their bid amounts by the opposite may be true if you don’t use
fees they charge. header bidding carefully. If you call ad
All of this does require some mainte- sources sequentially within a waterfall,
nance, and large publishers may even the final fill rate is often higher. Howev-
employ a separate staff to manage it. er, it all depends heavily on two settings
you make, which are the maximum
2. Increase in revenue – By the response time (i.e. the time before you
nature of header bidding technology discard the request for the demand
and the difference from waterfall, there partner so-called “timeout”) and the
should be an increase in revenue on the floor price. If you set a low floor price
publishers’ side. To review the details, and a long response time, then the fill
see Chapter 2, page 24. rate increases significantly.

3. Unified statistics – Using wrap-

pers, we can see independent data on
ongoing auctions. Thus, we can verify
the data supplied by demand partners
and resolve issues with them in case
of discrepancies. A typical problem
79
80

Benefits of header bidding

for advertisers
Header bidding also brings some ad- all the technologies that will exchange
vantages for advertisers: his bid before it reaches the publisher
(see Chapter 10, page 148)

1. Comparison of adver- He can’t even be sure at what final

tisers’ conditions – It is better for price his bid will compete, or whether
an advertiser if he knows that he is the final bid will compete with itself,
participating in an auction with clear because his bid will be written into two
conditions. Header bidding auctions are demand partners competing against
transparent, especially when compared each other in the auction.
to a waterfall, where the advertiser One way to find out which demand
does not know when his turn comes and partners a publisher uses is with the
whether in the meantime his competi- ads.txt file (see Chapter 10, page 155).
tor would take the offer which would be However, this method is not entirely
even cheaper. accurate and difficult for the average
advertiser to use in practice.
2. Greater reach to premium Basically, the advertiser should be
space – Header bidding forced publish- aware of these unknowns and adjust
ers to make their premium space, which his bids accordingly.
they had previously reserved for a se- At the same time, however, he has the
lected clientele, available. In case of ad- option of approaching interesting pub-
vertiser’s interest, he can offer a higher lishers, directly, and agreeing on the
price and get the premium space. exact way to send his offers. For exam-
ple, the form of programmatic private
deals, which are integrated into the
SSPs that the publisher has involved
What else the adver- in header bidding, could be useful in
mapping out the entire bid path and
tiser should know what factors influence it (see Chapter
Today, it is almost unrealistic for an 10 page 148).
advertiser to control or even know
whether they are participating in a
header bidding or waterfall auction.
Most of the time, he doesn’t even know
 81

Chapter conclusion: Header bidding

In this chapter we: ● explained the concepts of post-bid
● explained the concepts of client- and footer bidding,
side and server-side header bidding ● discussed the benefits of header
and the difference between them, bidding from the publisher‘s
and compared the advantages and perspective and from the
disadvantages, advertiser‘s perspective.
● described the main S2S solutions of
Google and Amazon,
This chapter was the most comprehen-
● discussed that it is possible to sive in this book and should provide a
combine S2S and C2S solutions,
theoretical background for the following
● discussed the specialties of video chapters
auctions, AMP sites and apps,
● discussed the intricacies of first and
second price events,
82

Chapter 5:
WRAPPER

Wrapper can be imagined as an ana- Before the advent of wrappers (Preb-

logue of Tag Manager (e.g., Google Tag id in particular), header bidding was
Manager). It is a few lines of JavaScript relatively problematic and faced major
code inserted into the header of the issues in terms of page loading speed
website, which then calls other func- and complex integration of individual
tions. It can be controlled and edited demand partners. Until the advent of
using online administration. wrappers, each demand partner had
The idea of wrappers is to secure the its own JavaScript that worked inde-
header bidding functions and give the pendently of the others, which resulted
publisher a simple interface for con- in slow page loading. Wrappers have
trolling it. So instead of a programmer, standardized this process and today,
header bidding can be handled by a most demand partners have created
yield manager, for example. connectors to the most widely used
wrappers.

Types of wrappers
There are currently many wrapper ● Proprietary – This type is developed
solutions, both open source and com- by a private entity and provided
mercial. Wrappers can be divided into for a fee (usually a monthly fee
three areas: + an impression fee). They are
● Open source – These are not transparent and independent,
independent, free, and developed but the provider solves many
by a wider independent community. technical problems and takes care of
Examples include Prebid or maintenance.
Pubfood. Their use requires greater ● Managed – A proprietary or open
development expertise including source solution, but managed by
subsequent, complex administration. experts in the field for a fee. The
However, for experienced developers largest offerings are from Prebid-
this is not a problem. based solution providers47.

47
https://prebid.org/product-suite/managed-services/
 83

Most used wrappers

Any publisher can use one or more
wrappers. And this is often the case in
practice, as shown by research from
the US market published on ROXOT.
com in 201948.

Demand channels combinations publishers use

56 %

27 %

6%

Google Adx Google Adx Google Adx

Client-side header bidding Client-side header bidding Client-side header bidding
Amazon Amazon Exchange Bidding
Exchange Bidding

Figure 5.1: Demand channels combinations publishers use. Source: ROXOT

As the attached chart shows, most

publishers are indeed using multiple
header bidding solutions. (The data is
from 2019 when the name Exchange
Bidding in Dynamic Allocation was still
used instead of Open Bidding for Goo-
gle’s solution.)

48
https://roxot.com/report/sell-side-programmatic-us-2019
84

As mentioned in Chapter 4, page 55,

However, if you decide to run multiple the integration of S2S tools from Goo-
wrappers at once, you should have a gle and Amazon is gaining popularity,
good understanding of their features even in parallel (i.e. they work togeth-
and operational costs (especially er on the page independently). At the
technical requirements).
same time, however, they are often
As you can see from the chart above, accompanied by one or more clas-
for larger publishers this situation sic wrapper solutions such as Prebid,
is quite common but mostly with whether in C2S, S2S or Hybrid ver-
S2S solutions. After all, running sions.
multiple C2S solutions is technically
challenging and can put a significant According to the Adzerk Header Bidding
burden on your website. Index49, the open source solution built
on Prebid.js has absolute dominance in
the number of implementations in Q3
2021:

Prebid's share of the number of implementations of open source

solutions in use (Q3 2021)

Proprietary 28 %

72 % Prebid

Figure 5.2: Prebid’s share of the number of implementations of open source

solutions in use. Source: Adzerk

Of the 28% of proprietary wrapper solutions, Index Exchange dominates:

49
https://www.kevel.co/hbix/
 85

Share of proprietary wrappers in the number

of implementations (2020)

30,5%
8

2 4% 3,7% 3,4% 2,9% 2,1%

0%
Index 3Across Sovrn Freestar Rubicon Prebid
Exchange Project Enterprise

Figure 5.3: Share of proprietary wrappers in the number of implementations.

Source: Adzerk

Main functions of the wrapper

There are several wrapper solutions, as ● price adjustment of bids
we’ll see in the next chapter, and they ● connection of other functions
all have several functions and features
in common: ● floor price definition

● common timeout ● special rules for displaying ads

● asynchronous loading
● connection to ad server Let’s take a closer look at all of it.

● easy management of demand

partners
● transparency and independent
monitoring
86

Common timeout
The more demand partners you use, number of line items (up to tens of
the more you run the risk that some thousands) in the ad server. Doing
of the demand partners won’t respond something like this manually is at the
fast enough and the whole page will limit of human capabilities. Wrappers
take significantly longer to load. Ex- can usually automate this function and
tending response time is one of the create line items in bulk. For more de-
things to watch out for with wrappers. tails, see Chapter 7, page 106.
Setting a central timeout that is com-
mon to all partners solves this prob-
lem.
Easy management of
demand partners
Asynchronous Wrappers usually have a predefined
interface for adding new demand
functionality partners so it can be done by a yield
This is one of the main features of manager in minutes without any pro-
wrappers. It ensures that all demand grammer intervention. Similarly, he
partners are called asynchronously, can simply remove a demand partner.
i.e., independently or simultaneously However, you do need to make sure
(or side by side). If, on the other hand, that the wrapper you are using is com-
the demand partners were called syn- patible with all your demand partners.
chronously (in succession), the entire Sometimes it happens that you want
page load would be disproportionate- to plug in a new partner and it doesn’t
ly longer and, in the case of a large have a connector to your wrapper yet.
number of called sources, significantly Here you can no longer do without the
longer than in the case of waterfall. help of developers. The situation is still
made easier by the fact that, thanks
to the wrapper, it is usually a well-doc-
umented environment with clearly
Connection to the ad defined standards.

server
For proper functionality of header bid-
ding, it is necessary to create a large
 87

Transparency and
independent monitoring
Most wrappers give you a lot of in- looks like the highest bid may win, but
formation about how header bidding after the operating fee is deducted, it
works. You will no longer have to col- will no longer be the highest. Within
lect data from all the demand partners the wrapper, you have the ability to
in different formats and try to con- adjust bids before you send them to
solidate it like you would with header auction. Set each demand partner’s bid
bidding for further work with them. reduction to be equal to its operating
Within the wrapper, you will have all fee. This is usually an across-the-board
relevant data in one place measured percentage reduction.
by the wrapper, so you don’t have to
rely on what the demand partners are
telling you. However, it is very useful to
occasionally do comparisons between
the wrapper data and the demand
Connection of other
partners’ ones. There will always be functions
differences between the two – usually You can integrate other technologies
a few percent. If you find a significant and transfer information in bulk within
difference, you should inquire into the wrappers. Examples include collecting
reasons for it. Thanks to that, you will consent from users to use their data
also get a new data set of all bids, (see Chapter 9, page 142), analytics
including non-winning bids, which will tools, and many other modules and
give you another interesting view of the connectors. See the list of options for
functionality of the demand partners Prebid50.
involved and allow you to configure
them better. See Chapter 6, page 96
for more details on statistics.

Floor price definition

Within the wrapper, it is usually pos-
sible to set a cascade of minimum bid
Price adjustment amounts (floor prices) for not only
of bids the entire site, but also in much more
Most demand partners send their bids detail down to the level of sub-ad units.
at the gross price, but then deduct a The floor prices set in this way are valid
commission for their operation. This for all demand partners.
skews the auction because a bid that

50
https://docs.prebid.org/dev-docs/modules/
88

Special rules for

displaying ads
Not all wrappers can handle refresh,
lazy load, responsive ads, and sticky
ads, but some can. We’ll cover many of
these features at the end of Part 3 of
this book (page 174).

The main disadvantages

of wrappers
The benefits of wrappers are enormous ple, you need to remember that the set-
and far outweigh their drawbacks. How- up for a publisher with premium space
ever, it is good to know them and work that sells a large portion of its space to
with them. They are: direct clients will be different than for a
● complicated initial installation and publisher who, on the other hand, relies
setup more on programmatic revenue. This
is not a question for programmers, but
● the need for maintenance and for the sales department, as you need
upgrades to consider your sales strategy with
● barriers between technologies respect to the opportunities offered by
header bidding technology.
Let’s break them down in more detail.

Installation and setup Upgrades

Technology in marketing is evolving
The operation of the wrapper is simple,
at a terrific speed. Even the wrappers
but for the initial setup you need some
and the demand partners involved are
know-how. In addition to knowing the
evolving and innovating. You need to
wrapper itself, you need to analyze your
be very careful and test new features
entire advertising system to choose the
well whenever you upgrade. Due to the
most appropriate form of setup. We’ll
complexity of the entire system and the
cover a few areas below, but in princi-
involvement of multiple technologies,
 89

there is no shortage of problems. With the publisher’s machine. But publishers

open source solutions (such as Prebid) I matter, and the bigger the publisher you
by no means recommend deviating from are, the more strongly you should insist
the official branch of development, or that the solution you use is transparent
you could have serious problems in the to you and not get backed into a corner
future. by some of the demand partners. They
are the parts of the car, but you are the
owner and driver.

Barriers between tech-

nologies
Some players in the market don’t like
to share information or lose control of
their solutions just to become a cog in

Response time control

Ad loading speed is a complex issue. Within the wrapper administration, you
Ultimately, it has a significant impact on can set a timeout, i.e. the time you give
the loading speed of a website, which is to demand partners to deliver their bids,
also a big issue in search engine optimi- between 500 and 2000 milliseconds.
zation. Make the exact settings in relation to
the tests you can perform. For more on
this, see Chapter 6, page 96.
A study51 conducted by the University
The issue of page load speed as a whole
of Crete found that header bidding
is covered in Chapter 11, page 164.
increases page latency by a median
of 0.6 seconds, and in more than
50% of actions, half of the bids
arrive late, reducing the return to
the publisher. So be wary

51
https://arxiv.org/pdf/1907.12649.pdf
90

Involvement of demand
partners
As time goes on, more and more de- This shows the average number of de-
mand partners are interested in get- mand partners involved per publisher,
ting involved in header bidding with with one publisher having an average of
publishers. They create connectors to seven demand partners participating in
the Prebid wrapper and other extended the header bidding auction in 2020 and
wrappers. This is well represented by a only five three years earlier.
comparison posted on adpushup.com52
showing the difference between 2017
and 2020 according to the Adzerk HBIX
tracker.

Average number of demand partners involved

0
2017 2020

Figure 5.4:Avg Adapters Per Site

Average number of demand partners involved. Source: Adzerk

However, it is also helpful to look at the

more detailed distribution provided by
AdPushup’s AdTech Insights Report for
Publisher53:

52
https://www.adpushup.com/blog/header-bidding-adoption-three-years-in-review/
53
https://adpushup.wistia.com/medias/swzqclb2xx
 91

Distribution of the number of demand partners

among publishers

32% 32%
28%

8%

1–3 Partners 4–6 Partners 7–9 Partners 10+ Partners

Figure 5.5: Distribution of the number of demand partners among publishers.

Source: adpushup.com

Thus, only 8% of publishers use more engaging large demand partners like
than 10 demand partners. Amazon, AppNexus/Xander, Pubmatic,
In 2020, there were about 100 possible Criteo, Adform, etc., plus demand part-
demand partners that publishers could ners that have a clear value-add that
engage in header bidding, but they makes sense for you. Don’t be afraid to
used only seven on average. Logic says ask a new partner where their enquiries
we should use the maximum possible are coming from.
sources to increase competition in the
auction, so why only seven? There are
several reasons.
1. Zero-value added of demand
partner – Many demand partners
have no unique offering and only com-
bine other resources that the publisher
already has engaged. We recommend
92

2. Difficulty of engaging a new 5. Technical limits on the num-

source – Even though the wrapper ber of web browser connections
makes it significantly easier to add a and the impact on page load
new demand partner, it’s still not insig- speed – If you are going to be asking
nificant. You should have a contract with for five or twenty-five partners, this will
each partner, do an initial implementa- affect your page load speed. We discuss
tion, verify their data regularly, perform this topic in detail in Chapter 11, page
monthly billing, analyze their runtime, 164.
and edit settings if necessary. Doing
this for seven partners is a lot less work
The world of advertising technology
than for twenty.
is constantly changing and it is
3. Inappropriate technical in- necessary to follow the current
tegration by the demand part- developments and adapt to them.
ner – Some demand partners have Even if you are currently using, say,
very unconventional ways of technical seven demand partners, keep an
integration, which leads to the fact that eye on the market and test. No one
their wrapper connector is significantly can tell you exactly which demand
partner they are and what number is
larger than your wrapper with all the
optimal for you.
other demand partners or they internal-
ly conduct other slow auctions and don’t
manage to respond to your demands
within the time limit.
4. Non-support of some other
solution you use – Header bidding
is no longer about the auction itself.
Wrappers are often connected to other
systems and connecting them can be
a necessity that some ad sources can’t
handle. For example, it may be a con-
nection to GDPR or User ID modules.
 93

Take a look at which demand partners

were most used according to the HBIX
tracker in 2021:

What adapters are header bidders using? (Top 10, 2021)

Amazon: 78%
Xandr: 75%
Index: 68%
Magnite: 65%
Pubmatic: 59%
OpenX: 49%
TripleLift: 38%
Criteo: 31%
Sovrn: 32%
District: 19%

This looks just at sites doing header bidding and shows what % of those sites are

Figure 5.6: Most commonly used query publishers. Source: Adzerk

94

Research on a similar topic was present-

ed on eMarketer.com54 in 2021 about
specifically what demand partners US
publishers used in the last 12 months.

Which SSP have US publishers used in the

past 12 months (March 2021)
Google ad manager 72%
Amazon publisher services 46%
Verizon Media ad platform 44%
PubMatic 44%
Xandr Monetize 42%
OpenX 42%
SpotX 42%
Index Exchange 40%
Magnite 36%
TripleLift 31%

Figure 5.7: Which SSP have US publishers used in the past 12 months. Source:
eMarketer

54
https://www.emarketer.com/chart/248339/which-supply-side-platforms-ssps-have-us-publishers-used-
past-12-months-of-respondents-march-2021
 95

Conclusion of the chapter: Wrapper

In this chapter we:
● explained what the purpose of
wrappers is and how they make
publishers’ jobs easier,
● discussed the main features,
advantages, and disadvantages of
wrappers,
● showed how publishers combine
multiple solutions and the market
shares of each wrapper,
● discussed how many demand
partners publishers typically engage
and why more is problematic
This chapter could assist publishers in
selecting the appropriate wrapper solu-
tion for their situation.
96

Chapter 6:
STATISTICS RELATED TO HEADER
BIDDING
If you’re a publisher who uses an ad If you’re a larger publisher and have
server and works with programmatic your campaigns in an ad server as
buying, you’re sure to use all sorts of well, then certainly think about how to
statistics. With the implementation of connect this data with header bidding
header bidding, your situation in this data to get a holistic view of your sales.
area will change. The wrapper data alone won’t help you
On one hand, you get a new set of data here, you’ll need an API to both the
and will be able to compare bids (even demand partners and the ad server
non-winning bids from demand part- and a tool that can unify and work with
ners) in one place. But you will also be all this data.
faced with the question of how to unify
your data not only from the header
bidding demand partners, but also from
the ad server and possibly from other
systems you use.
Most wrappers today offer some
statistics from the demand partners
involved. For example, Prebid offers a
separate module to connect with Goog-
le Analytics55.
Over time, many tools have evolved
that take a detailed and comprehensive
view of statistics and combine data
from many sources, some of them em-
bedded in wrappers, some built outside
of it as standalone tools. A good step-
ping stone for finding a suitable solu-
tion can be an overview of the modules
in Prebid56.

55
https://docs.prebid.org/overview/ga-analytics.html
56
https://docs.prebid.org/overview/analytics.html
 97

Bids data
One of the most important pieces of The bid request value ranges are usual-
information that wrapper statistics can ly very high.
give you is information about bids. Not Example:
just the winning bids, but all of them.
1 000 000 000 bid requests
This puts a whole new set of data in
your hands for your decision making.
Let’s break down all the important
features one by one. Again, recall the
definition of “demand partner”. This
Bid response
is a system that aggregates demands Bid response (sometimes also referred
and participates in an auction. In the to as ad response) is the partner’s re-
statistics, therefore, we monitor the ply to the bid request, if he decides to
parameters that the individual demand give one. It contains the bid price and
partners supply us with. the advertisement it wants to display if
it wins.
To give you a better idea, we provide
typical example values for each term or There are three possible outcomes
identify from which values it is calcu- of a bid request:
lated. 1. The demand partner sends its bid
response.
2. The demand partner does not re-
spond because they are not interest-
Bid request ed in the bid (no-bid).
This is a request generated by the 3. The demand partner fails to respond
publisher’s system (e.g. ad server) and in time (timeout).
sent to all demand partners, typically,
an offer of impression in a new auction
when a user visits the website. The bid
request carries with it to the auction
all the relevant information about the
impression being offered, according to
which the partners will decide whether
to bid and at what price. See page 96
for more information on the concept of
impression.
98

If the partner responds, then we can Timeout rate

see the amount of his bid (price). On
the other hand, if no response comes As we mentioned earlier, setting limits
(timeout or no-bid), then the publisher for partner response is one of the most
is blind and do not know why it hap- basic and important functions of the
pened. There can be several reasons wrapper (Chapter 5, page 86).
for a non-response: You need to set this value so that as
● the short time limit you set for them many partners as possible can submit
in the wrapper (see Chapter 5, page their bids while keeping it as short as
86), possible to avoid increasing the page
load time.
● slow internet connection,
● faulty integration with the CMP tool
If a partner fails to deliver a bid, even
(see Chapter 9, page 142),
if they are interested, this can have a
● overloading the user’s web browser negative impact on your eCPM. It is
with other tasks, his bid that may be the highest.
● other reasons.
On the publisher’s side, we cannot So you’re looking for an optimal length,
distinguish between no-bid and timeout which you set as the maximum re-
states, so we generalize both states as sponse time at which the vast majority
timeout in the statistics. of partners will manage to deliver their
In order to be able to analyze this bid, but you cut out the crumbs that
whole area, you need to know not only you would wait too long for and would
the information about the winning bid, unnecessarily extend the page load
which is then passed on to the ad serv- time.
er, but you need to know detailed infor- The percentage of responses that do
mation about the bids of all partners. not arrive is referred to as the timeout
The bid response value ranges are usu- rate.
ally very high. It is calculated by the formula:
Example:
Timeout rate = Total timeouts bid
- 500 000 000 bid responses responses / Total bid requests ×100%
- 100 000 000 timeout responses
 99

The value of the timeout rate is calcu-

lated as a percentage. If your bid rate drops significantly
for multiple partners, then there is
Example: let’s say the number of time-
probably a bug in your setup.
out responses in our example is
100 000 000.
Then we get the calculation: 100 000
000 / 1 000 000 000 ×100% = 10%

Timeout rate = 10%

Impression
This is the ad impression counter, i.e.,
the winning bid response. If a demand
partner wins the auction, he has the
right to display his ad in the winning
Bid rate space. Impression is one of the most
Bid rate is referred to as the percent- used metrics in online advertising and
age of bid requests returned with any many business models rely on the
positive response. That is, the partner number of impressions purchased.
is interested in the impression and has
The value of impressions is usually
submitted a bid price. The bid rate is
high, but is always lower than the total
dependent on many factors that are
number of bid requests.
not in the hands of the advertiser, but
often in the hands of the publisher, for Example:
example, integration of various User ID 20 000 000 impressions
modules, integration of CMP tools, or
visibility of the advertising space.
It is calculated by the formula:
Bid rate = Total bid responses /
Total bid requests ×100%

The bid rate value is also presented as

a percentage.
Example calculation: 500 000 000 /
1 000 000 000 ×100% = 50%

Bid rate = 50%

100

Win rate It is calculated by the formula:

Win rate is the ratio of how often the Fill rate = Total impressions /
partner wins the auction (and there- Total requests ×100%
fore wins the impression offered and The fill rate value is calculated as a
displays his ad). That is, how often he percentage.
placed the highest bid among all part-
ners who participated in the auction. Example calculation: 20 000 000 /
It is calculated by the formula: 1 000 000 000 ×100% = 2%
Win rate = Total impressions / Win rate = 2%
Total bid responses ×100%
The win rate is calculated as a percent-
age.
Example calculation: 20 000 000 / Revenue
500 000 000 ×100% = 4% Sometimes referred to as yield, revenue
is the total price paid for advertising. As
Win rate = 4%
a rule, it is not the price per impression,
but the total price for the number of
impressions displayed. When calculating
the price per 1000 ad impressions, the
Fill rate term eCPM is used.

Fill rate is the ratio of impressions sold Revenue value refers to advertising rev-
from your entire offer. Sometimes it is enue and is expressed in currency.
also referred to as buy rate. It indi- Example: $50 000
cates what percentage of your portfolio
you manage to sell out.

If partners are responding (low timeout

rate) and are interested in your impres-
eCPM
sions (high bid rate), but your fill rate is It indicates the price per 1000 ad im-
still low, then you probably have too a pressions. It is calculated as a ratio of
high floor price (Chapter 3, page 33). revenue per 1000 ad impressions and
is the average of multiple CPMs. eCPM
 101

indicates the combined average of all Viewability rate

partner bids per ad impression. In sim-
ple terms, we use CMP for one specific This is the ratio comparing the number
bid, eCPM for the average of multiple of times an impression was viewed to
bids. the total number of impressions.

It is calculated by the formula: It is calculated by the formula:

eCPM = Total Ad Revenues / Total Ad Viewability rate = Total visible im-

Impressions ×1000 pressions / Total requests ×100%

The value of the eCPM is given in cur- The viewability rate value is calculated
rency. as a percentage.

Example calculation: 50 000 / Example calculation: 12 000 000 /

20 000 000 ×1000 = 2,5 20 000 000 ×100% = 60%

eCPM = $2.5 Viewability rate = 60%

Visible impressions Clicks

This is the number of impressions that If an advertisement is displayed (im-
were visible, i.e. viewed by the user (full pression), then the visitor can click on
discussion of viewability in Chapter 12, it. Often this will take them to the URL
page 157). that the advert is promoting. There are
a few measurable parameters linked to
The value of visible impressions is usu- clicks, which we will introduce here just
ally similar to, but always less than, the quickly:
number of impressions.
● CPC – cost per click on an ad
Example:
12 000 000 visible impressions ● CTR – the percentage of how often
the click occurred in relation to the
total number of times the ad was
displayed
102

Examples: Bid Latency

- Number of clicks: 100 000 clicks
- CPC: $0.5 For bids that do not time out, the time
- CTR: 0.5% required to complete the bid request is
called the latency. A summary of laten-
cy by partner allows publishers to view
partners by how fast or slow they re-
spond on average. For more on latency,
Discrepancy report see Chapter 11, page 164.
Discrepancy refers to the difference
between data from different systems.
It can be approached from different
perspectives. For example, compare
the partner data with the data in the
ad server. The consistency will never
be 100%. There may be bid responses
that the partner considers to have been
delivered correctly, but for some reason
they are discarded on the publisher’s
side.

Discrepancy always occurs and

can be revealing to observe in the
context of statistics. For example, if
one partner has 2x the discrepancy
of the others, then you should look
for a reason.
103
104

Evaluating statistics
You should not perform analyses on your site as a whole, if possible, but break
it down into logical sections. With response times, for example, it will depend on
the location of your users. Users from Brazil may have a different response time
(latency) than those from Japan. You’ll also find major differences in response
time when dividing your portfolio by user device (desktop vs mobile) and of course
by specific ad unit (a banner visible after the first paragraph in the blog text will
have a significantly better eCPM than a banner in the footer).

Data can be displayed in many forms.

Try to choose a view that shows the
values you want to compare from all
demand partners in one chart. See for
example:

Sample statistics chart

100%

75%

50%

25%

0%
10/3/2015 10/4/2015 10/5/2015 10/6/2015 10/7/2015 10/8/2015 10/9/2015

Bidder1 Bidder2 Bidder3 Bidder4 Bidder5

Figure 6.1: Sample statistics chart. Source: Prebid

 105

In order to properly analyze it, you need

to put your statistics in context. For
example, you may discover a partner
that is significantly slower than others,
but on the other hand has a significant
number of offers that other partners do
not have. By reducing response time,
you could significantly reduce this re-
source and therefore, as a direct result,
reduce the average eCPM you achieve.
Try doing a few experiments and com-
pare the results.

Conclusion of the chapter: statistics

related to header bidding
In this chapter we: ● described how publishers can
● explained where a publisher can get increase revenue using relevant
interesting auction data and what it data.
is good for,
● discussed the main metrics to track This chapter shows what data can help
and showed what they are calculated publishers increase revenue.
from,
● showed how to evaluate the data
collected,
106

Chapter 7:
AD SERVER

Header bidding was originally conceived Below, we’ll discuss what an ad server
as a hack for Google’s ad server. The is good for and how to use it in combi-
presence of an ad server is therefore nation with header bidding. There are
expected and it is necessary to link many options and before you choose
header bidding and the ad server cor- one, you should think about your busi-
rectly to get the result we need. ness strategy for selling ad space and
only then set up header bidding in the
ad server.
However, some wrappers also
offer header bidding without an
ad server. They simply select the
winner and directly ensure the
winning bid (ad) is displayed on
the page. If for some reason you
don’t need an ad server and 100%
of your revenue is programmatic
advertising, then this is the ideal
solution.

Sales strategy
Over the past decade, the programmat- manager and thereby check off that they
ic ecosystem has evolved to the point have programmatic advertising sorted.
where it now accounts for the majority The technical setup of the entire system
of revenue for many publishers, whether and the interconnection of all its com-
in the form of header bidding, program- ponents is now so crucial (and sophisti-
matic deals, programmatic guarantee cated) that people with this know-how
or other forms of automated sales. The should be in charge of the entire sales
advertising world is shifting away from team.
a direct sales state towards automated The integration of header bidding into
sales. the ad server should reflect your unctiol
Header bidding and programmatic play a business strategy. Below are a few points
key role here. Today, it is almost unthink- you should consider when configuring it.
able for a larger publisher to hire a yield
 107

Ad server Ensure delivery of all orders and when-

ever possible, choose a demand partner
Probably the most used ad server, with a higher eCPM. Generally speaking,
Google Ad Manager, works directly with selling a guaranteed number of impres-
programmatic and allows you to create sions is a complication when choosing
programmatic deals with agencies and the highest price. The more sales you
clients. Other ad servers are independ- get from programmatic, the more you
ent of SSP functionality and provide should adjust the direct store to no
pure ad server functionality. longer sell guaranteed impressions
So from a header bidding perspective, unless they have a significantly higher
you are addressing the question of eCPM.
whether all of the demand partners will
compete in header bidding, or whether
you have some connected directly to
the ad server and therefore will not par-
First look options
ticipate in the auction directly. The ad If your sales include first look options,
server will then compare their bids with the advertiser expects a priority posi-
the final winner of the header bidding. tion in the impression option. Therefore,
This of course also affects the statistics, you should have first look priority spe-
which for some sources you will be able cifically considered. It is certainly not
to see in the ad server, and for some in a good option to leave this setting only
the wrapper. to the respective SSP where you have
such a programmatic deal set up. Be
aware that there are multiple SSPs that
compete with each other and the winner
eCPM then competes with other ad server
What percentage of sales are direct campaigns. The fact that this is a “first
sales versus programmatic? It’s com- look” should be reflected directly in the
mon for direct sales to sell at a higher ad server.
eCPM, so you need their orders to be
handled by the ad server as a priority.
On the other hand, it often sells at a
bargain eCPM but with a guaranteed
number of impressions. The ad server
setup should account for all variations.
108

How the ad server works

An ad server is a tool that allows you to Typical targets of ad server campaigns
insert ads into your site through a spe- can be for example:
cial interface, instead of having to insert ● Deliver 1 million impressions in
them directly into your HTML code. It’s 300x600px format to users in London
usually used when you have a sales and to users at the same time,
team that sells your ad space, so you between 1 January and 28 February,
need a system that will manage these with priority “high”.
ads and give you information about how
● Deliver 5 million impressions in
they are running.
300x600px, or 300x250px format to
The main unctionns of ad servers are: any users between 1 January and 31
● Setting the start and end date of the December, with priority “medium”.
ad display ● Deliver 500 clicks on a 300x250px
● Delivery of the number of ad format between January 15 and
impressions, clicks, or other target February 14 to UK users, but only
between the hours of 3:00pm and
● Alternation of ads from multiple
10:00pm, with priority “high”.
clients
● Deliver unlimited impressions on
● Setting a maximum number of ad
a self-promo ad in 300x250px or
impressions per visitor
300x600px format without any other
● Targeting the ad according to some restrictions, with priority “low”.
criteria, i.e., not all visitors see it
across the board, but only selected
ones If you don’t sell your own adver-
tising and don’t have to deal with
The primary goal of the ad server is such things, then the ad server
to deliver campaigns according to set may not be of interest to you and
priorities. The priority can be a dead- is best avoided.
line, number of impressions, number
However, as you’ll see in the next
of clicks, price, etc. At the same time,
chapter, having your own sales
the ad server uses restrictions to target channel is very convenient and
smaller audiences (a certain group of gives you a huge advantage and
sites with similar content, a group of some independence from other ad
visitors within one GEO location, etc.). space sellers.
 109

And so on. It depends on how big

the publisher is and how demanding
their clients are. The number of active
campaigns an ad server has to handle
can range from single units to tens of
thousands. Its primary goal is not to
select the highest bid, but to deliver all
the advertising it has commissioned in a
defined order of priority

Ad server settings related

to header bidding
In the context of header bidding to
The price of the winning bid is passed
the ad server, we set up another set
to the ad server within the variables.
of campaigns in the ad server, which
For example, Google Ad Manager (GAM)
sets certain priorities to the incoming
refers to these as “key-values”. So, for
requests from header bidding. These
example, Prebid sends the variable “hb_
help the ad server decide when to select
pb=3.5” to GAM.
a bid from header bidding and when to
select other internal bids. Price plays an The ad server tracks these variables
important, but not the only, role in the and targets campaigns to them. Thus,
decision. there is a special campaign in the ad
server targeting exactly the variable
“hb_pb=3.5”, which is triggered exactly
when a response with this value and
Setting price levels variable comes from the header bidding.
Header bidding sends the price at which The price can be rounded. Thus, instead
it was bid to the ad server along with of the price of the winning bid, the price
the winning bid. We will stop here for a level (range) closest to the actual price
moment. is sent from the header bidding.
110

This trick is used for a simple reason. The rounding rate can be adjusted to
For each price (or price level) there your needs. The more campaigns your
must be a line item for the ad server sales team has in the ad server, the
that will display the ad for that price. more they will want to have detailed
To avoid having to create a special header bidding campaign settings.
campaign in the ad server for each Conversely, if you have few campaigns
increment of 0.01, price levels are often of your own, then you’ll probably settle
used. For example, header bidding will for very rough rounding.
round the winning price of 3.52 to 3.5.
And in the ad server we have campaigns However, some ad servers (such
set up with increments not of 0.01, but as Adform) also offer features
only 0.1 (i.e. 10 times less). If you used where they work directly with the
an increment of only 0.01, then you price bid in header bidding, which
would need 1000 campaigns for a price is not rounded in any way.
range of 0 to 10.
There are different ways to work with
rounding prices, we have shown a rela-
tively trivial case above. In practice, we Setting other
often use a setup where there is very
little or no rounding for lower prices and
parameters in the ad
rounding increases as the price increas- server
es57. This setup correlates with the fact Price is not the only thing you need to
that at lower price levels the ad server pass to the ad server from the header
has more options to choose from, but if bidding. You definitely need to pass
you present it with a very high bid, it is other information to it as well, such as
unlikely to have as much competition. information about the winning demand
partner.
The ad server therefore does not If there were to be a special campaign
receive information about the ex- in the ad server for each combina-
act amount of the winning bid, but tion of incoming variables, then there
the rounded price. This is import- would have to be thousands of such
ant to keep in mind when working campaigns. This is obviously difficult to
with statistics in the ad server.
create and the long-term management
is complicated to say the least.

57
https://docs.prebid.org/prebid-mobile/adops-price-granularity.html
 111

Setting up multiple
Therefore, there are ways to
simplify this setup and minimize header bidding
the number of campaigns in the
ad server58 or to create campaigns
solutions
in the ad server automatically. For As mentioned earlier, you can use mul-
more information, I refer to the tiple header bidding solutions. At the
Prebid website, for example59 very least, linking Amazon, Google and
one other independent wrapper makes
sense. In the context of ad server inte-
gration, for example, such an integra-
tion looks like this:

Setting up multiple header bidding solutions

in the ad server

AD SERVER Bidder 4
Bidder 1

Header Price Priority UAM Bidder 5

Bidder 2
Bidder
Price Priority Bidder 6
Bidder 3
ADX
Dynamic Allocation
Ad EBDA 1
Guaranteed Deals
EBDA 2

EBDA 3

Figure 7.1: Setting up multiple header bidding solutions in the ad server.

Source: headerbidding.co

58
https://docs.prebid.org/overview/how-to-simplify-line-item-setup.html#how-to-simplify-line-item-setup
59
https://docs.prebid.org/tools/line-item-manager.html
112

How to connect an ad server

to header bidding
There are several ways to link an ad Hybrid engagement
server with header bidding, so let’s
discuss three basic approaches. The ideal model is a hybrid that com-
bines the advantages of the above
solutions. That is, it balances the needs
of the ad server and clearing your
internal orders against the possibility
Ad server takes of a higher quote coming through the
priority header bidding.
This model gives the ad server absolute For this option to work best, a large
priority in selling ads. That is, only if part of your business models must be
the ad server does not choose a given CPM-based. Alternatively, the ads in
impression will it send it to header bid- the ad server should be converted to
ding to be bid on by demand partners. CPM in some way. Only then can head-
A great thing about this method is that er bidding in cooperation with the ad
you have absolute decision-making server assess and select the best bid.
control over the sale of advertising. On
the downside, you are missing out on
revenue from impressions that demand
partners would be able to sell at a The problem of set
higher CPM than what your ad server is
selling them for. priorities
In practice, it is often the case that the
ad server is prioritized at higher price
levels and header bidding is only ap-
Header bidding takes plied at lower prices. This is a consider-
able simplification of the whole issue.
priority
For example, let’s say your direct cam-
This model leaves the option to sell im- paigns have an average eCPM 3 times
pressions to demand partners involved higher than the average eCPM from
in header bidding first. If there is no header bidding. Thus, it seems logical
bid above the floor price, the impres- that you will automatically give direct
sion is sent to the ad server. campaigns priority in clearance (they
will be on higher priority).
 113

But realistically, in specific cases, it Hopefully the publisher has had enough
often happens that a bid from header foresight to set up their business
bidding comes with a higher CPM than model in order to minimize the number
what a direct campaign can currently of times they force the ad server to
offer in the ad server. If in such cases choose a bid other than the one with
the ad server does not decide on the the highest CPM.
current price but on the priority, then
the publisher loses potential revenue.

It is better to decide on a case-

by-case basis than to give direct
campaigns a higher priority just
because they have a higher eCPM
on average.

So ideally, the ad server should always

make decisions based on the current
CPM. However, this is often not pos-
sible in practice because other factors
come into play when, for various rea-
sons, direct campaigns need to be han-
dled as a priority regardless of price.
If it were to always make decisions
based on price alone, then the ad
server would lose its relevance. So try
to maximize opportunities to decide by
price, but at the same time take ad-
vantage of the other ad server features
that will sideline the auction based on a
price at an opportune moment.
114

Ad server as an auction
participant
In header bidding, the ad server does This setup has the advantage of letting
not usually participate directly in the direct campaigns compete in header
auction. It just compares the result that bidding on price only. But if we need
comes from the auction with the direct to, we have a first pair that allows us to
campaigns and decides whether to give check in direct campaigns on a priority
preference to the winning bid from the basis.
header bidding or to give preference The whole setup is important and
to another direct campaign that it has needs to be well thought out with re-
entered. spect to the types of orders you have
set up in the ad server.
But there is another option of-
fered by some ad servers (such as
Adform). The ad server can partic-
ipate in the auction directly as one
of the demand partners.

If the ad server can participate di-

rectly in the header bidding auction,
this allows for much wider possibilities
for setting up and managing internal
campaigns. For example, it is possible
to have two pairs for each ad position
in the ad server: ad tag + campaign.
One pair will be used as a classic ad
server and will handle direct campaigns
in priority. The other pair will only
participate in header bidding auctions
where internal campaigns will compete
on price.
 115

Ad server as an auction participant

Bidder 1 AD SERVER
Ad
Header Direct
Bidder 2
Bidder Header Bidder

Ad servever
Bidder

Figure 7.2: Setting the ad server as a participant in header bidding

116

Why it is good to have a direct

sales and ad server
You may have found this chapter com- eCPM on average, and conversely, more
plicated. You are right, setting up and programmatic demand with higher CPMs
operating an ad server is a separate pushes direct sales.
field. Large publishers used to employ But the same is true one level down;
separate staff to operate the ad server it would be a mistake to assume that
before header bidding came about. plugging Google AdX into Google ad
Over time, this field has transformed server will solve automated sales. Goog-
towards automated selling. And that le is still just Google, and if you don’t
puts demands on staff who manage this provide it with enough competition, your
setup, especially with regard to overall revenue won’t be as high as it could be.
ties into the sales strategy.
Setting up a campaign correctly in an
From a publisher’s perspective,
ad server is no longer just a matter
it is essential to have a properly
of one project. The right configuration set up ad server that will try to
can significantly increase or decrease a handle programmatic sales from
company’s overall revenue. many sources while also having a
In general, it is advisable to have as large amount of additional unique
many competing resources as possible demand. And this can be provided
by, for example, an in-house sales
in the ad server, ideally a combination
team selling advertising to direct
of direct sales with individual demand
clients. By appropriate ad server
partners (i.e., programmatic buying). setup, we mean that demand is in
However, with optimal use of demand balance and there is minimal arti-
partners, you will gradually hit a cap. ficial favoritism to certain sources.
Adding more demand partners will no
longer make sense. And here is the
huge power of your direct sales force.
It is the one that can create an addition-
al and unique competitive offering for
Automated demand partners.
But that’s exactly the fuel – the driving
force. If programmatic partners had no
competition, they would sell at a lower
 117

Chapter conclusion: Ad server

In this chapter we: This chapter was the most comprehen-
● explained what a good ad server is sive in this book and should provide a
for, theoretical background for the following
chapters
● how important it is to have a well
thought-out sales strategy,
● how to set up header bidding
campaigns in an ad server,
● why it’s a good idea not to rely on
programmatic alone and to have
direct sales as well.
118
 119

PART
3

Areas with
a strong relationship
to header bidding
120

Chapter 8:
COOKIES AND OTHER USER ID
SOLUTIONS
The online advertising industry needs services where the user can be unique-
a way to identify specific users. It then ly identified, for example, by their
associates information about the user’s login. Using the same login on multiple
behavior with that identification. Users devices or web browsers will always
are very often categorized into interest identify a single user. A typical example
groups based on the sites they visit, of such a service is Facebook, where
or carry information about visiting a users are usually logged in.
particular online store. In contrast, without a login, a single
The identification of a specific user is user on multiple devices is treated as
very challenging and in practice can different identities and pairing them
only be handled on a large scale by is not easy (though not impossible).

User identification in an online environment

mobil PC home PC work TV

Google
Firefox
Chrome

Figure 8.1: User identification in an online environment

https://blog.google/products/chrome/updated-timeline-privacy-sandbox-milestones/
60

https://blog.mozilla.org/en/products/firefox/todays-firefox-blocks-third-party-tracking-cookies-and-crypto-
61

mining-by-default/
 121

Therefore, we often identify a partic- For many years, the online advertis-
ular user not by a person, but by their ing industry was built on cookies, and
web browser or device. Nowadays, these unique IDs were stored in cook-
most users access the Internet from ies.
multiple devices, such as a computer
and a mobile phone. Some use multiple
web browsers within a single comput- Google Chrome60 will stop using
third-party cookies in 2023.
er. However, even identifying a user in
FireFox61 and Safari62 have already
only one device and browser is benefi-
blocked them
cial for the advertising industry.
For each user (whether we are talking
about a login or a web browser), each Given the strong dominance of Google
advertising technology assigns a Chrome (over 60% share63), the use of
unique identifier (hereafter referred cookies for advertising purposes will
to as ID), to which it then associates only make sense as long as Google
everything it knows about that user. Chrome uses them – i.e., until 2023.

Web browser Market Share

Chrome 62,9%

Safari 17,7%

Internet Explorer & Edge 4,9%

Firefox 4,8%

Opera 1,1%

View Monthly Trends

Figure 8.2: Web browser market share. Source: w3counter

62
https://webkit.org/blog/10218/full-third-party-cookie-blocking-and-more/
63
https://www.w3counter.com/globalstats.php
122

This change will have huge implications they realize its impact on their rev-
for the entire online advertising eco- enues. According to Adform, most
system and will affect everyone in the European publishers are ready for the
market. Therefore, for several years changes. For example, according to
there have been efforts to find an al- their research64, 93% of UK publish-
ternative solution for user identification ers have already applied some kind of
that would replace the core functions of identity system based on first-party
cookies while maintaining legality and cookies.
ethical passability.
In practice, it turns out that publishers
are very quick to apply the new user
identity solution to their websites as

What are Cookies?

can retrieve all the information it has
Cookies were first introduced previously stored about you.
in 1994 by Netscape
In practice, cookies are used to speed
Communications employee Lou
up your work. When you arrive at a
Montulli65, and activated by
Internet Explorer in 1995 page where you have created a profile,
the login form will immediately offer
your login and password, or the page is
Cookies are short text files created preloaded from the last time you left it.
by a website you visit and stored on If you’ve added something to your cart
your computer by your internet brows- in the store, the cart contents are still
er (Internet Explorer, Firefox, Google there on your next visit. Contact forms
Chrome, etc.). They serve to send the remember your address and offer it
stored cookie back to the same website to you so you don’t have to type it in
when you return, so that the browser again. The benefits go on.

64
https://site.adform.com/company/press-releases/adform-calls-on-cmos-to-ramp-up-adoption-of-first-par
ty-ids-after-publishers-prove-they-are-ready/
65
https://en.wikipedia.org/wiki/Lou_Montulli
 123

How cookies are Cookies themselves therefore have an

impact on page loading speed. Although
classified it’s usually not a substantial impact, it
There are three types of cookies: can become significant when a site uses
too many cookies.
● First-party cookies – used by web
developers to store data about
users, but can only be used by the
website itself. Among other things,
they are used to store access data Why third-party
and personalized settings and to cookies will disappear
remember what the user has added
to their cart or wish-list. Cookies are an amazing tool that have
served advertisers for decades. Un-
● Second-party cookies – essentially fortunately, time shows that they can
first-party cookies, but which are easily be misused or used for purposes
passed on to a third party by their that users are completely unaware of.
owner.
Users have increasingly called for re-
● Third-party cookies – used by strictions on tracking and data sharing.
external technologies. So, for As a direct result of this pressure:
example, on TopPublisher.com,
Google has embedded third-party ● Legal regulations such as GDPR and
cookies for advertising purposes. others have emerged (see Chapter 9,
page 134).
The advertising industry is based on
third-party cookies and each domain ● Web browsers have slowly
usually contains several of them. But stopped using third-party cookies
third-party cookies are also used for and replaced them with other
non-advertising purposes, for instance technologies.
by social networking buttons or chat
applications. One thing they have in
common – these cookies are third-par-
ty applications that are embedded in
websites.
124

Impact of cookies on without cookies, limiting the frequency

of specific users’ engagement will also
publishers’ revenues become impossible.
When cookies are disabled, there is Disabling them without an adequate
minimal transfer of important data substitution could therefore jeopardize
necessary for ad targeting. While it is publishers’ revenues.
possible to use other targeting meth-
ods, most online advertising tools
used third-party cookies at the time
of their existence. This is not just a
matter of targeting ads to interests;

Several measurements have been made on the topic of disabling cookies (pay
attention to the sources and check each study’s conditions):
An experiment conducted by Google66 in 2019 resulted in a drop in revenue of
more than 50%.
A study on Digiday.com67 shows a 40% drop in revenue for publishers after
restricting cookies within the Safari browser.
The Rubicon Project found out68 that disabling cookies on Safari led to a 60%
drop in earnings for publishers.
Research69 from 2019 noted revenue for users with AdChoices disabled was
52% less than for users with the feature enabled.
This study70 compares the revenue attributable to new cookies with longer-
existing cookies. The result shows a difference in yields in the upper tens of
percentages.
Another study (by Marotta, Abhishek, and Acquisti)71 showed a drop in
revenue of only 4%.

66
https://services.google.com/fh/files/misc/disabling_third-party_cookies_publisher_revenue.pdf
67
https://digiday.com/media/the-scale-of-the-problem-is-enormous-apple-flexes-strong-anti-tracking-stance/
68
https://www.theinformation.com/articles/apples-ad-targeting-crackdown-shakes-up-ad-market
69
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3020503
70
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2421405
71
https://weis2019.econinfosec.org/wp-content/uploads/sites/6/2019/05/WEIS_2019_paper_38.pdf
 125

This document72, on the other hand, attributes the substantial losses not to
third-party cookie losses, but rather to technology fees in the “supply chain”,
and estimates a much smaller decline for publishers.
There are other studies. As can be seen, predicting what will happen is not
easy and other influences come into play other than the actual removal of the
third-party cookie.
But we already have the first hard data from a related area. In April 2021,
Apple released a new version of the operating system, 14.5. In it, each app
must be separately enabled in order to collect data about them. Apps can’t do
this without explicit consent. According to an analysis by the Financial Times73,
this move has cost the biggest social networks such as Facebook, Snapchat,
Twitter and YouTube an estimated $10 billion in their first six months of
existence.

Cookie Matching (Cookie Syncing)

Each advertising technology stores
its own set of cookies in the TopPub-
its own cookies with a unique user
lisher.com domain and cannot read
identifier on each domain where it
other partners’ cookies. This is crucial
can advertise. However, these cook-
because the advertising ecosystem is
ies are only readable by their creator
made up of many entities (SSPs, DSPs,
(or the domain that stored them). For
DMPs, etc.) that pass information be-
example, if a demand partner stores a
tween each other.
cookie on TopPublisher.com under the
domain doubleclick.net, then this file In order to be able to target advertis-
is inaccessible to any other domain. ing to users and to use multiple tech-
Each demand partner therefore stores nologies in the purchase chain, Cookie

72
https://acfou.medium.com/abstract-2fefd374edb2
73
https://www.ft.com/content/4c19e387-ee1a-41d8-8dd2-bc6c302ee58e
126

Matching, or cookie syncing, must This is a very simplified description.

occur. In reality, it is important to note that
The whole process is based on each cookies are time-varying, in the sense
technology storing its own user ID for that some users disable them entirely
the domain in a cookie with a unique in their browser, or occasionally delete
URL. Each time an ad request is made, them, or use some blocking soft-
the DSP buying platform also retrieves ware. In these cases, matching has no
these URLs from other technologies. It chance of taking hold.
compares these URLs to a table where The problem with cookie matching is
user IDs with these URLs are stored also that there is some drop-off each
long term. time a match is made, because the
If technologies are able to “match” matching is never 100%. See the ex-
unique identifiers to their users in this ample in the figure. If the match rate
way, it means they are also able to in the pairing was 80%, then after the
match them with the user data they third step you only have a 40% match
collect. rate.

Data loss during retransmission

100M users 80M users 64M users 44,8M users

START Sync 1: Sync 2: Sync 3:

Advertiser’s With DMP 20% With DSP 20% With SSP 30%
target audience audience loss audience loss audience loss

Active cookies Lost cookies

Figure 8.3: Data loss during retransmission. Source: The Trade Desk
 127

Alternative solution to Cookies

– User ID
Cookies have two major problems: ● Several dozen companies
1. They will stop working on the latest representing both buy-side and
Google Chrome browser in 2023. sell-side technology providers
have formed the Advertising ID
2. Cookie matching decreases substan- Consortium75, whose mission is to
tially with the number of iterations. democratize the use of identity in
The advertising industry has been look- advertising technology.
ing for a better solution than cookies for There are already many solutions for
some time now, one that would allow user identification. If you’re looking for
users to be identified across devices a specific one, you can start with the
and systems and easily targeted. Prebid ID module overview page76. For
For example: example, PubCommonID, which is di-
rectly developed by the Prebid commu-
● Google announced the SandBox
nity, is an effective one. Neither solution
project with the end of cookies (see
has yet been dominantly adopted by the
Chapter 8, page 131).
market, and the situation in this area is
● AIB Tech Lab launched Project Rearc74 fragmented.
to provide a framework for cookie-
free ad targeting.

Adoption of User ID Modules by Prebid Adapters

Netid 5,8%
Britepool 6,9%
Parrable 8,1%
LiveIntent 9,2%
DigiTrust 9,8%
Criteo 9,2%
Trade Desk 16,2%
Pubcommon 12,7%
ID5 11,6%
Liveramp 10,4%

0 5 10 15 20

Figure 8.4: Adoption of User ID Modules by Prebid Adapters. Source: ID5

74
https://iabtechlab.com/project-rearc/
75
https://www.adidentity.org/
76
https://docs.prebid.org/dev-docs/modules/userId.html
128

For better explanation, we will explain

Solving the user identification three specific solutions.
issue is a hot topic and companies
are investing a lot of resources in
it. According to a study published
on eMarketer.com, U.S. marketers’
spending on identity solutions will SharedID
grow from $900 million in 2018 to (PubCommonID)
$2.6 billion in 202277.
SharedID is a third-party cookie sync
solution and a simple pop-out solution
that makes it easy for users to manage
their preferences. PubCommon ID has
Deterministic and existed since 2018 as a user identity
probabilistic matching solution for cookie-free environments
(Apple world).
Most alternative solutions to user iden-
tity are based on either deterministic or Prebid merged these two solutions into
probabilistic matching, or a combination one SharedID in 2020 and continues to
of the two. operate as open source. The goal of this
project is to be an open norm for the
● The deterministic approach consists
entire market and give publishers full
of using “strong signals”, such as
control over the solutions they use.
encrypted personal information like
an email address or phone number, SharedID combines the benefits of
to link a device to a given identity. storing IDs in third-party cookies with
Deterministic matching is inherently first-party cookies. It therefore takes
more accurate. advantage of the current state of the
art in header bidding and is ready for
● The probabilistic approach uses “soft
the world after 2023 when third-party
signals”, such as device type, IP
cookies end.
address, software version, operating
system, screen resolution, location,
and more, to create a graph of user
identities. Probabilistic matches will
never achieve 100% accuracy, but
can achieve a reasonable level of
confidence.

77
https://www.emarketer.com/content/will-the-advertising-id-consortium-solve-marketers-audience-identi-
ty-problems
 129

Unified ID 2.0 from Universal ID from ID5

The TradeDesk The French company ID5 offers an inde-
pendent identification solution under the
(UID 2.0) Universal ID label.
The UID 2.0 solution uses a user’s email
Universal ID works on the principle of
address to identify them. The solution
tracking users using soft signals (prob-
is therefore suitable for publishers who
abilistic model) combined with tracking
can motivate their users to log in to
strong signals. Identity is stored in
their site. The email address is encrypt-
first-party cookies.
ed and assigned a unique ID which is
stored in first-party cookies. This approach has an advantage over
UID 2.0 because a large proportion
The solution is open source – free for
of users are unwilling to provide their
anyone interested in using it. Even
email when visiting a website.
though it was created by The Trade
Desk, it was given to the Prebid com-
munity. As a result, the solution can
be considered independent and further
open source development is ensured.

Interoperability
Many cookie-free ID solutions are being Some vendors are building interoper-
developed independently. This means ability into their own systems, while
that a publisher using Universal ID will others are building solutions that could
not be able to pass identity to a DSP address this problem across the board.
that uses UID 2.0, and vice versa. Tapad’s Switchboard is an example of an
This problem did not exist with across-the-board solution. Switchboard
third-party cookies. The solution was can link UID 2.0 with Universal ID and
available to all and ensured compatibili- several other solutions.
ty between systems.
130

Effect of Cookie and user ID

on header bidding
The principle of header bidding is based Header bidding also partly decides the
on the fact that the highest bidder gets bid path and therefore the SPO issue
the impression. In order to be willing (see Chapter 10, page 148).
to bid higher, the advertiser must have
enough information about how inter-
esting the user to whom the impression The cookie matching principle
works very well for C2S header
will be displayed is to them.
bidding, but the success rate is
Each system knows certain information worse for S2S header bidding.
about the user and associates it with However, this is not the case
the user’s identifier stored in a cook- with alternative user ID solutions
ie or alternative solution. As we have and due to the discontinuation of
shown, the issue of matching this infor- cookies, this significant advantage
mation and passing it along the entire of C2S header bidding solutions
will disappear. Combined with
bidding path between the advertiser
the other advantages of the S2S
and the publisher is crucial.
solution (page 55), this brings up
Header bidding plays an important role the question of whether most pub-
here as the starting point. Within head- lishers will gradually switch to the
er bidding, the connection of various S2S solution.
alternative solutions for user identity
recognition can be addressed. From
the publisher’s point of view, it makes
sense to integrate several different
solutions, as they should be prepared
for all solutions that advertisers will
use. For example, if there are five main
user ID solutions that advertisers use,
then each advertiser can only use one
of them. But the publisher needs to
be prepared for most of the options to
maximize their revenue, so they need
to work with all five.
 131

Google Privacy Sandbox

With the announcement of the end of
third-party cookies in Chrome, Google
has announced a new technology that
it plans to replace cookies with called
The Privacy Sandbox78. At the time of
this book’s publication, this technology
is still in the testing phase. Pay atten-
tion to news for updates. The Sand-
box assumes that Chrome (and other
browsers may join) will have several
APIs available.

78
https://privacysandbox.com/
132

Relationship of user identity

to header bidding
User ID solutions need to be included can use several79. So, a publisher
in your header bidding and connected should study the possible solutions
to the ad server. In addition, you must and choose which one to support.
also connect to the CMP tool (see page ● If you have some first-party data,
142) to determine which users will give then you should also attach it to
you the rights to handle their data. the user ID and include it in your
auction80. CMPs offered by partners
So this is an example of an extremely will certainly help this.
complex issue that a publisher needs
● Before each auction, the system
to connect to their entire ad ecosys-
should see what consent it has and
tem. While you’ll need to dive into the
adjust the outbound offer to the
developer documentation for a detailed demand partners accordingly and
understanding, we’ll describe a few pass the available information to
basic points here for a general under- them. For example, if there is an
standing. option to forward user ID, then it will
● The publisher can only handle the do so.
data if it has the user’s consent. ● It is also advisable to pass the user
The type of consent is dependent on ID to the ad server, as campaigns
local legislation and is discussed in are also targeted there based on
more detail in the next chapter (see identity, or this information is shared
Chapter 9, page 134). In general, for further to programmatic deals.
example, in the European Union, all In the case of Google, then it still
users must give you explicit consent. comes down to sharing the user ID
This is usually done using CMP tools to Google Open Bidding. See for
(see Chapter 9, page 142) that store example the description of the Prebid
who has given what type of consent. x GAM link81.
● Both the CMP tool and the user
ID solution are usually part of
the header bidding. For example,
Prebid has modules for this that can
be integrated into your solution.
The important thing is that the
integration is done correctly and
legally.
● As part of header bidding, a publish-
er usually has one consent collection
tool (CMP), but a user ID solution

79
https://docs.prebid.org/dev-docs/modules/userId.html
80
https://docs.prebid.org/features/firstPartyData.html
81
https://github.com/prebid/Prebid.js/blob/master/modules/userIdTargeting.md
 133

Chapter conclusion: Cookies and other

solutions for user ID
In this chapter we: ● discussed the direct impact of
● explained how user identity identity on the amount of eCPM in
is handled in the advertising auctions
environment, This chapter explains why user identity
● defined what cookies are and is important to the advertising industry.
that they will be replaced by new This has long been addressed by cookie
technology, technology, which is now gradually
being replaced by other technologies
● explained the matching of identities under the label user ID.
from different solutions,
134

Chapter 9:
PROTECTING USER RIGHTS

In recent years, there has been a grow- ● Personal Information Protection and
ing mistrust among users around the Electronic Document Act (PIPEDA) –
effectiveness of personal data protec- Canada
tion on the Internet82. At the same time, ● Lei Geral de Proteção de Dados
the willingness of users to share their (LGPD) – Brazil
data varies considerably from country
to country83. ● Personal Data Protection Act (PDPA)
– Thailand

For example, most of us remem-

ber the scandal surrounding
the misuse of Facebook data by
Cambridge Analytica84 and its im-
pact on elections.

This has been gradually followed up

by legal action in various parts of the
world, with perhaps the most publicized
being the European legislation known as
GDPR85.
The whole field of legal privacy pro-
tection is very broad and undergoing
dynamic development. For example, the
following regulations were in force in
2021:
● General Data Protection Regulation
(GDPR) – European Union
● California Consumer Privacy Act for
publishers (CCPA) – California

82
https://www.mckinsey.com/business-functions/risk-and-resilience/our-insights/the-consumer-data-opportunity-and-the-privacy-imperative
83
https://www.statista.com/chart/21081/countries-most-comfortable-to-share-personal-data-online/
84
https://www.marketingcharts.com/customer-centric/privacy-and-security-83114
 135

But there are many more. Global regu-

lations in 2021 have also been graphed
by the IAB86 based on data from
Dlapiperdataprotection.com:

Global data protection and privacy regulation

Canada: Digital Privacy Act reforming

PIPEDA (Personal Information and
Protection and Electronic
Documents Act))

China:PIS Standard
(Personal Information
EU:ePrivacy Security Specification)
Regulation
California:CCPA and GDPR
(California Consumer India:PDPB (Personal
Privacy Act) Data Protection Bill)

Brazil:LGPD New Zealand:

Thailand:PDPA
(General Data Privacy Bill 34-2
(Personal Data
Protection Law)
Protection Act)
South Africa:POPIA
(Protection of Personal Australia:Privacy Act
Information Act) and Amendements

Heavy Robust Moderate Limited

Figure 9.1: Global data protection and privacy regulation. Source:

Dlapiperdataprotection.com

85
https://ec.europa.eu/info/law/law-topic/data-protection_en
86
https://iabtechlab.com/wp-content/uploads/2020/08/IAB_TechLab_Privacy_IABEurope_EN.pdf
136

The topic of protecting users’ rights on

the Internet could be the subject of its
own book. In this publication, we will
only provide a general introduction to
the GDPR and points that directly affect
header bidding, in particular how to
obtain users’ consent to use their data
for ad targeting and how to pass it on
to the demand partners.

General Data Protection

Regulation (GDPR)
The GDPR87 is a work of the European
Parliament that came into force on 25 It is also possible to see an over-
May 2018 and establishes the protection view of the fines imposed and who
of EU citizens’ data. has been fined online using the
Enforcement Tracker88.
It presents a lot of changes for publish-
ers, but it primarily affects entities that
handle data on EU citizens. For exam-
ple, the GDPR has a big impact on finan-
cial and health institutions and local and Content of the GDPR
state authorities and employers. The GDPR defines several areas of data
The GDPR caused a lot of controversy at that it protects. These include:
the time of its introduction as it intro- ● Personal information (name,
duced new major obligations for many address, date of birth, social security
entities dealing with personal data. But number...)
it has also attracted a lot of attention
● Web-based data (IP address, cookies,
due to its hefty fines for breaches – up
location...)
to €20m or 4% of annual revenue.
● Health data, including genetic data

87
https://gdpr-info.eu/
88
https://www.enforcementtracker.com/
 137

● Data on membership of an ethnic or What preceded

other group
the GDPR
● Biometric data and sexual orientation
The GDPR didn’t come out of the blue.
● Political opinions and attitudes It was preceded by several iterations of
similar legislation by the European Un-
And it also defines the rights that data ion, and certainly the GDPR is not final
owners have: or definitive. There will most likely be
other regulations regarding the handling
● People can request the deletion of personal data to follow.
of their personal data from the
company’s database or withhold
consent to its handling. Already in 1995 the Data
Protection Directive89 was is-
● Companies must inform their clients sued, followed by the ePrivacy
and users about what data they Directive90 in 2002 and the ePriva-
collect about them and how they cy Regulation91 in 2009.
handle it before they do so. Personal
data must not be used without the
consent of its owner.
● Users and clients can ask companies
to modify their data and how it is Opt-in versus Opt-out
used. The GDPR requires the use of the op-in
Simply put, the GDPR mandates under method to collect user consent for per-
what circumstances data can be col- sonal data processing.
lected and how it can be further han- The opt-in method means that the data
dled. They give the owners of personal cannot be used without explicit consent.
data the right to prohibit anyone from This is a significant difference to, for ex-
collecting and working with it, and once ample, the California CCPA, where only
they have it, owners can request that it opt-out consent is required. GDPR is the
be deleted or altered. only legislation that requires opt-in.

89
https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex%3A31995L0046
90
https://eur-lex.europa.eu/legal-content/EN/ALL/?uri=CELEX%3A32002L0058
91
https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex%3A32009L0136
138

The opt-in method has one unpleasant

feature. If you ask a user to consent to
the use of their data they have not two
but three choices:
1. Give consent
2. Not to give consent
3. Ignore your request
The third point is a bigger problem
than it may seem at first glance. Under
the GDPR, point 3 is equal to point 2.
Whereas in other jurisdictions around
the world, point 3 is equal to point 1,
meaning publishers can continue to
work with the data and use it commer-
cially.
So within the EU, it is not enough to
have a CMP tool to collect consent (see
chapter Consent Management Plat- Figure 9.2: CMP tool example. Source:
OneTrust
form below)-publishers must actively
try to have the CMP tuned to get as
many consents as possible (maximizing
point 1)
The example below shows the CMP tool
(page 142 for a more detailed expla-
nation of CMP) configured for a user
entering the website from EU territory.
Consents to the use of cookies are not
automatically given and must be active-
ly allowed (opt-in method). Until this is
done, advertising cannot be targeted.
 139

If we were in California, consent to use Sites with international traffic, there-

cookies would be by default enabled and fore, must work with CMP tools that can
the user could choose to disable them recognize the user’s location and ac-
(opt-out method). Ads could be targeted cordingly pass information to the header
on the first visit. bidding that consent has already been
Understanding the difference between granted:
opt-in and opt-out is also very impor- ● in EU territory after active user
tant from a header bidding perspective. consent,
Before GDPR, ads were automatically ● in places like California, where
targeted to users. Prior to that, opt-outs consent is predefined and valid until
were required for consent, which few us- the user revokes it.
ers took advantage of, but the vast ma-
jority did not address it in any way. Now,
however, you can’t target ads in the EU
unless the user has previously given you
consent. This has led, for example, to
Prebid, as of version 4, requiring consent
and not running the auction without it.

Relationship of GDPR to
header bidding
From a header bidding perspective, user This data is essentially one of the cor-
data is data that identifies the user. nerstones of ad targeting and without it,
That is, the IP address, location, cookie, untargeted ads would be served. Price
or any data that the user has provided per ad is also influenced by data – the
as part of registration or otherwise. more data a publisher has, the higher
This data is used in programmatic buy- price they can earn for ad space (see
ing to target advertising and is passed Chapter 8, page 120 for full discussion
between the different entities partici- of user identification and transfer of
pating in header bidding (SSP, DSP, Ad identifiers between solutions).
Exchange, DMP, etc.).
140

It is imperative that publishers work

with GDPR and have consent from web- A DHL study92 illustrates well the
site visitors to use their data and to pro- variability of consents granted.
vide it to other entities in the advertis- DHL experimented with different
ing market. Or rather, publishers need CMP bar designs, showing that
consents granted may vary from
to work with GDPR (they are required to
28% to 70%, depending on design
do so by law) and should aim to maxi-
used.
mize the number of consents from their
users.

So if half of your users don’t give

you consent, then you’ll have a
noticeably lower eCPM on half
of your impressions. Engage UX
experts and try to maximize the
percentage of consents from your
users.

Transparence and Consent

Framework (TCF)
GDPR defines the legalities, but the
reality of commissioning is always more In other words, the TCF is a tool
challenging. That’s why the IAB Tech that publishers can use to be
Lab has released the Transparency and GDPR compliant. This framework
is an open source and fully avail-
Consent Framework (TCF). This has
able on Github93.
provided a standard mechanism for
requesting, storing, and sharing user
consent in the ad tech supply chain that
is GDPR compliant.
 141

As a result, the Global Vendor List94, an

evolving list of all entities that want to
be part of this standard, was launched.
Publishers then use the CMP (see page
142) to choose which vendors to share
their visitors’ consents with.
TCF also defines exactly how consents
are shared between entities and defines
the exact purposes of data collection.
The IAB Tech Lab maintains an inven-
tory of CMP95 providers compatible with
the IAB TCF framework. TCF v2.0 is also
supported by Google96.

92
https://analytics.dhl.com/despite-gdpr-up-to-70-analytics-opt-in-rates-why-extensive-testing-is-worth-er-
very-minute-of-effort/
93
https://github.com/InteractiveAdvertisingBureau/GDPR-Transparency-and-Consent-Framework
94
https://iabeurope.eu/vendor-list-tcf-v2-0/
95
https://iabeurope.eu/cmp-list/
96
https://support.google.com/admanager/answer/9805023?hl=en
142

Consent Management Platform

(CMP)
CMP tools have been developed so that In practice, this is usually a special web
publishers do not have to implement page add-on that, when it registers a
TCF or any other tool themselves. new user it does not know yet, shows
Consent management platforms (CMPs) a banner asking for consent to han-
are tools that allow businesses to define dle user data, including the option to
cookie policies, set geolocation rules, change predefined settings.
record user consents, and keep audit
trails to comply with specific privacy Exactly how the banner should
regulations such as GDPR, CCPA, and behave is defined in the IAB
others. Europe Transparency & Consent
Specifically for publishers, the CMP tool Framework Policies document97.
allows them to:
● inform users about what data is This is done proactively, i.e., the user is
collected about them and how it is prompted to do so. The main reason for
handled (for example, what other this aggressive technique is that some
entities it is provided to), legislation, led by the European GDPR,
● store this data according to the prohibits working with user data unless
users’ wishes – each user has several the user has given explicit consent. If a
choices about what data can be publisher has consent to use this data,
collected and how it can be further it has a very strong revenue impact for
handled, them. This consent is necessary so that
the auctions can be organized in header
● share user data with other entities
bidding (as we showed in the chapter on
throughout the supply chain,
cookies and user identity, page 120).
according to each user’s specific
preferences, Without consent, no personalized ad-
vertising can be served to the user and
● enable users to change consent
only non-personalized targeting (e.g.
settings for their data, including a
contextual) can be used.
complete ban on its handling.

97
https://iabeurope.eu/iab-europe-transparency-consent-framework-policies/
 143

CMP can be implemented by the pub- This consent must then be correctly in-
lisher themselves or with the help of tegrated into the header bidding before
one of the open source tools, but due to the auction can be executed. You may
the complicated legal and technical en- have noticed that many publishers in
vironment, there are specialized compa- the EU territory will not show you any
nies offering proprietary solutions that ads before you first consent to the use
work in this field. of your data using the CMP tool. At the
same time, when configured correctly,
the same URL will show the ad if you
From a publisher’s perspective, we are coming to this site for the first time
need to remember that CMP tools
from a non-EU location.
are general tools for managing
user consents. However, TCF is a
special framework for publishers. Let’s show this with the example
Most CMP tools can use TCF, but of Prebid. In order for the consent
you need to activate this option. transfer to take place in header
bidding, the Consent Management
Module99 must be integrated in
As we have already mentioned, CMP
Prebid. The way the consent
tools do not only address GDPR but also
transfer works is that the Prebid
the legislation of other countries, which library first reads the consent
is why their use is spreading rapidly at the beginning of the auction
outside the European Union. and then sends it to all the de-
mand partners. That is, it directly
attaches a “consent string” to the
For example, in the US, over ¼ of request for an advertisement and
all sites had some kind of CMP tool starts the auction. If the Prebid
activated by 202198. library does not have consent, the
auction will not be triggered at all.

CMP can usually also distinguish the

geolocation of the user and ask the user
for the correct type of consents in their
language. Among other things, it can
set up opt-in consent requests within
the EU and opt-out consent requests
outside the EU.

98
https://www.kevel.co/cmp/
99
https://docs.prebid.org/dev-docs/modules/consentManagement.html
144

Checking the settings

Checking whether a publisher has the
correct GDPR treatment is not a simple You can run this feature in Google
matter, and although there are several Chrome console:
online tools, they will only ever give you __tcfapi(‘getTCData’, 2, console.log)
an indicative overview. This is an area
about which you need to consult and get specific information.
a lawyer.
However, from a header bidding per- You can examine how a website han-
spective, it is crucial that your users’ dles granted consents with the Google
consents are passed on the demand Chrome plugins “CMP Check”101 and
partners. So, for example, it is para- “Azerion Ad Expert”102, which check the
mount that they are implementing TCF site settings for TCF functionality.
correctly.

Improperly passing on consent

can lead to a reduction in ad sup-
ply, and therefore a reduction in
revenue. This is why this area is
key.

CMP tools have a built-in API for com-

municating with other tools100. Part of
this API is the __tcfapi() function, which
provides information about the consent
granted.

100
https://github.com/InteractiveAdvertisingBureau/GDPR-Transparency-and-Consent-Framework/blob/mas-
ter/TCFv2/IAB%20Tech%20Lab%20-%20CMP%20API%20v2.md
 145

Figure 9.3: Information provided in the web browser. Source: Azerion Ad Expert

If you are using the Prebid solution for header bidding, then you can add the
“?pbjs_debug=true&debug=1” parameter to any URL and Prebid will start
outputting a detailed report to the developer console, including to whom it is
passing the “gdprConsent” parameter and in what form.

Figure 9.4: The bid request section in the Prebid environment

101
https://chrome.google.com/webstore/detail/cmp-check/odemilkphklhfejgajeaohjcdfmmfjde?hl=en
102
https://chrome.google.com/webstore/detail/azerion-ad-expert/nndadbimjipilgfojofhpjjkhgflkihc?hl=en
146

However, I recommend checking the

consent transmission directly with the
demand partners. Most of the checking
mechanisms are available online, and
with S2S integration, checking cannot
be done this easily. It is therefore safer,
although slower, to verify the correct
transfer of consents to demand part-
ners by querying them directly. Some
demand partners already present this
data in their online dashboards.

CMP and the world beyond

header bidding
Publishers need to realize that CMP is a CMP is a tool with relevance beyond
universal tool for collecting and trans- header bidding and you need to consid-
mitting user consent. We desperately er the additional features you need from
need it in the realm of ad monetization, it when choosing one. Learn to use it
and thus header bidding as we’ve cov- well yourself or hire someone who can.
ered it so far.
But we need to work with these con-
Setting up a CMP tool correctly
sents on other levels as well. For exam- can not only make you money
ple, if the user has not given consent from ad monetization, but also
to use analytics tools we must not run save you fines for illegal use of
Google Analytics or other cookie-based user data.
analytics tools. Similarly, we need to
be careful with marketing tools using
retargeting and conversion tools such as
Google Ads, Criteo, Facebook, etc.
Simply setting up CMP will not solve
these other problems on its own. So
 147

Chapter conclusion: Protecting user rights

In this chapter we: This chapter explains the legal stand-
● discussed the different legal ards that publishers must follow. It also
provisions relating to the rights of shows the way to best meet all obliga-
Internet users in different parts of tions and minimize the potential nega-
the world, tive impact on revenue from targeted
advertising sales.
● described in detail the most stringent
GDPR legislation in the European Very simplistically, one could say: CMPs
Union, are specific tools that use the TCF
framework in order to comply with legal
● explained the difference between opt- regulations (GDPR and others).
in and opt-out,
● listed the tools for collecting user
consent and how to share these
consents with demand partners,
● explained how to check that your
setup is technically correct.
148

Chapter 10:
SUPPLY PATH OPTIMIZATION AND
DEMAND PATH OPTIMIZATION
Supply Path Optimization (SPO) is a a few months. We will discuss all the
very important part of programmatic tools in this area in more detail at the
buying, and header bidding has a major end of this chapter (page 155).
impact on it, so we shouldn’t overlook it.
The move by publishers from waterfall SPO is important for both
to header bidding has had a profound advertisers, helping them to
impact on purchasing paths. Publishers optimize their buying paths,
have integrated several demand part- and for publishers, who should
ners and each of them has dozens, or wish to promote transparency
more likely hundreds, of links to other and optimization of their bidding
demand partners. An advertiser adver- paths towards advertisers, as this
will have a long-term effect on
tising in a single DSP can then ultimate-
revenue. On the contrary, laxity
ly compete with itself several times
on the part of publishers may lead
over due to these links, as we will show to a gradual reduction of offers by
below. advertisers.
Moreover, each system in the demand
path has different fees and works with
different types of auctions. This com- Gradually, the term Demand Path Opti-
plexity has made the entire bidding path mization has started to emerge, which
incredibly non-transparent, which has is more relevant to publishers and is
resulted in a substantial increase in ad discussed at the end of this chapter
fraud. (page 163).

These fundamental inefficiencies have

led to efforts in the market to clean The term SPO probably first
up the whole system. SPO probably appeared in 2016 in an article103
sums them up most thoroughly, but the by Brian O’Kelley (AppNexus). In
matter is so fundamental that it has 2020, IAB Europe published “The
made its way into the new RTB protocol IAB Europe Guide to Supply Path
Optimsation”104 , which summa-
version 3.0. The importance of SPO be-
rizes SPO and offers concrete
came apparent in the first cleanup effort
practices.
with the ads.txt file, which the market
jumped to integrate within

103
https://bokonads.com/preventing-bad-behavior-in-header-bidding-supply-path-optimization/
104
https://iabeurope.eu/wp-content/uploads/2020/09/IAB-Europe-Guide-to-SPO-Sept-2020-2.pdf
 149

The principle of SPO

Imagine you want to buy a product, say Header bidding has made the whole
a perfume. You type your favorite brand process even more complicated. Where-
into a search engine or price compar- as in waterfall the number of buying
ison tool and it will show you several paths was limited, in header bidding the
retailers offering that product, but some processes discussed above are expo-
of them will probably differ in price. nentially amplified thanks to the inter-
Anything can make a difference in price connectedness of all the technologies
– the manufacturer’s purchase price, involved in the process:
shipping, the retailer’s margins, the
number of resellers in the shopping trip,
etc. It’s the same with programmatic
advertising. As an advertiser, you have
the chance to buy the same impression
in several different ways and at different
prices.

Partners' participation in demand by type of implementation

WATERFALL SETUP HEADER BIDDING SETUP

SSP 1 DSP ADVERTISER SSP 1 DSP ADVERTISER

If not
filled

SSP 2 SSP 2

If not
INVENTORY filled
INVENTORY

SSP 3 SSP 3

Figure 10.1: Partners’ participation in demand by type of implementation

150

In fact, most large SSPs have direct A simplified diagram of a single bid’s
links to most DSPs. At the same time, request passage might look like this, for
ad exchanges come into play, where example:
demand and supply from multiple sys-
tems accumulate. Ultimately, we then
find out that a bid request has flowed
to the publisher from, say, ten different
systems. Each of these ten bid requests
has a completely different path behind
it and is affected by the systems it has
passed through along the way. So in the
end, the advertiser is competing with
itself.

Example of a simple bid request pass

Advertiser DSP

Ad
SSP 1 SSP 2 exchange
A

Publisher
Wrapper Publisher
ad server

Figure 10.2: Example of a simple bid request pass

 151

The bid request in this case can come

in three ways. However, this is a rather
simple example; a more complex one
might look like this:

Example of a complex bid request pass

Ad exchange 1

Advertiser DSP 1 Ad exchange 2

Wrapper 1
SSP 1 Ad exchange 3

Ad exchange 1

SSP 2
Publisher
Publisher
ad server

DSP 2 SSP 3 Ad exchange 3 Wrapper 2

Ad exchange 1

Figure 10.3: Example of a complex bid request pass

As you can see, shopping paths have and quality space, and secondly, they
become much more complicated and want to buy it ideally from the shortest
need attention. The complexity of the possible path to save on intermediary
system brings with it a higher level of fees.
ad fraud. Optimization by DSPs and advertisers
Advertisers and DSP platforms are pay- is not only price-oriented, but also
ing a lot of attention to the SPO issue performance-oriented. Whether it is a
because optimization is in their direct person or an algorithm monitoring the
interest. Firstly, they want to buy real performance of specific spaces, SSPs,
152

and ad exchanges, those that bring ● Improve performance for the

the lowest performance (i.e., visibility, advertiser – whether by eliminating
clicks, leads, cost savings) will gradu- unnecessary fees or finding the most
ally fall out of usage. And, importantly, efficient path.
systems today can also perform optimi- ● Reduce the number of instances
zations in an automated way. where the advertiser competes with
itself.
It is therefore in the interest
of publishers to be as open as
possible to advertisers and their
buying platforms and deliver
maximum performance. Anything
else may lead to a temporary
increase in revenue, but in the
long run it is not beneficial for
publishers.

So it is clear that the SPO issue is

helping:
● Eliminate unnecessary fees – simply
by seeking the most direct or
efficient path between advertiser
and publisher. But it can also justify
the importance of some systems and
therefore their fees. Knowing the
buying path and understanding the
importance of all the players along
the way can also give important
know-how for negotiating fees.
● Bring transparency and reduce fraud
– by using projects such as ads.txt
(ads-app.txt), sellers.json and supply
chain, transparency is achieved
for the entire purchase path for all
involved (see page 155).
 153

The SPO procedures

The programmatic ecosystem was not according to the results. Some
simple even before the massive expan- optimizations can be done in real-
sion of header bidding, and now it is time, but many are done with a delay.
almost literally opaque. What can we Therefore, not all changes made
do about it? Most of the transparency by advertisers or publishers will be
efforts are led by advertisers and could immediately apparent, which needs
be divided into two areas: to be kept in mind throughout the
● Manual buying path optimization – optimization.
This puts demands on staff in the
industry to try to map the routes
through which they buy advertising.
They uncover all participants’ added
value as well as fees and then make
optimizations.
● Automated processes – Buying
platforms have buying path data
from projects like ads.txt, sellers.json
and supply chain available. This data
can be analyzed and changes can
be made to purchasing strategies
154

Suppy chain
As we have already mentioned, the
programmatic ecosystem is very robust In 2020, ISBA conducted a study
and each technology requires a certain on this topic in the UK market105.
fee for its use, so the difference be- The study included 15 advertis-
tween the price paid by the advertiser ers, 12 agencies, five DSPs, six
SSPs, and 12 publishers, repre-
and the price that reaches the publisher
senting approximately £0.1bn of
is significant.
UK programmatic ad spend and
nearly two-thirds of AOP (premium
publisher) digital ad revenues. You
can see the result in the chart

Supply chain findings

%
100
7%
8%

10%
75
100% 15%

8% 1%
50

51%
25

0
Advertiser Agency DSP Technology Uknown SSP Technology Publisher
spend fee fee fee delta* fee fee revenue
(demand side) (suply side)

Figure 10.4: Supply chain findings. Source: ISBA

105
https://www.isba.org.uk/system/files?file=media/documents/2020-12/executive-summary-programmat-
ic-supply-chain-transparency-study.pdf
106
https://www.emarketer.com/content/us-programmatic-fees-2019
 155

Over the years, you can see that the

According to another study, “US chain fees have been gradually de-
Programmatic Fees 2019”106, a creasing. This is due to increased com-
full third of the total spend within petition, a larger market, and gradually,
the US market in 2019 went to greater transparency throughout the
the area between publisher and
industry.
advertiser. We get similar data
from other sources. For example, But there is one thing these findings
Magna Global’s private network107 can and should teach both advertiser
from 2017 shows system-wide and publisher: to be selective about
fees at 55%, or a chart from where and when these chains bring val-
Pubmatic108 from 2010, which is ue based on the knowledge they gain
even less hopeful for the publish- from the systems.
er, shows that only $1.8 of a $5 ad
went to the publisher.

Transparency projects
In order to make the system more to 50%110 in the first year. Today, ads.
transparent, the IAB consortium defined txt is already strictly required by some
the ads.txt standard in 2017 and added demand partners and without it, they
sellers.json and supply chain in 2019. won’t accept demands at all. For larger
publishers, this is practically a necessi-
ty. If you are a smaller publisher or your
site is monetized by one specific sys-
Ads.txt tem, you may not need ads.txt.

After the IAB introduced the ads.txt Ads.txt is a simple way for publishers to
project in 2017109, it quickly gained declare who is authorized to sell their
popularity and the number of publish- inventory, increasing transparency for
ers that started using it skyrocketed advertisers.

107
https://www.isba.org.uk/system/files?file=media/documents/2020-12/executive-summary-programmat-
ic-supply-chain-transparency-study.pdf
108
https://pubmatic.com/blog/ad-tech-tax-or-a-reality-of-doing-business/
109
https://iabtechlab.com/press-releases/iab-tech-lab-launches-assault-on-illicit-advertising-inventory/
110
https://www.emarketer.com/content/ads-txt-adoption-continues-its-steady-growth
156

Like a robots.txt file for example, ads.

txt can only be published on a domain Example: Let’s say there is
by the publisher’s webmaster, making it the domain, publisher.com. An
authentic. Moreover, it is a text file that advertiser receives an offer
is easily editable and readable by all through the XYZ system for an
impression on this domain and
entities.
is interested in buying it. He can
The ads.txt file can be periodically easily verify that the XYZ system
scanned by advertisers, i.e. DSP sys- is listed in the publisher.com/ads.
tems, to check the eligibility of sellers to txt file. If it is not, it is probably
offer these domains. an ad fraud and the advertiser can
discard the offer.
The following picture may help to ex-
plain:

Explaining the function of the ads.txt file

Content
Ads.
owner 1
txt

Ads. Content Supply

owner 2 DSP Advertiser
txt chain
:

Content
Ads. owner n Crawled
txt ads.txt
data

Figure 10.5: Explaining the function of the ads.txt file. Source: IAB

The ads.txt file has a clearly prescribed structure:

< SSP/Exchange Domain >, < SellerAccountID >, < PaymentsType >, < TAGID >
 157

For example111: Within the world of mobile apps, there is

greenadexchange.com, 12345, DIRECT, an equivalent to ads.txt in the app-ads.
AEC242 txt file, but it has minor differences that
you need to pay attention to.
blueadexchange.com, 4536, DIRECT
For many publishers, the ads.txt file
silverssp.com, 9675, RESELLER has become a repository of text records
that demand partners dictate to them
and publishers don’t think about. The
Explanatory notes for each item:
goal of ads.txt, along with sellers.json
● SSP/Exchange Domain – the official and supply chain, is to make the market
name of the demand partner that is transparent and optimize buying pro-
authorized to sell advertising on the cesses, so publishers should pay atten-
domain. This name is supplied by the tion to the ads.txt file and keep it up to
partner itself. date. Good policies include:
● SellerAccountID – a unique identifier ● Checking what records have already
that the seller assigns to the been added. Often demand partners
publisher’s account. use the same ad exchange and their
● PaymentsType – describes the type lines are then repeated.
of relationship between the publisher ● Removing entity lines from your file
and the seller. It can only take the when the cooperation has terminated.
following options:
● Inserting text notes in your file to
DIRECT – indicates that the publisher help you keep track of the records.
directly controls its account with the For example, it is useful to note
partner. who asked for particular lines to be
RESELLER – indicates that the re- inserted.
seller is authorized to sell the do- ● Being the only one who marks your
main. direct sellers as DIRECT, marking
TAGID – an optional parameter indi- other resellers with the RESELLER
cating the partner ID in the Trustwor- flag.
thy Accountability Group (TAG)112. ● Regularly checking your ads.txt file
with one of the checking tools.

111
https://iabtechlab.com/ads-txt-about
112
https://www.tagtoday.net/
158

As an example of a maintained file, platforms available to it. Other exam-

here is ads.txt from the NY Times ples: theguardian.com: 55 entries, bbc.
publisher. It contains only 21 records co.uk: 32 entries, bild.de: 30 entries...
(as of the time of publication of this And so we could go on. Clearly the big
book – 2022) and all of them are with publishers are careful about who they
the PaymentsType parameter DI- do business with.
RECT113. Thus, the publisher does not
allow any reselling of its ad space and
anyone who wants to buy it must be
in a direct relationship with the pub-
lisher. However, it has several different

Figure 10.6: Content of the ads.txt file on nytimes.com in February 2022

113
https://www.nytimes.com/ads.txt
 159

Sellers.json and supply The sellers.json project, on the other

hand, is published on the domain of the
chain seller (the demand partner) and de-
The sellers.json project is directly re- clares its relationship to the publisher.
lated to ads.txt and was introduced by The actual path between the advertiser
the IAB consortium in 2019114. and the publisher (i.e., each entity on
The ads.txt project helps publishers to it) is stored in a supply chain object
declare which demand partner is that is part of the RTB protocol.
authorized to offer them and is pub- Technically, this can be shown as fol-
lished on the publisher’s domain. lows:

Explanation of supply chain functionality

ADS.TXT

3
SELLERS.JSON SELLERS.JSON

‹› ‹› ‹›

2
DSP/
SSP Exchange
Buyers
4
1 Publisher sends a bid request.
2 Buyer receives bid request
and data from the SupplyChain object.
3 Buyer looks up the identities of
1 all intermediaries who resell inventory.
4 Buyer crawls and verifies vendors
authorized to sell inventory

Figure 10.7: Explanation of supply chain functionality. Source: Google

114
https://iabtechlab.com/press-releases/iab-tech-lab-announces-two-new-technologies-to-build-more-
transparency-trust-in-the-programmatic-supply-chain/
160

This gives advertisers an idea of how Below is an example of how such op-
far their offer has to travel before it timization can take place. The figure
reaches the publisher and how many shows two examples, before and after
participants are involved in the path. optimization:
With this, DSP tools can optimize their
buying paths and eliminate certain
players along the way.

Difference in the bid path before and after optimization

$1 bid Google Adx 80 cents

With
SELLERS.JSON

ADVERTISER PUBLISHER

$1 bid Adsense Google Adx Reseller 50 cents

Without
SELLERS.JSON

ADVERTISER PUBLISHER

Figure 10.8: Difference in the bid path before and after optimization. Source: Google

Setting up the sellers.json record must ● Name: the name of the seller
be done by the publisher in the demand ● Domain: the domain of the seller
partner system that manages their ads.
In most cases, this is part of the system ● Seller_type: indicates the type of
administration, where they must fill in: seller. Can take values:

● Seller_id: the unique code of your PUBLISHER – if the publisher directly

account in the system manages the account with the seller.
 161

● INTERMEDIARY – if it is a reseller. Tools for supply chain

● BOTH – if it is a combination of both control
of the above.
There are a number of helpers for
controlling and optimizing the above
An example of a sellers.json file can be tools. Just type the phrases “ads.txt” or
found for each participant in the pro- “sellers.json” into a search engine and
grammatic ecosystem. See, for exam- add “validator” or “checker”. Some tools
ple: https://cdn.pubmatic.com/sellers/ are online services, others are browser
data/sellers.json and the entry for the add-ons.
NY Times publisher contained therein: Here are just a few to get you started:
“seller_id”: “158945”, ● Sellers.guide116
“name”: “NY Times”, ● Ads.txt Guru117
“domain”: “Nytimes.com”, ● Ads.txt Checker118
“seller_type”: “PUBLISHER”.

The Demand Path

Ads.cert Optimization (DPO)
While ads.txt, sellers.json, and supply The Demand Path Optimization (DPO) is
chain have reduced fraud, they cer- basically the opposite of SPO. That is, it
tainly haven’t eliminated it completely. is primarily a technique for publishers
Already, another standard from the IAB to boost revenue from advertisers by
has been announced115 to further clean clearing the buying path. It is not just
up the market under the name ads.cert. a technique to increase eCPM, but the
Ads.cert is based on the principle of bid goal of DPO is to bring transparency
encryption and two keys, one public and and trust to the buying process and
one private, known only to the publish- make sure the entire system is set up
er. The launch will take place with the optimally.
transition to the RTB 3.0 protocol.

115
https://iabtechlab.com/press-releases/openrtb-3-0-beta/
116
https://sellers.guide/
117
https://adstxt.guru/
118
https://chrome.google.com/webstore/detail/adstxt-checker/bjddoinjnabgbnloplmjdfdcdlefdbhj
162

Knowing SPO is very important for regularly delivering poor quality

publishers, as we have shown, but pub- ads, then part ways with them. Each
lishers also have other possibilities. For publisher may see the term “poor
example: quality” as something different. For
● Auction data from header bidding, example, promotion of services on
such as win rate, response time the edge of ethics, delivery of ads
(latency), CPM of the bid, etc., allow exceeding certain data limits, or
the publisher to decipher the quality overly aggressive look or feel of ads
of the bids of each demand partner may not be worth the investment.
and thus understand the contribution ● Make sure your bid requests include
of a particular demand partner. as much additional information as
● Inviting major advertisers to the possible, such as the user identity
Private Market Place (PMP) – like that tools discussed in earlier chapters
you minimize the number of systems (Chapter 8, page 120), or proper
between you and the price is clearly linkage to CMP tools (Chapter 9, page
negotiated including technology fees. 142), and forwarding approvals to all
demand partners.
● Monitor the quality of the ads
delivered and if any bidder is

If your analysis shows that a demand partner is not relevant to you, don’t
be afraid to say goodbye. By removing their entry from ads.txt, you’ll let the
whole market know that your space is no longer available via that avenue,
but at the same time you won’t block specific advertisers because they still
have many ways to reach you.
On the other hand, analyze carefully if you are going to cut off any major
demand partners. In general, you should start cutting back on smaller
partners and gradually move to larger ones. Smaller demand partners
generally replicate the demand of the larger ones and rarely have their own
added value.
Also, always ask demand partners about the volume of unique demands
in their system, that is, something that cannot be obtained other than by
engaging just their system.
 163

In 2021, the first definition of buyers.

json was created by the IAB119. It is the
opposite of sellers.json and its essence
is to uniquely identify the advertiser
behind the offer.

Conclusion of the chapter: Supply Path

Optimization and Demand Path
Optimization
In this chapter we: This chapter shows how important it is
● discussed the concepts of SPOs and to follow the bid request path between
DPOs and explained the principles on advertiser and publisher and how inter-
which they operate, esting optimizing can be financially.

● presented an example of the

complexity of the offer’s journey from
publisher to advertiser,
● demonstrated how much money is
“lost” along the way,
● explained the ads.txt, sellers.json and
supply chain projects.

119
https://iabtechlab.com/wp-content/uploads/2021/03/buyers-json-1-0.pdf
164

Chapter 11:
PAGE LOAD SPEED

As we’ll show in the following pages, In the next chapter (Chapter 12, page
loading speed is extremely important. 174), we will also address ad viewability,
And not just the speed of loading ads, lazy load, ad reloading, and other com-
but also the speed of loading the entire ponents that are very closely related
page. Without a fast-loading page, often to the page code, and therefore page
an ad won’t even appear, and if the ad speed.
has low visibility (because it doesn’t
load in a large number of cases), then
the DSPs on the advertiser’s side will
gradually stop targeting them.

Latency
The term “latency” usually refers to the
time that elapses between the time a
request is sent and the time the content
is displayed. So usually the request is
a visitor entering a new web page and
latency is the time it takes for the entire
page to load.
Latency can be measured for the entire
page or for specific elements on the
page. For example, in advertising, we
are interested in the speed from the
time the header bidding starts until all
ads have loaded.
This time is affected by many parame-
ters, some of which are hard for us to
control. We will see some that can be
influenced in the next chapter.
 165

Latency

Latency
click to open
new URL

Figure 11.1. Latency

Studies on speed
In 2017, Google conducted a large test on 11 million mobile ads in 213
countries120. The result provided, among other things, a comparison of the
probability with which a user leaves a page according to the length of the
page load.

Probability of bounce

1s to 3s
the propability of bounce increases 32%

1s to 5s the propability of bounce increases 90%

1s to 6s the propability of bounce increases 106%

1s to 10s the propability of bounce increases 123%

Figure 11.2. Probability of bounce, Source: Google

120
https://www.thinkwithgoogle.com/marketing-strategies/app-and-mobile/mobile-page-speed-new-indus-
try-benchmarks/
166

Alternatively, older information From today’s perspective, this

from Google says that every information published just years
extra 0.5 seconds of generating ago seems prehistoric: A study
a search page reduced traffic by found that every 100ms of delay
20%121. costs it 1% of revenue127. Another
eCommerce study by Radware128
says that 20% of users will
Even though mobile pages load abandon a shopping cart if the
87% slower122, around 2014 the whole buying process is lengthy.
number of mobile users exceeded
desktop users123 and by 2025 it
is estimated that ¾ of users will This and other data say one thing:
access the internet via mobile Speed matters. After all, if we stick to
devices124. a pure header bidding environment, we
get the same response. The “Evolution
of the Header AdMonsters”129 study
In 2018, the growth in the number of shows that page latency is considered
mobile users and the slower loading by publishers to be one of the biggest
of pages on mobile devices led Google challenges in the header bidding space.
to change the parameters by which it
ranks search results. Now, one of the
criteria on mobile devices is loading
speed125.

The BBC has posted126 tips on how

to make pages load faster, and
additionally states that they have
found that they lose an additional
10% of users for every additional
second that their site takes to
load.

121
http://glinden.blogspot.com/2006/11/marissa-mayer-at-web-20.html
122
https://backlinko.com/page-speed-stats
123
https://speedy.site/average-page-load-times-websites-2018/
124
https://www.cnbc.com/2019/01/24/smartphones-72percent-of-people-will-use-only-mobile-for-internet-by-2025.html
125
https://developers.google.com/search/blog/2018/01/using-page-speed-in-mobile-search
126
https://www.creativebloq.com/features/how-the-bbc-builds-websites-that-scale
127
https://www.gigaspaces.com/blog/amazon-found-every-100ms-of-latency-cost-them-1-in-sales
128
https://www.youtube.com/watch?v=UuePkp6TuhQ
129
https://www.admonsters.com/playbook/admonsters-playbook-header-evolution/
 167

What is your top challenge associated with the header

Page latency
Amazon:
82% 78%
Auction timeouts
76%
Low-quality ads or malvertising
48%
Lack of demand
27%
Poor CPMs
24%
Other
6%

Figure 11.3: What is your top challenge associated with the header. Source:
AdMonsters

How to increase the speed of ads

Within header bidding itself, we can most time-consuming parts of a page
increase ad speed via various avenues. to load. Asynchronous processing will
Ultimately, loading speed has a direct make everything run in the background,
impact on the functionality of header in turn reducing the overall page load
bidding and therefore the financial yield time.
of the entire system.
a. However, take into account the need
1) Load JavaScript files to obtain user consent through CMP
asynchronously – For some ad (Chapter 9, page 142). Sometimes the
sources this can be a problem. Howev- activation of header bidding takes place
er, nowadays asynchronous loading is before CMP and the auction will not be
mostly standard and is solved by the executed without consent. Thus, the
wrapper (Chapter 5, page 86). CMP tool code should precede the head-
er bidding code.

2) Place the wrapper in the 3) Use caching and compression

page code first – Set the wrapper – Once you’ve implemented header
JavaScript completely at the beginning bidding for crisp traffic, take advantage
of your page code. Ads are one of the of all compression and caching options.
168

4) Check wrapper loading speed system as a whole must be taken into

– A wrapper is a set of JavaScript. account. In other words, the system will
These have to be loaded into the page be as fast as its slowest component.
on their own and take some time to
execute. If you use one of the open
7) Use lazy load – The principle of
source wrappers, make sure you put it
lazy load is very simple: content starts
on a fast CDN and only use the modules
loading the moment the user is close
you actually need. For example, Prebid
to viewing it. Thus, ads placed in the
is a very complex system made up of
footer won’t load the first time the page
separate modules – only integrate the
loads, when the user hasn’t had time to
modules you use.
scroll all the way down (full discussion in
Chapter 12, page 178).
5) Set a timeout for demand
partners – Setting a timeout for
demand partners within wrappers
(Chapter 5, page 86) is essential and
should never be skipped. Don’t just
settle for setting it, though. Bench-
mark the response of your participating
partners and adjust the timeout period
accordingly. We recommend 1000ms
as a default value and then decrease
or increase according to the result and
monitor the impact.

6) Client-side vs Server-side
header bidding – Server-side solu-
tions work faster than client-side (differ-
ence fully discussed on Chapter 4, page
53), but have other pitfalls of their own
to consider. When combining a hybrid
solution or even using multiple wrappers
at the same time, this difference plays
less of a role because the speed of the
 169

How to increase the speed

of the whole page
Even if your header bidding solution and developers and content people and
all parts of your ads are superfast, you’ll advertising people. Often the ad depart-
still be relying on the loading speed ment employees don’t even know that
of other elements on the page. Below some HTTP/2 is essential for them, and
are some basic tips you can reference the developers in turn have no idea how
to speed up the whole page. However, important this is for ads.
this is only general and more detailed
information is beyond the scope of this
publication. To find out if your website uses
HTTP/2, simply run one of the
online tests. Type “HTTP/2 test”
1) Make use of compression – into Google and you’re sure to find
Try to minimize the size of images with the results you want. Alternatively,
special compression tools. Introduce discuss whether you might be
compression for scripts and HTML. Even interested in switching to the new
HTTP/3 protocol straight away.
simply removing spaces, line breaks
and annotations can have a big impact.
Alternatively, you can also use GZIP. 4) Postpone loading some ele-
ments – Some parts of the website
are too extensive and users don’t need
2) Move your site to CDN – Con-
them in the first seconds. For example,
sider moving not only your ad scripts,
on Learniv.com130 loading verbs into
but your entire site or parts of it to a
the header filter occurs after the entire
CDN. That means HTML code, images,
page has loaded. Footer bidding (Chap-
CSS, and other scripts.
ter 4, page 76) is based on the same
idea.
3) Implement HTTP/2 protocol
– HTTP/2 protocol includes features
5) Consider AMP – If page loading
for faster asynchronous page loading.
speed is critical to you, consider using
This affects not only ads but the overall
the AMP framework131. These are funda-
loading speed of the web page. Check
mental changes at the page code level,
if you are using it and if not, consider
but they take your site to a whole new
implementing it. Or, if you are not using
dimension in terms of speed. But you’ll
it, there should be a very good reason
also need to take a special approach
for it, and even that should be tempo-
to ads and header bidding (Chapter 4,
rary. There is a deep divide between
page 65).

130
https://en.learniv.com/have/
131
https://amp.dev/
170

Tools for analyzing ad loading

speed
There are several tools that can meas- ensure that the page does not scroll
ure the loading speed of ads or an en- when the ad is delayed, but only fills the
tire web page. I will name probably the empty space provided for this purpose.
most used ones: A limited version of SpeedPage is also
available directly in the Chrome browser
in ChromeDevTools.

Google SpeedPage
Insight
This tool is mainly used by webmas-
ChromeDevTools
ters to optimize the loading time of the An integrated tool built directly into
website as a whole. Loading time has an the Google Chrome browser, Chrome-
impact on the page’s ranking in Goog- DevTools shows a time chart of how
le’s search results and probably other content, including ads, is loaded onto
search engines as well. The tool can the website.
analyze all scripts loaded and executed
in the website, including the JavaScript
Use this procedure to find ads:
that makes up the wrapper, and suggest
optimization. Open DevTools by pressing Option
+ z + J (macOS) or Shift + CTRL
SpeedPage shows one more value that
+ J (Windows)
is often related to ads: the Cumulative
Layout Shift. This is an index132 that Navigate to the Network panel
takes into account the shifts of content In the filter, search for the content
on the page as the page loads, typically you are interested in, for example
a shift in text due to an ad appearing “ads?”
above it. People responsible for display-
The results you find can be ex-
ing ads should address this in addition
panded for more detailed informa-
to the webmaster.
tion and a separate timeline
In practice, this means that the site
must define in advance how much space
will be available for each ad unit, spe-
cifically, the maximum height of each
ad unit, and only allow such ad formats
in each ad unit accordingly. This should

132
https://web.dev/vitals/
 171

Figure 11.4: Networks sample. Source: ChromeDevTools

ChromeDevTools can also run the limit-

ed analysis available in Google Speed-
Page Insight. It is available under the
“Audit” or “Litehouse” tab (depending on
your browser version). ChromeDevTools

Figure 11.5: Litehouse sample. Source: ChromeDevToole

172

Headerbid Expert
This is a Google Chrome plugin133 that
shows when communication with a
particular ad demand partner starts and
how long it takes. The program directly
analyzes any problems and gives sug-
gestions for solutions.

Figure 11.6: Headerbid Expert sample. Source: Headerbid Expert

133
https://chrome.google.com/webstore/detail/headerbid-expert/cgfkddgbnfplidghapbbnngaogeldmop?hl=en
 173

Conclusion of the chapter:

Page load speed
In this chapter we: This chapter shows how important it is
● showed several studies proving that to consider page load speed not only
page speed is important, from the perspective of ads, and that
this topic should not only be addressed
● presented tips for improving ad and by technicians, but also by people re-
page loading speed, sponsible for monetizing the site.
● reviewed some tools for speed
analysis.
174

Chapter 12:
VIEWABILITY

The visibility value is determined as a

Viewability is a metric designed to percentage of visible impressions to
measure how much an ad is visible the total number of impressions. So if
on a user’s device. The IAB134 you display one million impressions on
(Interactive Advertising Bureau) has a page and only 500,000 of them meet
defined a standard where a banner
the viewability condition defined above,
ad is considered viewable if it meets
then the viewability value of your ads
the following parameters:
is 50%.
● 50% of the ad unit in view for
a minimum of 1 second for Within the Google Ad ecosystem,
standard ad formats such as viewability is also referred to as Active
leaderboards and MPUs; View135.
● 30% of the ad unit in view
for 1 second for large canvas
formats such as skyscrapers
and takeovers;
● 50% of the ad unit in view for 2
consecutive seconds for video
ads.

Viewability is important
In general, all advertisers want their ads buying platforms are already optimized
to be visible, and they monitor these to this target and use machine learning
statistics closely. It is one of the most algorithms that learn incrementally from
watched statistics ever. the data collected.
It’s logical. If the user doesn’t see the Visibility is crucial for publishers and it
ad, they are not influenced by it and is also in their interest to maximize it on
cannot click on it. There is a significant their site. On the other side, publishers
effort on the part of advertisers and are struggling with the fact that content
all systems on the advertiser side to is even more important to them, and
buy ads with maximum visibility. Most are therefore trying to strike a balance
between ad viewability and content.

134
https://www.iabuk.com/news-article/quick-qa-viewability
135
https://support.google.com/google-ads/answer/7029393?hl=en
 175

In 2013, a study from Comscore136 was presented showing that on average,

only 46% of ads met the definition of viewability. Thanks to advertisers’
focus on this issue, ad viewability is gradually increasing. The 2020 study137
shows an average viewability of about 72%.
In 2014, Google138 released an interesting series of data139 on average view-
ability. In it, it shows that 56% of ads are never seen140 by users and that
above-the-fold ads have an average of 68% viewability, and below-the-fold
ads have an average of 40% viewability. An explanation of these concepts is
given on page 176.

Difference in above-the-fold and below-the-fold visibility

Median viewability for Above-the-fold ad units Median viewability for below-the-fold ad units

68% 40%
% OF PUBLISHERS DOMAINS

% OF PUBLISHERS DOMAINS

ATF VIEWABILITY DISTRIBUTION BTF VIEWABILITY DISTRIBUTION

35% 25%
30% 20%
25%
20% 15%
15% 10%
10%
5% 5%
0% 0%
%

%
20 %
30 0%

40 %
50 0%
60 0%
70 0%
80 0%
90 0%

20 %
30 0%

40 %
50 0%
60 0%
70 0%
80 0%
90 0%
10 %

10 %
00

00
0

0
10

10
-2

-3
-4

-5
-6
-7
-8
-9
-1

-2

-3
-4

-5
-6
-7
-8
-9
-1
0-

0-

Figure 12.1: Difference in above-the-fold and below-the-fold visibility. Source:

TheDrum

136
https://www.comscore.com/Insights/Blog/Viewability-Benchmarks-Show-Many-Ads-Are-Not-In-View-but-
Rates-Vary-by-Publisher
137
https://go.integralads.com/us-mqr-2020-h1.html
138
http://think.storage.googleapis.com/docs/5-factors-of-viewability_infographics.pdf
139
https://www.thedrum.com/news/2014/12/03/56-online-display-ads-are-not-seen-consumers-claims-goo-
gle-first-ad-viewability
140
https://martech.org/googles-report-56-percent-ads-arent-seen-isnt-shocking-heres/
176

For publishers, this sends a clear Even the “viewable CPM” (or vCPM)
signal: advertisers are watching view- metric has gradually evolved and some
ability and trying to buy ad space that systems, including Google141, allow you
meets this definition. This means that to buy just the impressions that meet
these ad spaces are in higher demand this condition.
and have higher average eCPMs. Nat-
urally, ad spaces with lower visibility
have lower eCPMs.

Above-the-fold
and Below-the-fold
In the context of visibility, the terms
Above-the-fold (ATF) and Below-the-fold
(BTF) appear. Originally, the designa-
tion has sources in print newspapers.
These generally have a fold, which is
significant because only information ABOVE
above that fold was visible at a glance
THE FOLD
on newsstands.
In the online advertising world, they THE FOLD
indicate whether an element on a page
is visible on the first load of the screen
without the user scrolling. So it depends BELOW
a lot on the screen size of the user’s THE FOLD
device. A desktop monitor can fit more
visible elements than a mobile device
with a small screen.
For each ad position on your page, you Figure 12.2: Explanation of the
should be aware of whether it is ATF or difference between above-the-fold and
BTF with respect to the most common below-the-fold

141
https://support.google.com/google-ads/answer/3499086?hl=en
 177

device type and monitor resolution. This is due to a simple reason – often
These values are typically very different the user scrolls on the page before the
for desktop and mobile versions of a top ad has a chance to load. Ads placed
website. at the border of ATF and BTF, or at the
However, not every ATF position neces- beginning of an article or other content
sarily needs to have high visibility. On for which the user visited the page, usu-
the contrary, positions completely above ally have the best visibility.
the page header have lower visibility.

How to increase ad visibility

● Play with ad positioning – used. Publishers can be advised to set
Above-the-fold placements are visible size limits on the ads that demand part-
when the screen is first loaded, so ners can pass through to the site.
they generally have higher visibili- Optimize page load speed – Ads are
ty. We already mentioned the lower a drop in the bucket of the total num-
visibility of full-header ads. However, ber of elements that load on a page,
you should also consider the so-called and the whole process takes time. In
“sticky” format (page 181) in your ad practice, it may happen that the user
placement. That is, an ad that stays on has already read the article and the ads
the screen even when the user scrolls. are gradually loading to the positions
From a viewability perspective, this is a they have already skipped. Therefore,
good method, just consider that some the activation of ads should be handled
demand partners may have special as high and asynchronously as possible
requirements and rules for these for- in the entire cascade of HTTP requests.
mats142. The topic of page loading speed is cov-
Take advantage of ad loading speed ered more closely in this book in sepa-
– This one is easy. A 50kB ad banner rate chapters on loading speed (Chapter
loads faster than a 500kB banner. That’s 11, page 164) or lazy loading (page
why most ad systems also give adver- 179).
tisers maximum limits on the size of ads

142
https://support.google.com/admanager/answer/7246067
178

Consider the viewability of different of the refresh function on page 181).

ad formats – In 2014, emarketer.com143 Be careful here though, each ad should
released a survey monitoring different be seen for a sufficient amount of time.
ad formats with respect to their aver- Changing ads that have not yet been
age viewability. The data has probably in the user’s field of view at all can be
changed since 2014, but the basic idea counterproductive. Some demand part-
remains the same. A 300x600 format, ners may also have problems with the
on average, will have higher viewability refresh function and have prescribed
than a 300x250 for the simple fact that conditions under which it can be used.
it’s larger in height and takes a user a — Keep track of miscellaneous
while to rescroll. factors – Visibility is also significantly
Keep your site simple – Sometimes, impacted by a number of parameters
the user has a slow connection. This is such as whether it works asynchronous-
a difficult problem for publishers to ly, how many ad demand partners are
solve. In general, the simpler the site, calling, and how much time (timeout) it
the faster (or if at all) it will load for gives them to respond (see Chapter 5,
users with slow connections. Video con- page 86 for full discussion).
tent in particular can be problematic for
slow connections, so consider when and
where to implement it.
Consider using the refresh function
– If a user has been seeing an ad for a
while, it may be possible to replace it
with another one (see the description

Lazy load
One of the essential tools to increase Lazy load is a technique that allows ads
visibility and speed up page loading is to be shown at the exact moment when
lazy load. It is also sometimes referred the user is likely to actually see them.
to as Smart loading or Just-in-Time But it is not just ads, lazy load can also
loading. load images or other entities on a web
page later.

143
https://www.emarketer.com/Article/What-Display-Ad-Sizes-Drive-Best-Viewability-Rates/1011670
 179

To give an example, if I am a visitor to a Possible complications

site and I am reading an article, I scroll
down gradually. So the advertisement of lazy load
at the end of the article is not loaded by As great as lazy load looks, it has down-
lazy load when the page is first loaded, sides that you need to keep in mind:
but only when I scroll to the end of the
● Regardless of lazy load functionality
page. This greatly reduces the number
within programmatic buying, you
of HTTP requests that the web brows-
should also consider the suitability of
er has to make. Page content loading
this feature within campaigns offered
is delayed until the point where it is
directly by your ad server. Keep
needed, or it doesn’t load at all because
in mind that a lazy load increases
it is not needed (the user left the page
visibility but may reduce the number
before he/she scrolled down).
of impressions.
So, by definition, it only makes sense to
● If a user is scrolling through the site
apply the lazy load function to ads out-
quickly, they may not have time to
side the first screen (see Above-the-fold
load and view the ad before scrolling
on page 176).
down and may not see it at all, which
will affect viewability value.
● Lazy load has a direct impact on
Search Engine Optimization (SEO);
Advantages of lazy simply because if some elements
load aren’t loaded, the Google bot can’t
There are two major advantages of this possibly know about them. This may
strategy: not be a problem for ads, but for
image indexing it is crucial. However,
● The web page loads faster because this issue goes beyond the scope of
it doesn’t transfer unnecessary this book, so I would recommend
amounts of ad and image data that consulting sources dealing with SEO
isn’t needed at the beginning. or Google144 directly.
● It fundamentally increases the ● Lazy load is not a simple feature–
visibility of ads, because they only its deployment needs to be done
appear when they are actually visible cautiously and tested before a sharp
to the user. start to see what impact it will

144
https://developers.google.com/search/docs/advanced/javascript/lazy-loading
180

have on your site. This is a task for General uses of lazy

developers. Moreover, like everything
else, lazy load needs occasional load
servicing and maintenance, and Exactly how much lazy load can speed
yield managers should double-check up web page loading is hard to gen-
once in a while in analytics tools that eralize and depends on the content of
everything is working as it should. the page. Generally speaking, the more
data a page contains beyond the first
visible screen, the more you can speed
A nice description of the technical
it up. An experiment on how a very lazy
integration of the lazy load fea-
load reduces loading speed has been
ture can be found on the Google
published on Researchasahobby.com148.
developer site145 or on Web.dev146.
Alternatively, you can find an help-
ful description of the integration of
the lazy load feature for Adsense
ads on Gulshankumar.net147.

Page Load Test

Average Data (25 tests for each option)
14
12
10
8
Seconds

6
4
2
0
Lazy Load is NOT used Lazy Load is used
Start Render Visually Complete Load Time

Figure 12.3: Page load test. Source: Researchasahobby.com

145
https://developers.google.com/publisher-tag/samples/lazy-loading
146
https://web.dev/browser-level-image-lazy-loading/
147
https://www.gulshankumar.net/setup-lazy-loading-google-adsense-ad-units/
 181

Sticky ads
Another feature for increasing ad visibil- ● A mobile ad pinned to the top or
ity is sticky ads, sometimes referred to bottom edge of the mobile browser
as Anchor ads. and remains visible even if the user
With this feature, the ad is “pinned” scrolls.
and moves with the user’s screen. It ● Sticky ads are usually subject to
is therefore visible for a much longer special requirements and rules by
period of time. demand partners which you should
Probably the most well-known examples know about before venturing into this
of sticky ads are: area.

● A desktop ad in the right or left

column next to an article. It remains
visible even if the user scrolls lower
as they read the article.

Refreshing ads
The last feature we’ll show that can rent status with your demand partners
make a big difference to visibility is the before you decide to use this strategy.
refresh feature.
Technically, it is possible to refresh ads
on a page without refreshing the entire It only makes sense to refresh ads
page. This topic is slightly controversial that are valuable, i.e. highly visi-
because it allows for potential abuse. On ble, or have a high CTR. Refresh-
the other hand, when used properly, it ing ads with minimal visibility is a
can be beneficial. bad idea. Most buying platforms
As with sticky ads (page 181), refresh already have machine learning
is subject to special requirements from that analyzes data over time and
demand partners. Some even strict- learns where it makes sense to
ly prohibit them (for example, Google compete for ads next time and
AdSense). Others allow them to be used where it doesn’t.
under strict conditions. Check the cur-

148
https://researchasahobby.com/display-images-only-when-need-lazy-load/
182

● For demand partners who prohibit

That’s why generally low-visibility this (for example, Google AdSense).
ads, and possibly even refreshed ● An ad position that has not yet been
ads, severely damage the score visible (the user has not yet scrolled
that this ad position–or even the to it).
whole page–gets with shopping
platforms.

There are several opportune moments Bid caching

to consider refreshing an ad position:
In certain cases, it is possible to store
● When the page content is refreshed the results of demand partners within
(we’re still talking about not the auction for later use. This is possible
refreshing the entire page) – e.g., with refreshing ads.
when the result of a filter is loaded
Imagine you are running an auction for
after a user interaction, like when,
a specific ad unit that is about to be
viewing a photo gallery.
displayed to a user. You get three bids
● After a certain amount of time (e. g. from your demand partners. Your ad
60 seconds) the user has been on server selects one as the winning bid
the page. This is particularly useful and displays it. But you have two others
for sticky ads on pages where users that have not been satisfied. Yet the
spend longer periods of time. partners are clearly interested in this
● After the page has been automatically ad.
refreshed by the server, for instance, If you’re running an auction for the
when watching the result of an online same user on the same ad unit and the
sports match. same URL after 60 seconds, when using
On the contrary, sometimes you should bid caching, you can use the bid from
avoid refreshing the advertising posi- the previous auction.
tion:

51
https://www.adpushup.com/blog/header-bidding-adoption-three-years-in-review/
51
https://www.kevel.co/hbix/
53
https://adpushup.wistia.com/medias/swzqclb2xx
 183

Combination of functions
in header bidding
All three features listed above can be
linked to header bidding in an synergetic
way. We already discussed how it’s useful to
In certain cases, it may make sense not implement the refresh function when
to perform the auction in the header the ad has been visible for a certain
of the page, but to wait until the user period of time. Thus, the auction for the
scrolls to the ad position (lazy load), demand partners is re-run and a new
then perform the auction and display winner is selected. Under certain condi-
the ad. At the same time, it might be a tions, we can also use the bids response
good idea to pin the ad (sticky ads) to from the previous auction (bid caching)
the screen and refresh it after a certain to fill the empty unit.
period of time, as it will be visible for a At the same time, these functions often
long time. offer a wrapper solution and can there-
fore be managed relatively easily.

Conclusion of the chapter: Visibility

In this chapter we: This chapter describes how a publisher
● showed that visibility is important can improve revenue by increasing ad
and how we can increase it, viewability.

● explained the concepts of Above-the-

fold and Below-the-fold,
● introduced the special practices of
lazy load, sticky ads, refresh and bid
caching.
184

Conclusion
This book has attempted to describe The code below is for registration to the
header bidding in its complexity without HeaderBiddingBook.com website. If you
singling out or diminishing any particular register you will gain access to the on-
possible solution or technology. The goal line version of this book and to follow-up
was not to push one ideal approach, information that we plan to publish in
because none exists. the future for free to registrants.
I wanted to present the principle of the If you find this book beneficial, I would
whole solution because when you un- appreciate a brief review on Amazon or
derstand the principle and the context, another global retailer.
you have a much better chance of put-
ting the whole puzzle together to meet
your unique needs. Your code to the bonus materials:

Certainly not everything has been ENF100

described in its full breadth or depth.
On the contrary, I have tried to avoid
too much detail, which can vary within
specific technologies.
At the same time, many things are
constantly evolving and some technol-
ogies and practices will eventually be
surpassed and replaced by others. But
I believe that the principle of head-
er bidding will endure for a relatively
long time because of its simplicity. It
is essentially auction selling, and it has
been known and used by humanity for
hundreds of years.
I welcome any comments and sugges-
tions, please do not hesitate to contact
me. I find the exchange of experience
and knowledge very beneficial and
prefer face-to-face rather than online
discussion if at all possible.
185
186

Used terms
Term Description
Elements on a web page that are visible on the first screen view (see page 176
Above-the-fold (ATF)
for details).
Accelerated Mobile
A technology designed to load web page content very quickly.
Pages (AMP)
A programmatic marketplace where DSPs and SSPs communicate (see page 15
Ad exchange
for details).
Ad fraud Abbreviation for online advertising fraud.
Ad server A system for publishers to easily manage ads (see page 106 for details).
Ad unit Abbreviation for advertising unit, usually used in an ad server.
The successor of the Ads.txt file within the RTB 3.0 protocol (see page 161 for
Ads.cert
details).
A simple text file that publishers use to declare who is authorized to sell their
Ads.txt
space (see page 155 for details).
Advertising ID Con- An association of several major companies in the field of online advertising with
sortium activities related to user ID solutions.
Asynchronous A method of loading content where individual elements are loaded simultane-
loading ously (in parallel).
Auction of auctions One of the terms historically used for header bidding.
Elements on the web page are visible only after the first screen is scrolled past
Below-the-fold (BTF)
(see page 176 for details).
A method to indicate the use of an unwinnable bid request at a later time (see
Bid caching
page 182 for details).
Bid latency The time from sending the bid request to displaying the ad.
Ratio of positive responses to all bid request responses (see page 99 for
Bid rate
details).
A request for a new impression offer generated by the publisher’s system and
Bid request
sent to the bidders (see page 97 for details).
A response sent by a prospective advertiser to a bid request (see page 97 for
Bid response
details).
An algorithm that tries to estimate the optimal bid amount in an auction (see
Bid shading
page 41 for details).
Blackbox A term for a system or tool whose internal workings are not known.
The opposite of sellers.json, used to uniquely identify the advertiser behind the
Buyers.json
bid (see page 161 for details).
A way of storing content for subsequent quick use without having to recreate it
Caching
or recall it from a slower source.
 187

Term Description
California Consumer
Privacy Act for pub- A designation for legal protection of users in California.
lishers (CCPA)
Generally a percentage reduction in the amount of bids with respect to the
Bid adjustment
technology fees (see page 87 for details).
Client-side head-
A form of header bidding taking place in the user’s web browser (see page 48
er bidding (C2S or
for details).
CSHB)
Consent Management Platforms that manage users’ consent to the use of their data (see page 142 for
Platform (CMP) details).
Cookie matching/ A label for how information in cookies is matched between different domains
Cookie syncing (see page 125 for details).
Short text files created by a website and allowing data to be stored on the
Cookies
browser side (see page 120 for details).
Cookies, third-party Cookies used by external domains (see page 123 for details).
Cookies, first-party Cookies used by the website domain (see page 123 for details).
Cookies, second-par-
Special type of first-party cookies (see page 123 for details).
ty
CPM Price per ad impression per 1000 impressions (see page 100 for details).
Daisy chain Another term for waterfall.
Demand Path Optimi- A set of procedures on the publisher’s side to optimize the paths through the
zation (DPO) purchase chain to the advertiser (see page 161 for details).
Demand Side Plat- A tool to represent the interests of the buyer (advertiser) in programmatic
form (DSP) buying (see page 15 for details).
Deterministic match-
A method of matching users based on an explicit principle.
ing
Discrepancy Indicates the difference between data from different systems.
The time required to perform a certain operation, for example, to load a page
Latency
(page latency).
Technology capable of being connected to an auction operated by the publisher
Demand partner
(usually an SSP) (see page 17 for details).
DoubleClick Predecessor of Google DV360.
DoubleClick for Pub-
Original name for Google Ad Manager – Google’s ad server solution.
lishers (DfP)
DV360 Google’s DSP tool.
A feature in Google Ad Manager that allows Google to continue to bid higher
Dynamic allocation
after an auction has ended.
188

Term Description
eCPM Average cost per impression for 1000 ad impressions (see page 100 for details).
Exchange Bidding in
The original name for Open Bidding – Google’s S2S header bidding solution
Dynamic Allocation
(see page 57 for details).
(EBDA)
Indicates the proportion of impressions sold out of the entire bid (see page 100
Fill rate
for details).
A type of programmatic deal that guarantees the buyer’s participation in the
First look
preferred auction.
A method of auctioning advertising where the winner pays the price he or she
First price auction
bid (see page 36 for details).
The minimum price a publisher asks for the provision of its advertising space
Floor price
(see page 34 for details).
A type of header bidding triggered after all other page elements have been
Footer bidding
loaded (see page 76 for details).
General Data Pro-
Designation for the legal protection of users in the European Union (see page
tection Regulation
134 for details).
(GDPR)
Ghost bids Fake bids at auction.
Global Vendor List Part of the TCF standard listing all participating technologies.
Google Ad Manager Ad server solution of Google.
Google Ads (earlier
Google’s buying platform for advertisers.
AdWords)
Google AdSense Google’s monetization platform for publishers.
Google AdX Google Ad Exchange.
Minimum price below which advertisers’ offers are never accepted (see page
Hard floor price
40 for details).
A method of online auctions in a programmatic environment that ensures equal
Header bidding
conditions and time for all auction participants (see page 45 for details).
Calling URL. Usually a web page that contains many requests to other resourc-
HTTP request
es that are loaded into the page, such as CSS stylesheets, advertisements, etc.
A version of the HTTP protocol that speeds up asynchronous loading of ads
HTTP/2
(see page 167 for details).
Hybrid header bid-
A form of header bidding combining C2S and S2S (see page 57 for details).
ding
Impression Ad impression counter (see page 99 for details).
The name used for header bidding in the environment application (see page 65
In-app bidding
for details).
 189

Term Description
Interactive Advertis- An organization established to standardize, research, and promote the online
ing Bureau (IAB) advertising industry.
Designation for pairing different user ID solutions with each other (see page
Interoperability
129 for details).
One of the basic programming languages used in the World Wide Web environ-
JavaScript
ment.
A piece of programming code written in JavaScript, in an advertising technolo-
JavaScript tag
gy environment usually presenting an ad call trigger from an external source.
Just-in-Time loading Another name for lazy load (see page 178 for details).
A technique that allows ads to be shown at the exact moment they are likely to
Lazy load
be shown to the user (see page 178 for details).
Lei Geral de Proteção
Term for legal protection of users in Brazil.
de Dados (LGPD)
The term for an advertising campaign or one specific targeting within a larger
Line item
campaign, usually used in an ad server.
Midroll Designation for a video ad placed in the middle of video content.
A state where no response is received to a bid request because there is no
No-bid
interest in the bid (see page 97 for details).
Open Bidding S2S header bidding solution from Google (see page 57 for details).
An open tool developed by the wider community with the possibility of further
Open source
customization.
OpenRTB Consortium A consortium founded in 2010 to standardize Real Time Bidding.
A method of collecting consent from users for the use of their data based on
Opt-in
the need to manually give consent (see page 137 for details).
A method of collecting consent from users to use their data unless they manu-
Opt-out
ally revoke consent (see page 137 for details).
Page latency Time required to load a web page.
A marker for header bidding in the enviroment of an application (see page 69
Parallel bidding
for details).
Method of calling the next demand partner in waterfall (see page 29 for
Passback
details).
Personal Data Pro-
Designation for legal protection of users in Thailand.
tection Act (PDPA)
Personal Information
Protection and Elec-
Designation for legal protection of users in Canada.
tronic Document Act
(PIPEDA)
190

Term Description
Type of header bidding executing directly from the ad server (see page 73 for
Post-bid
details).
Postroll Designation for a video ad placed after the video content.
Prebid The most widely used open source header bidding solution.
Pre-bid One of the historically used names for header bidding (see page 45 for details).
Preroll A designation for a video ad placed in front of a video content.
Usually a commercial tool operated by a private company in which the user has
Proprietary solutions
no right to make modifications. The opposite of open source.
Probabilistic match- A method of matching users based on the probabilistic principle (see page 128
ing for details).
Programmatic deal A commercial agreement between buyer and seller in an RTB environment.
Programmatic buying An automated way of buying online advertising (see page 14 for details).
Programmatic direct/
One form of programmatic buying based on the method of selling/buying at
Programmatic guar-
fixed agreed volumes.
anteed
An IAB initiative to develop a solution for ad targeting after the end of cookies
Project Rearc
(see page 127 for details).
PubCommon ID User identity solution (see page 128 for details).
Real-Time Bidding One form of programmatic buying based on the auction selling/buying method
(RTB) (see page 14 for details).
Real-time configura- A method of header bidding in the AMP site environment (see page 65 for
tion (RTC) details).
A method used to increase ad visibility by refreshing ads after a short period of
Refresh
time (see page 181 for details).
A general term for ad revenue, also otherwise referred to as yield (see page
Revenue
100 for details).
Search Engine Opti-
A set of techniques to increase traffic to a website from search engines.
mization (SEO)
A method of auctioning advertising where the winner pays not the price they
Second price auction bid, but the price of the second highest bid plus one cent (see page 35 for
details).
A file stored on the demand partner side, by which publishers declare who is
Sellers.json
authorized to sell their space (see page 159 for details).
Server-side head-
A form of header bidding running on the publisher’s server (see page 48 for
er bidding (S2S or
details).
SSHB)
SharedID User identity resolution (see page 128 for details).
 191

Term Description
Smart loading Another name for lazy load (see page 178 for details).
A price that, although relatively low for the publisher, below which they are still
Soft floor price
willing to accept an offer (see page 40 for details).
Sticky ads An element that moves with the user’s screen (see page 181 for details).
Supply Path Optimi- A set of procedures, especially on the advertiser side, to optimize the paths
zation (SPO) through the purchase chain to the publisher (see page 148 for details).
Supply Side Platform A tool to represent the interests of the seller (publisher) in programmatic buy-
(SSP) ing (see page 15 for details).
An object that stores the bid path between the advertiser and the publisher
SupplyChain
(see page 159 for details).
Switchboard The interoperability solution (see page 129 for details).
A method of loading content where individual elements are loaded sequentially
Synchronous loading
in a row.
A term for a tool that allows external pieces of code to be inserted into a web
Tag manager
page through the administration.
Tagless One of the terms historically used for header bidding (see page 45 for details).
The Privacy Sandbox Google’s cookie replacement initiative (see page 131 for details).
A condition where no response to a bid request arrives at the requested time
Timeout
(see page 97 for details).
Ratio of timeout responses to all bid request responses (see page 98 for de-
Timeout rate
tails).
Transparence and
Standard published by the IAB in the area of CMP tools (see page 140 for
Consent Framework
details).
(TCF)
Transparent Ad Mar- Amazon’s S2S header bidding solution for larger publishers (see page 60 for
ketplace details).
Unified Ad Market- Amazon’s S2S header bidding solution for smaller publishers (see page 65 for
place (UAM) details).
A term for header bidding in the environment of applications (see page 62 for
Unified bidding
details).
Unified ID/Univer-
User identity resolution (see page 127 for details).
sal ID
User ID User identification technologies (see page 127 for details).
A ratio used to indicate what percentage of ads were visible to the user (see
Viewability
page 174 for details).
The ratio of how many impressions were seen to the total number of impres-
Viewability rate
sions (see page 101 for details).
192

Term Description
Indicates the CPM price converted to viewable impressions only (see page 101
Viewable CPM (vCPM)
for details).
A way to engage multiple demand partners and pass the ad offer between
Waterfall
them (see page 24 for details).
Waterfall tag Another name for waterfall (see page 24 for details).
Win rate The winning ratio of a given resource in an auction (see page 100 for details).
Wrapper Control interface for header bidding (see page 82 for details).
A generic term for advertising revenue. Also otherwise referred to as Revenue
Yield
(see page 100 for details).
A type of job in which an employee handles programmatic systems on the
Yield manager
publisher’s side.
193
194

Sources
https://www.admonsters.com/playbook/admon-
1 1
sters-playbook-header-evolution/

2 https://www.iab.com/guidelines/openrtb/ 2

3 https://prebid.org/blog/happy-birthday-prebid-js/ 3

4 https://prebid.org/blog/announcing-prebid-1-0/ 4

https://www.emarketer.com/chart/227616/us-head-
5 er-bidding-adoption-jan-2018-march-2019-of-total- 5
hbix-sites

https://www.emarketer.com/content/five-charts-the-
6 6
state-of-header-bidding

https://www.emarketer.com/content/more-publish-
7 7
ers-are-using-a-hybrid-header-bidding-approach

https://digiday.com/media/telegraph-used-head-
8 er-bidding-increase-programmatic-revenue-70-per- 8
cent/

https://digiday.com/media/header-bidding-publish-
9 9
ers-boosting-cpms-much-50-percent/

https://admanager.google.com/home/resources/pub-
10 10
lift-uses-exchange-bidding-to-deliver-higher-cpms/

https://admanager.google.com/home/resources/
11 11
dazn-programmatic-video-exchange-bidding/
 195

https://admanager.google.com/home/resources/vid-
12 12
eo-spacestation-24-exchange-bidding/

https://www.jisem-journal.com/download/header-
13 bidding-as-smart-service-for-selling-ads-in-the-dig- 13
ital-era-8483.pdf
https://www.adexchanger.com/the-sell-sider/as-goo-
14 gle-enters-heading-bidding-game-whats-the-impact- 14
on-client-side-integrations/

https://law.stanford.edu/publications/why-goo-
15 15
gle-dominates-advertising-markets/

16 https://www.bbc.com/news/technology-57567799 16

https://adage.com/article/digital/googles-digi-
17 tal-ad-dominance-harming-marketers-and-publish- 17
ers-says-new-study/2257576

18 https://en.wikipedia.org/wiki/Demand_curve 18

https://www.emarketer.com/content/more-publish-
19 19
ers-are-using-a-hybrid-header-bidding-approach

https://www.blog.google/products/admanager/
20 simplifying-programmatic-first-price-auctions-goo- 20
gle-ad-manager/
https://www.blog.google/products/admanager/roll-
21 ing-out-first-price-auctions-google-ad-manager-part- 21
ners/

https://www.emarketer.com/api/RedirectAs-
22 22
set?a=chart&r=242771
196

https://www.emarketer.com/chart/218560/digital-
ad-impression-share-among-us-supply-side-plat-
23 23
forms-ssps-by-auction-type-dec-2017-march-2018-
of-total-impressions-analyzed-by-getintent

https://www.anderson.ucla.edu/faculty_pages/rob-
24 24
ert.zeithammer/SoftFloors_0719.pdf

25 https://arxiv.org/pdf/2009.01360.pdf 25

https://www.adexchanger.com/online-advertising/
26 rubicon-grows-revenue-but-bid-shading-is-driving- 26
down-publisher-cpms/

https://www.adexchanger.com/the-sell-sider/shin-
27 27
ing-a-light-on-the-darker-side-of-bid-shading/

https://www.statista.com/statistics/1227654/wrap-
28 28
pers-usa/

https://www.emarketer.com/content/more-publish-
29 29
ers-are-using-a-hybrid-header-bidding-approach

https://www.blog.google/products/admanager/an-
30 30
nouncing-exchange-bidding-open-beta/

https://support.google.com/admanager/an-
31 31
swer/7128453

https://support.google.com/admanager/an-
32 32
swer/7128958

https://www.accc.gov.au/system/files/Google%20
-%20Report%20from%20Daniel%20Bitton%20
33 33
and%20Stephen%20Lewis%20%285%20May%20
2020%29.pdf
 197

https://aps.amazon.com/aps/2018/03/05/unified-ad-
34 marketplace-uam-vs-transparent-ad-marketplace- 34
tam/

https://iabeurope.eu/knowledge-hub/iab-europe-
35 35
adex-benchmark-2020-report/

https://pubmatic.com/wp-content/uploads/2020/01/
36 Video_Advertising_Requires_New_Monetization_ 36
Strategies_TLP.pdf

https://github.com/ampproject/amphtml/blob/main/
37 37
extensions/amp-a4a/rtc-documentation.md

38 https://docs.prebid.org/formats/amp.html 38

39 https://prebid.org/product-suite/prebid-mobile/ 39

https://on.emarketer.com/rs/867-SLG-901/images/
40 Branch-Mobile%20Trends%202020%20Report%20 40
Sponsorship.pdf

https://www.statista.com/statistics/377808/distribu-
41 41
tion-of-facebook-users-by-device/

https://pubmatic.com/wp-content/uploads/2017/08/
42 42
PubMatic-UnderstandingAuctionDynamics.pdf

43 https://docs.openx.com/#understanding-first-price 43

https://support.google.com/admanager/an-
44 44
swer/9298008
198

https://docs.prebid.org/overview/what-is-post-bid.
45 45
html

https://www.admonsters.com/what-is-footer-bid-
46 46
ding/

47 https://prebid.org/product-suite/managed-services/ 47

https://roxot.com/report/sell-side-programmat-
48 48
ic-us-2019

49 https://www.kevel.co/hbix/ 49

50 https://docs.prebid.org/dev-docs/modules/ 50

51 https://arxiv.org/pdf/1907.12649.pdf 51

https://www.adpushup.com/blog/header-bidding-
52 52
adoption-three-years-in-review/

53 https://adpushup.wistia.com/medias/swzqclb2xx 53

https://www.emarketer.com/chart/248339/which-
54 supply-side-platforms-ssps-have-us-publishers- 54
used-past-12-months-of-respondents-march-2021

55 https://docs.prebid.org/overview/ga-analytics.html 55

56 https://docs.prebid.org/overview/analytics.html 56
 199

https://docs.prebid.org/prebid-mobile/
57 57
adops-price-granularity.html

https://docs.prebid.org/overview/how-to-simplify-
58 line-item-setup.html#how-to-simplify-line-item-set- 58
up

59 https://docs.prebid.org/tools/line-item-manager.html 59

https://blog.google/products/chrome/updated-time-
60 60
line-privacy-sandbox-milestones/

https://blog.mozilla.org/en/products/firefox/to-
61 days-firefox-blocks-third-party-tracking-cook- 61
ies-and-cryptomining-by-default/

https://webkit.org/blog/10218/full-third-party-cook-
62 62
ie-blocking-and-more/

63 https://www.w3counter.com/globalstats.php 63

https://site.adform.com/company/press-releases/
64 adform-calls-on-cmos-to-ramp-up-adoption-of-first- 64
party-ids-after-publishers-prove-they-are-ready/

65 https://en.wikipedia.org/wiki/Lou_Montulli 65

https://services.google.com/fh/files/misc/disabling_
66 66
third-party_cookies_publisher_revenue.pdf

https://digiday.com/media/the-scale-of-the-prob-
67 lem-is-enormous-apple-flexes-strong-anti-tracking- 67
stance/
200

https://www.theinformation.com/articles/apples-ad-
68 68
targeting-crackdown-shakes-up-ad-market

https://papers.ssrn.com/sol3/papers.cfm?abstract_
69 69
id=3020503

https://papers.ssrn.com/sol3/papers.cfm?abstract_
70 70
id=2421405

https://weis2019.econinfosec.org/wp-content/up-
71 71
loads/sites/6/2019/05/WEIS_2019_paper_38.pdf

72 https://acfou.medium.com/abstract-2fefd374edb2 72

https://www.ft.com/content/4c19e387-ee1a-41d8-
73 73
8dd2-bc6c302ee58e

74 https://iabtechlab.com/project-rearc/ 74

75 https://www.adidentity.org/ 75

https://docs.prebid.org/dev-docs/modules/userId.
76 76
html

https://www.emarketer.com/content/will-the-adver-
77 tising-id-consortium-solve-marketers-audience-iden- 77
tity-problems

78 https://privacysandbox.com/ 78

https://docs.prebid.org/dev-docs/modules/userId.
79 79
html
 201

80 https://docs.prebid.org/features/firstPartyData.html 80

https://github.com/prebid/Prebid.js/blob/master/
81 81
modules/userIdTargeting.md

https://www.mckinsey.com/business-functions/
82 risk-and-resilience/our-insights/the-consumer-da- 82
ta-opportunity-and-the-privacy-imperative
https://www.statista.com/chart/21081/coun-
83 tries-most-comfortable-to-share-personal-data-on- 83
line/

https://www.marketingcharts.com/customer-centric/
84 84
privacy-and-security-83114

https://ec.europa.eu/info/law/law-topic/data-protec-
85 85
tion_en

https://iabtechlab.com/wp-content/uploads/2020/08/
86 86
IAB_TechLab_Privacy_IABEurope_EN.pdf

87 https://gdpr-info.eu/ 87

88 https://www.enforcementtracker.com/ 88

https://eur-lex.europa.eu/legal-content/EN/TX-
89 89
T/?uri=celex%3A31995L0046

https://eur-lex.europa.eu/legal-content/EN/
90 90
ALL/?uri=CELEX%3A32002L0058

https://eur-lex.europa.eu/legal-content/EN/TX-
91 91
T/?uri=celex%3A32009L0136
202

https://analytics.dhl.com/despite-gdpr-up-to-70-an-
92 alytics-opt-in-rates-why-extensive-testing-is-worth- 92
ervery-minute-of-effort/

https://github.com/InteractiveAdvertisingBureau/
93 93
GDPR-Transparency-and-Consent-Framework

94 https://iabeurope.eu/vendor-list-tcf-v2-0/ 94

95 https://iabeurope.eu/cmp-list/ 95

https://support.google.com/admanager/an-
96 96
swer/9805023?hl=en

https://iabeurope.eu/iab-europe-transparency-con-
97 97
sent-framework-policies/

98 https://www.kevel.co/cmp/ 98

https://docs.prebid.org/dev-docs/modules/consent-
99 99
Management.html

https://github.com/InteractiveAdvertisingBureau/
GDPR-Transparency-and-Consent-Framework/
100 100
blob/master/TCFv2/IAB%20Tech%20Lab%20-%20
CMP%20API%20v2.md

https://chrome.google.com/webstore/detail/cmp-
101 101
check/odemilkphklhfejgajeaohjcdfmmfjde?hl=en

https://chrome.google.com/webstore/detail/azeri-
102 102
on-ad-expert/nndadbimjipilgfojofhpjjkhgflkihc?hl=en
 203

https://bokonads.com/preventing-bad-behav-
103 103
ior-in-header-bidding-supply-path-optimization/

https://iabeurope.eu/wp-content/uploads/2020/09/
104 104
IAB-Europe-Guide-to-SPO-Sept-2020-2.pdf

https://www.isba.org.uk/system/files?file=media/
105 documents/2020-12/executive-summary-program- 105
matic-supply-chain-transparency-study.pdf

https://www.emarketer.com/content/us-program-
106 106
matic-fees-2019

https://www.emarketer.com/content/why-tech-firms-
107 obtain-most-of-the-money-in-programmatic-pur- 107
chases

https://pubmatic.com/blog/ad-tech-tax-or-a-reality-
108 108
of-doing-business/

https://iabtechlab.com/press-releases/
109 iab-tech-lab-launches-assault-on-illicit-advertis- 109
ing-inventory/

https://www.emarketer.com/content/ads-txt-adop-
110 110
tion-continues-its-steady-growth

111 https://iabtechlab.com/ads-txt-about 111

112 https://www.tagtoday.net/ 112

113 https://www.nytimes.com/ads.txt 113

204

https://iabtechlab.com/press-releases/iab-tech-lab-
announces-two-new-technologies-to-build-more-
114 114
transparency-trust-in-the-programmatic-supply-
chain/

https://iabtechlab.com/press-releases/openrtb-3-0-
115 115
beta/

116 https://sellers.guide/ 116

117 https://adstxt.guru/ 117

https://chrome.google.com/webstore/detail/ad-
118 118
stxt-checker/bjddoinjnabgbnloplmjdfdcdlefdbhj

https://iabtechlab.com/wp-content/uploads/2021/03/
119 119
buyers-json-1-0.pdf

https://www.thinkwithgoogle.com/marketing-strate-
120 gies/app-and-mobile/mobile-page-speed-new-indus- 120
try-benchmarks/

http://glinden.blogspot.com/2006/11/marissa-may-
121 121
er-at-web-20.html

122 https://backlinko.com/page-speed-stats 122

https://speedy.site/average-page-load-times-web-
123 123
sites-2018/

https://www.cnbc.com/2019/01/24/smartphones-
124 72percent-of-people-will-use-only-mobile-for-inter- 124
net-by-2025.html
 205

https://developers.google.com/search/blog/2018/01/
125 125
using-page-speed-in-mobile-search

https://www.creativebloq.com/features/how-the-
126 126
bbc-builds-websites-that-scale

https://www.gigaspaces.com/blog/amazon-found-ev-
127 127
ery-100ms-of-latency-cost-them-1-in-sales

128 https://www.youtube.com/watch?v=UuePkp6TuhQ 128

https://www.admonsters.com/playbook/admon-
129 129
sters-playbook-header-evolution/

130 https://en.learniv.com/have/ 130

131 https://amp.dev/ 131

132 https://web.dev/vitals/ 132

https://chrome.google.com/webstore/detail/head-
133 erbid-expert/cgfkddgbnfplidghapbbnngaogeld- 133
mop?hl=en

https://www.iabuk.com/news-article/quick-qa-view-
134 134
ability

https://support.google.com/google-ads/an-
135 135
swer/7029393?hl=en

https://www.comscore.com/Insights/Blog/Viewabil-
136 ity-Benchmarks-Show-Many-Ads-Are-Not-In-View- 136
but-Rates-Vary-by-Publisher
206

137 https://go.integralads.com/us-mqr-2020-h1.html 137

http://think.storage.googleapis.com/docs/5-fac-
138 138
tors-of-viewability_infographics.pdf

https://www.thedrum.com/news/2014/12/03/56-
139 online-display-ads-are-not-seen-consumers-claims- 139
google-first-ad-viewability

https://martech.org/googles-report-56-percent-ads-
140 140
arent-seen-isnt-shocking-heres/

https://support.google.com/google-ads/an-
141 141
swer/3499086?hl=en

https://support.google.com/admanager/an-
142 142
swer/7246067

https://www.emarketer.com/Article/What-Display-
143 143
Ad-Sizes-Drive-Best-Viewability-Rates/1011670

https://developers.google.com/search/docs/ad-
144 144
vanced/javascript/lazy-loading

https://developers.google.com/publisher-tag/sam-
145 145
ples/lazy-loading

146 https://web.dev/browser-level-image-lazy-loading/ 146

https://www.gulshankumar.net/setup-lazy-loading-
147 147
google-adsense-ad-units/

https://researchasahobby.com/display-images-only-
148 148
when-need-lazy-load/