Tracking Growth. A case study.

risktrainer.medium.com/tracking-growth-a-case-study-ea81760368be

Jawwad Ahmed Farid July 1, 2023

Modeling ecommerce GMV growth in Pakistan for ‘23.

Pakistan Ecommerce GMV by quarter paid via digital payment channels. SBP Payment Systems
report.

a) Projected ’23 full year estimate for SBP ecommerce spend? USD 654 million to USD
723 million. Divide by your assumption of CoD market share compared to digital payment
share to find the total ecommerce market size.

b) Two models. One, average GMV per merchant x registered merchants. Two, average
consumer spend x number of unique consumers.

c) Conventional analysis suggests a down year. But when we add growth in mobile
banking users and mobile banking spend, there is a strong case for ’23 being a growth
year.

d) Supported by surprising growth registered in ’22 in $ terms despite the exchange rate
shock, import restriction, economic downturn and collapse of investor funding.

e) Average ecommerce consumer spend per year US$ 440. Average order basket up to
$22 per order. 3–5 orders per quarter. 4 quarters per year. 1.3 to 2 million unique
ecommerce consumers. 5–7 million unique mobile banking consumers. The average
consumer spends high if we just think of food delivery as ecommerce. But consider the
size of the pool and the share of consumer electronics, household appliances and fashion
in the ecommerce basket.

f) Data source and modeling using SBP Payment systems report. With some changes,
SBP payment systems report can be a powerful validated source of real transaction
based assessment of living standards compared to the PBS PSLM surveys. The PSLM
survey is the primary source of household income for many researchers but has multiple
challenges and known data quality issues. Engaging with SBP to see if we can make that
happen sooner.

Context and background.

This has been a hard year in numerical terms for ecommerce in Pakistan. Over 18
months the segment has been hit by multiple shocks.

a) A 60% depreciation in purchasing power of the rupee reducing consumer demand and
average spend. Over a 10-year period from 2013 to 2023 the rupee tied for the 6th worst
currency in terms of loss of purchasing power with Nigerian Naira and Russian Rouble. If

1/16
it is any consolation we managed to out perform Egypt, Ukraine, Ghana, Turkish and
Argentina local currencies.

b) A severe economic recession with GDP growth at less than half a percentage point
(0.39%).

c) Sweeping import restrictions reduced supply through formal channels and local
inventory levels of imported consumer goods.

d) Collapse of investor interest in the segment and consequent premature failures of the
biggest names in this space. Resulting in significant reduction in marketing spend,
customer mindshare, brand awareness, user trust and sales conversions.

e) Absence of visibility and continued uncertainty on economic, political, judicial and

financial outlook for the country. One out of any of these four is bad enough. Having all
four crisis in tandem isn’t really a recipe for building investor, founder or business
confidence.

USD:PKR. Cumulative annual depreciation. 2018–2023.

The paradox. Building a growth outlook for 2023?

Despite these challenges on a year-on-year basis, overall ecommerce growth was
positive. This presents an interesting paradox. Rather than shrinking, ecommerce GMV
as measured by SBP figures has grown in PKR (nominal) and USD (real) terms. An
unexpected display of resilience on part of the local market. Given the negative overhang
created by PKR depreciation. Hence the need to build a more nuanced growth projection
model than one based on superficial analysis.

Two questions that we want our analysis to answer.

a) How did the market manage to grow in USD terms?

b) What are the drivers of that growth?

2/16
c) What is the likely comparable year end ’23 figure for digital ecommerce spend reported
via SBP payment systems report.

Given the uncertainty in previous six quarters and the outlook for both local and global
economy how should we go about building an outlook for 2023? Should we expect a
continued decline or a turnaround? What would the timing for that turnaround? Would
total ecommerce sales as represented by Gross Merchandise Value (GMV) in ‘23 be
above or below ‘22 levels?

Growth model design.

One decision was to measure and report all figure in USD terms. The objective of our
analysis was to focus on real growth. While USD growth in not a perfect proxy for real
growth it adjusts for imported inflation. While a business based in Pakistan has no need
to report numbers in USD terms, if its primary input (imports) are priced in dollar terms,
and its valuation is linked to rounds priced in dollar terms, it makes sense to quote
growth, revenues and valuation linked metrics in dollar terms.

Beyond a baseline for measuring performance and growth there are other design
considerations for our growth model. In the ecommerce world, growth can be driven by
multiple parameters. Consumers, merchants, average spend, average sales per
merchant. Growth or decline in any of these factors can impact top line figures. While a
high level market growth model can serve well, a parameter driven model is robust and
resilient to changes in external factors. It allows us to examine a market from different
lenses and decide which one of these views is likely to dominate growth in future years.

Higher level models may only look at total projected ecommerce sales, ecommerce sales
as a percentage of retail sales, retail sales as a percentage of GDP. While there is nothing
wrong with this view it is difficult to reconcile it with more important secular trends.

Why bother with multiple methods and tools? Why not stick with one? Shouldn’t each of
these methods reconcile to one approach or design rather than multiple results. How wide
is the range likely to be?

From a modeling perspective we are only modeling State Bank of Pakistan (SBP)
ecommerce GMV estimates. This is the figure which is booked and paid via digital
payment channels for registered ecommerce vendors. We have a documented history of
these figures as they are first recorded via banking channels and then reported to SBP on
an aggregate basis. The figure excludes Cash on Delivery (CoD) transactions which
represent a much larger share of market. We can easily apply a share of CoD assumption
and convert SBP ecommerce figures into total ecommerce GMV estimates.

Ecommerce GMV. Quarterly figures. Q1, 2023.

Before we answer these questions, lets take a look at where we are as per the latest SBP
payment system report and how does that compares with the historical trend for these
figures. We also want to see if we can form any impressions of what the historical trend

3/16
suggests about future outlook. The most recent release of SBP’s payment system report
shares data up till 31 March ’23.

Pakistan. Ecommerce GMV via digital payment channels. Avg. digital merchant GMV. Registered
digital ecommerce merchants.

The chart above shows 3 data plots. The blue bar chart at the bottom of the plot is
number of ecommerce merchants. The two plots in orange and grey represent gross
merchandise value (GMV) per quarter and the average GMV by merchant by quarter.

Both line plots are quoted in USD terms. The index for the orange plot is on the right
(LHS). The index for the grey plot is on the left (LHS).

The blue bars show ecommerce merchant registrations with payment systems providers.
Registered merchants have grown from 2,523 merchants at the end of Q1, 2021 to 6,562
merchants at the end of Q1, 2023. 160% growth in 2 years. Registrations are still growing
at 10% a quarter in 2023, (47% YoY). Registrations are likely to continue growing at this
pace as more stores, storefronts and payment systems solutions come online.

The downside of this growth is visible in the second line plot.

The grey line shows the average quarterly sales per merchant. In 2017 that number was
over US$ 70,000 per quarter. In Q1, 2023, it is hovering around US$ 21,300 per quarter.
This translates into average annual sales receipts for ecommerce stores in 2017 at
$280,000. In 2023, that number is down to US$ 85,000.

The third plot, the orange line, shows the official SBP reported ecommerce sales
proceeds received through digital channels. While this number has been steady around
US$ 150 million per quarter for the last five quarters, in 2023, Q1, it dipped to US$ 140
million.

Should we worry about this dip? I would say no. Why? One word, seasonality.

4/16
Pre-Covid-19, in 2018, 2019 and 2020, sales always dipped in the first quarter, compared
to the final quarter of the prior year. This trend broke in 2021 and 2022 but we have gone
back to it in 2023. See the plot below.

The thick blue line at the bottom is QoQ growth for 1st quarter of the new year, compared
to last quarter of the prior year. It is negative in 5 of the 6 years. The only exception was
2021.

Compared to the double digit slide for Q1 (new year) compared to Q4 (prior year) in ‘19
and ‘20, ‘23 looks harmless with a mildly negative 8.5%. Remember this is in USD terms.
In PKR terms Q1 ‘23 over Q4 ‘22 is still positive.

Historically QoQ growth in the three remaining quarters of the year tends to be positive.
Between 2017 and 2023, it was only negative in two of the 15 quarters in our dataset.

Seasonality. QoQ ecommerce GMV growth. 2017–2023.

Model One. Merchants and average GMV.

Our first model uses projected merchants and average GMV by merchant per quarter.
Take the number of merchants, multiply by average store sales to get the quarterly figure.
Project both to get your projected figures.

6,562 merchants as at end of March ‘23. US$ 21,370 in average quarterly GMV. US$ 140
million in quarter GMV.

If we project merchant growth at a steady rate of 10% per quarter for the next 3 quarters,
we end up with 8,570 merchants end of Dec ’23. Assuming average GMV per merchant
dips to US$ 20,000 per merchant.

There would be a number of adjustments that would need to be made to this model and
estimate but for our first pass we will take these numbers as is. Our estimate:

5/16
US$ 654 million = 8,570 x 20,000 = Crude best case estimate for ecommerce GMV
reported by SBP at the calendar end of ’23.

This compares with US$ 609 million figure as at year end ’22. Representing 7.5% YoY
growth in USD terms over ’22.

How do these projections fit in with existing models? The plot below shows quarterly
historical trend and two forward looking projection models using a 5 period moving
average and a polynomial curve fit on the trend line. Both show quarterly GMV flattening
out in the next 3 quarters. Do we have any evidence to the contrary that supports a 7%
growth outlook for ‘23.

Growth projection models. Ecommerce GMV by quarter.

An alternate model. Mobile banking application users.

To answer this question we have to explore an alternate model. A model that relies on
SBP data but uses a different head from that data set. We use mobile banking application
users and average mobile banking user transactions in US$ terms.

Why? Mobile banking application users need a smartphone to download and run a mobile
banking application. They need access to 4G / 3G internet connectivity. Some educational
exposure to use and run the application. Sufficient balance in their account to value the
convenience factor of mobile banking applications over IBFT transfer fees or a physical
visit to their nearest bank branch or ATM. Comfort and trust with technology, banking
application, network infrastructure and security to enable and use their phone for financial
transactions.

Mobile banking users and their quarterly transaction volume serve as a proxy for how
large the underlying market is. Not all of these transactions are ecommerce transactions.
The largest and most popular is utility bill payments. Still the pool represents an upper

6/16
cap on potential users and a part of their transaction volume, available discretionary
spending power that may one day find its way to online ecommerce transactions.

Modeling tweaks. Unique users and their transaction volumes.

SBP data reports mobile banking users and transaction volume in aggregate. For us to
use this data effectively we first have to translate it to unique users and transaction
volumes for unique users. We don’t have a lot of data points that could help us do this
accurately and in a scientific fashion. But we do have anecdotal evidence about the
number of accounts a typical urban user has and as a result the number of mobile
banking applications they use. We start off with assuming 2 active bank accounts, 2 credit
cards and an average of 2.2 mobile banking applications per urban user. Please note
that this is a working assumption.

We can also apply a similar analysis on number of ecommerce transaction by a unique

user. While the bank account assumption can be static, the ecommerce transaction by
unique user will likely change on a quarter by quarter basis. Making this change to the
data set gives us a sense of unique ecommerce users and their quarterly spend.

Our final step would be to compare the number of unique ecommerce users with mobile
banking application users and compare the average spend for both users on the two
channels.

The difference between the two figures should give us a sense of room to grow within the
ecommerce industry and should serve as a possible cap or check on future growth
projections.

Here are the results from the adjustments we have proposed. The chart below shows
three trends. The orange plot shows our estimate for unique mobile banking user by
quarter. The blue line our estimate for unique ecommerce users. The green bars the
difference between the two metrics.

Unique mobile banking users versus unique ecommerce users. Opportunity or disconnect?

7/16
The number of unique ecommerce users hasn’t grown at the same clip as mobile banking
application users. The difference between the two figure clearly shows that there is
significant room to grow for ecommerce market if we can figure out answers to trust,
fulfillment and payment processing. Using our working assumptions the model shows that
there are roughly 4 to 5 million users who use mobile banking applications but are not
ecommerce customers in Pakistan.

What is the average spend of these missing users? Using our working assumptions, US$
22 average order value, 5 orders per quarter, US$ 110 per quarter or US$ 440 per
year in transaction volume. Multiply that by 4 million customers and we end up with
US$ 1.76 billion per year.

That is the money we are leaving on the table every year.

How did we get to this figure? The chart below shows the average transaction size for an
ecommerce order (green bars) and multiplies that by the number of unique orders placed
by a unique customer in a given year. The number of unique orders is a working
assumption in our model and is modeled to change every year as customer profiles and
their trust with ecommerce vendor changes.

Average spend. Unique ecommerce customers.

Solving these three challenges (trust, logistics and payments) is worth 3 times the
current SBP ecommerce estimates. We are making progress in all three areas. Integrated
online storefronts are making it easier to bring vendors online. Electronic money, payment
solutions providers and digital banking licensee are speeding up the onboarding process.
Last mile logistics have seen a whirlwind of activity in the last 4 years, originally driven on
by Covid-19 and then by the wave of startups focused on ecommerce.

Read all of the above with the mobile banking user and spend curves below and you can
clearly see that there is room to grow in this market.

8/16
 Evidence A. Future growth outlook.

The average ecommerce spend by a unique user was $US 110 per quarter in March ’23.
The average mobile banking unique user’s spend was US$ 1,702 per quarter at the same
point. The difference between the two figures is US$ 1,592.

The mobile device on which these two sets of transactions happen is the
same. The user is the same. Only the channel is different. Our thesis is that
over the next two years these two channels and their customers will merge
and when that happens the two estimates are likely to converge. These are
the stakes everyone should be playing for.

Money on the table?

9/16
A follow on point is growth trajectory of mobile banking users and their average spend per
quarter. Similar to our projections above we use two separate projections models to
anticipate the trend. The increase in dollar value of these figures indicates that the use
case for mobile banking payments is moving beyond utility bill payments. As more
vendors, schools, electronic stores, pharmacies, retail outlets and service providers offer
digital payments as an option, more customers are likely to use them.

Growth models. Revisited.

Bringing it together. Model Two projections.

Model two builds up on the trends we have detailed above using the mobile banking and
ecommerce unique user dataset.

a) We first grow and project average ecommerce user basket at the rate it has been
growing for the last few quarters.

b) Then apply that figure to estimate average quarterly spend by a unique ecommerce
user.

c) Then multiply the average quarterly spend by the number of unique ecommerce users
as at 31 March 2023.

The table and chart below presents the results of this framework applied to quarterly
ecommerce projections.

10/16
 Models at work.

The Q1 ’23 figure is based on actual figures. Q2–Q4 are projected results. Our model 1
estimate is US$ 654 million. Our model 2 estimate is US$ 723 million. The difference
between the two is due to differences in methodology.

Models at work. Illustrated edition.

Looking at the SBP trend in isolation one would feel the safe choice would be to project
’23 as a down year for ecommerce. Once we add the mobile banking story to the context
and the underlying trend in average spend during the last few quarters, the growth year
arguments become more plausible. The truth and realized future is likely to be
somewhere in between the two positions and estimates.

Taking the models forward. Impact factors.

What do we expect to change in ‘23?

a) More buyers and customers will be using digital payments. A higher proportion of
payments will come through digital channels in ‘23. Why? Take a look at the mobile user
growth and averages spend chart above.

11/16
Net impact assessment? Positive.

b) This will create a natural upward growth pressure on payments systems figures. Part of
the growth in SBP payment systems figures may simply reflect conversion of CoD
customers to payment channels. This implies that while the payment systems number will
continue to grow, it may or may reflect real growth in the underlying ecommerce market.
There is no simple way to adjust for this but we should anticipate the impact.

Net impact assessment? Neutral to negative on overall market size. Strongly positive
for SBP ecommerce spend estimate.

c) Seasonality. Q1 is down compared to Q4. It should be. Give economic conditions Q2

for ’23 number is also likely to disappoint. But we should see an uptick in both Q3 and Q4
figures for ’23.

Net impact assessment? Neutral.

d) USD:PKR FX rate change impact. USD:PKR exchange rate is likely to depreciate in

favor of USD. Rupee value of purchases is like to go up. Despite this as we have seen in
’22 net real change in USD terms is still likely to be marginally positive.

Net impact assessment? Negative.

e) Number of merchants under ecommerce market. Number of merchants registered with

EMI and PSP providers has been on an upward tear as new players ramp up onboarding
efforts. While overall value per merchant will come down, number of transactions in the
system and total transaction value should go up.

Net impact assessment? Positive.

f) Average ecommerce consumer spend. This had been trending downwards for a while.
However for both mobile users and ecommerce transactions, despite the hard downward
shift in exchange rate, there is a clear upward tick in both figures. The upward trend is
likely to continue as demand starts to come back after the contraction experienced in ’22
and the first two quarters of ’23.

Net impact assessment? Positive.

g) The rationale and reasoning for this? As number of merchants and consumers were
growing, the average ticket size was falling. As the growth in number of users slows
down, ticket size is likely to trend upwards. This happens because of two reasons. One,
as user trust with ecommerce vendors goes up, their ticket size increases. Two, user trust
also increases the frequency of purchase orders. Collectively both trends have a direct
impact on average spend.

Net impact assessment? Positive.

Notes. Dataset challenges.

12/16
What exactly is an ecommerce sale or transaction? How do we define ecommerce
sales?

There needs to be a standard definition of what is considered an ecommerce sale in the

SBP payment systems report. The current implied and assumed definition is:

a) Sales of stores and payments received by such stores registered with banks and
ecommerce payment gateways / internet payment gateway (IPG) providers.

While our assumption is that this is the case, segregation of additional data in the PSD
report would help if it could also show the following transaction types. One question would
be to clarify if all payments received by registered ecommerce vendors included in the
ecommerce total or just payments received via IPGs.

a) IBFT payments made by consumers that indicate payments to suppliers or merchants

for good and services

The SBP payment systems report already provides a breakdown of IBFT transfers and
transaction types by IBFT transfers through mobile banking applications. A clarifying note
indicating that these figures are not included in the ecommerce total above would also
help.

Motivation for analysis.

Two years ago we introduced the concept of scale multiples in Founder Puzzles. One way
of assessing growth potential of a market is to look at the number of customers that exist,
but haven’t been reached as yet.

We were trying to get a sense of how do unique mobile banking users compare with
unique ecommerce users. Because that would give us an indication of the room to grow
in the local market when it comes to ecommerce.

If we already have 2 million unique ecommerce customers and 2 million credit cards or
bank accounts with the right profile then we are tapped out. But if we have 5–6 million
unique users with mobile banking applications and only 2 million unique ecommerce
users then there is a potential conversion opportunity. Even though ecommerce may be
low margin business, but it is a documented business. It helps us put the local market in
the right light in front of regional investors.

Why not internet banking? Why mobile Banking? Because internet banking data and IVR
data does not distinguish between retail and corporate clients. Internet is a mix of both
corporate and retail clients. Mobile is mostly retail but that is changing slowly as more
corporate clients move to mobile banking applications.

This is not to say that phone, IVR or internet banking channels are not important
channels. Every channel counts and reinforces the motivation to purchase. Just that
mobile banking data was in a form where this analysis could be run to generate data

13/16
driven insights on the local market.

Hence the case to review the possibility of additional metrics, tables and reporting labels
in the SBP payment systems to improve downstream modeling by financial analysts and
investor teams.

Notes. Proposed modification to SBP dataset.

Here is what SBP can do to help make the payment systems report more powerful and
useful. Without compromising on user privacy:

a) Add a breakdown of mobile banking users and transactions value and volume by
nature of account. Retail customers vs institutional and corporate customers. Today we
assume mobile banking users are mostly retail consumers while Internet banking users
include both retail and corporate clients. This is likely to change as banks roll out mobile
banking solutions for their corporate clientele.

b) A distribution of users, transaction value and volume by province or city of residence of

the account on an aggregate basis.

c) A break down of users, transaction value and volume by gender on an aggregate

basis.

d) A break down of users, transaction value and volume by devices, if possible.

Alternatively by the device operating system.

e) A distribution of transaction values by broad rupee buckets for retail consumers and
accounts.

SBP has already an admirable effort in place to start providing data in Excel templates.
Availability of this dataset and its proposed modification in Excel will also go a long way in
making analyst efforts more productive.

Why bother with the changes? The mobile banking application and consumer dataset
is one of the most valuable economic estimation datasets in Pakistan. Researchers today
rely on the PBS’s PSLM dataset which is open to multiple challenges of accuracy,
representation and data validation. By making the above changes, SBP can create an
actual transaction based profile for the digital economy of Pakistan. More curated and
filtered data makes it easier for investors to size the potential of a given market as well as
make smarter investor decisions.

Why should we care about the ecommerce sector?

There is debate within the retail sector if growth in the ecommerce segment comes at the
cost of cannibalizing existing physical retail store sales. The answer to that question is
yes.

14/16
From a logistics, margins and efficiency perspective a dark store serving customers
online should be cheaper than brick and mortar stores. It takes a while for order flow to
reach a stage where unit economics for a given store turn positive. Once that stage is
reached the math begins to work for both consumers and store owners. Historically the
challenge has been time and cost of reaching break even. When large venture funded
start up entice customers with deep discounts and marketing campaign they don’t change
behavior just for their customers, they change behavior for the entire sector. Store owners
as well as customers.

What are the sources of this value?

One, documented transactions using a documented channel that were previously

undocumented. Documented transactions bring retailers, bank accounts and customers
within the tax net and improve sector sizing estimates. Documented sales also reduce
footprint of the cash economy. While the tax benefits are clearly quantified, the biggest
uptick is additional data. Good data drives decisions, market sizing and segmentation
efforts such as this note and helps improve decision making processes.

Two, more efficient fulfillment reducing cost of sale and reallocating margins across the
distribution channel. Dark stores, improved logistics, payment systems efficiencies all
help improve margins. Some of these savings are passed on to consumers. Other
become source of savings and additional capital for store owners.

Three, the most interesting and useful impact, a ready market for liquidation of inventory
at discounted prices of older models, returns and damaged goods. This increases the
depth of the overall segment and market and becomes a source of additional liquidity
adding to market efficiency.

Related readings.
For a more detailed dive into the methodology used in this write up, please see the sizing
market, segment selection and scale multiple notes from earlier this year.

Sizing markets.

Sizing Pakistan’s e-commerce foot print as at year end 2022 (31-Dec-22)

using a mix of methods and data sets. TPD class…

risktrainer.medium.com

The Right Segment?

Four simple tests to pick the right customer segment for your products.

blog.startupstash.com

15/16
The Zen of building financial models.

It takes more than Excel to build a model for the real world.

risktrainer.medium.com

Scale Multiples. 50 minutes on idea selection filters.

I taught the valuation and financial modeling boot camp to founders for 7
years. I have been a serial has been for much…

risktrainer.medium.com

16/16