McKinsey on Payments

Special Edition on
Advanced Analytics in Banking
4 13 20

The analytics-enabled How machine learning can Combating payments fraud

collections model improve pricing and enhancing customer
performance experience

28 36 44

Using data to unlock the Hidden figures: The quiet Using analytics to increase
potential of an SME and discipline of managing satisfaction, efficiency, and
mid-corporate franchise people using data revenue in customer service

52 60 68

Designing a data Building an effective “All in the mind”: Harnessing

transformation that delivers analytics organization psychology and analytics to
value right from the start counter bias and reduce risk

80 86

Mapping AI techniques to Data sheet: Advanced

problem types analytics

Volume 11 Number 28 August 2018

 Building an effective analytics
organization

As companies recognize the predictive power of advanced analytics,

many are hoping to use AA to drive their business decisions and
strategies. While most companies understand the importance of analytics
and have adopted common best practices, fewer than 20 percent,
according to a recent McKinsey survey, have maximized the potential and
achieved AA at scale.

Gloria Macias-Lizaso In working with a wide range of organiza- pertise, beyond specific sector expertise, will
tions, McKinsey has seen many companies become more and more relevant.
start their analytics journey eagerly, but
With this in mind, McKinsey conducted an
without a clear strategy. As a result, their
extensive, primary research survey of over
efforts often end up as small pilots that fail
1,000 organizations across industries and
to scale or have significant impact. Some
geographies to understand how organizations
of these pilots have been mere exercises in
convert AA insights into impact, and how
“intellectual curiosity” rather than a serious
companies have been able to scale analytics
effort to change the business. Consequently,
across their enterprise (see sidebar on page
they are not designed with an end-to-end
61). In this article, we will discuss how to
approach that incorporates the necessary
design, implement, and develop the right or-
conditions for implementation. Instead, the
ganization and talent for an AA transforma-
pilots are carried out in small labs with lim-
tion. An AA transformation usually requires
ited connection to the business, and fail to
new skills, new roles, and new organizational
provide the answers the business needs to
structures.
move forward. Even if a pilot does answer the
right questions, it may not address the cul- Building an AA-driven organization
tural aspects that would, for example, make a Top-performing organizations in AA are en-
sales representative trust a model more than abled by deep functional expertise, strategic
her own experience. partnerships, and a clear center of gravity for
organizing analytics talent. These companies’
These companies quickly become frustrated
organizations usually include an ecosystem
when they see their efforts falling short while
of partners that enables access to data and
more analytically driven companies are le-
technology and fosters the co-development of
veraging their data. Democratization of data
analytics capabilities, as well as the breadth
is blurring sector boundaries; businesses will
and depth of talent required for a robust pro-
increasingly find themselves disrupted not
gram of AA.
by the company they have been monitoring
for the last several years, but by a newcomer For a company aspiring to an AA transforma-
from another industry. Being the best in an tion, these elements can be incorporated into
industry is no longer enough; now companies any of several organizational models, each of
must aspire to be at least at par across indus- which is effective as long as there is clear gov-
tries to compete effectively. Functional ex- ernance, and the company encourages an an-

Building an effective analytics organization 1

 McKinsey’s Insights to Outcome Survey
In the fall of 2017, McKinsey performed quantitative research (using a survey-based approach) of approximately
1,000 organizations across industries and geographies. The survey contained 36 questions, most of which measured
respondents’ degree of agreement or asked respondents to choose their top three responses. The 1,000 responses
encompassed more than 60 responses per geography and over 50 responses per industry, which ensured statistical
relevance in various cuts of the data. The responding companies represent more than $1 billion in revenues.
The survey targeted analytics leaders and C-level executives with a broad perspective on their organization’s analyt-
ics capabilities across the enterprise. These respondents included 530 individuals in analytics roles and 470 in busi-
ness roles.
The industries covered by the survey included: A&D, automotive, banking, insurance, energy (including oil and gas),
resources (including mining and utilities), telecom, high tech, consumer, retail, healthcare, pharmaceuticals, transpor-
tation, and travel. The geographies covered included: US, UK, France, Germany, Spain, Brazil, India, Australia, New
Zealand, Singapore, China, Japan, and the Nordics.

alytical culture across business units to learn work effectively, as long as governance is
and develop together. Answering a few key established to prevent the various units
questions can help to identify the best model. from becoming islands. The proposed or-
ganization depends somewhat on how ad-
1. Centralized, decentralized, or a hybrid:
vanced the company and the business units
First, the company should decide whether
are in their use of analytics.
to create one centralized AA organization,
in which AA stands alone in a center of ex- It is important to note that any organiza-
cellence (COE) that supports the various tion will change over time as the AA trans-
business units; a decentralized organiza- formation evolves. Some companies start
tion, in which analytics is embedded in out decentralized and eventually move AA
individual businesses; or a hybrid, which into a centralized function, while others
combines a centralized analytics unit with that are centralized later move into a hy-
embedded analytics areas in some units. brid model of hubs and spokes. Top-per-
forming companies prepare for these
Our benchmark of several organizations
eventual changes.
indicates that any of these models can

One organizational example

A large financial and industrial conglomerate created a separate COE that reports directly to the CEO and supports
the organization with AA expertise, AA resources (on “loan”), use case delivery, infrastructure to execute use cases,
and technical interviewing. The center also manages data partnerships, develops new businesses by designing and
deploying cross-company and ecosystem use cases on the company’s own infrastructure, facilitates aggregated AA
impact calculation, reports progress to the executive committee, and executes the data committee’s mandates. The
center started out as a small cost center but aspires to transform into a self-standing profit center within two years.

2 McKinsey on Payments August 2018

Two successful strategies
One industry conglomerate addressed this scale requirement by starting with a centralized COE serving all business
units. As the use and understanding of analytics grew across the organization’s companies, they demanded more
support, and the COE was split into sub-groups that were fully dedicated to the largest companies. Over time, owner-
ship of these groups was transferred to the “client” company—but not until they had built a sense of community and
common methodology across the entire conglomerate. This sense of community was further reinforced by requiring
all new recruits to spend six months at the COE and to go through specific AA training and networking events. Since
fragmentation of the analytical talent across functions is almost inevitable over time, it is critical to start out with the
appropriate processes and mechanisms to ensure consistency and community across these new profiles.
A leading pharmaceutical company developed an integrated talent strategy that merged business and analytics func-
tions. The company recruited technology and analytics executives in key management roles and developed analytics
career paths for them. Placing analytics professionals in key business roles enabled the company to identify and op-
erationalize new analytics opportunities before their competitors could. The organization successfully embedded ana-
lytics in key elements of the business—for example, analytics on clinical trial data to enable more cost-effective data.

The choice between centralization and tegic partnership with an external vendor
decentralization is not an all-or-noth- be the best approach? AA will effectively
ing decision but should be decided per become the “brain” of the organization,
sub-function. Data governance, how- so companies should be careful not to
ever, should be centralized, even if data outsource too much. Top-performing com-
ownership is not. For data architecture, panies often keep analytics that provide a
top-performing companies often have competitive advantage—such as pricing an-
data centralized within business units. alytics—within the organization. A central,
This data typically includes data from internal unit can oversee all AA outsourc-
marketing, sales, operations, and so on. ing, and partnerships can be established
Most top-performing companies centralize for specific AA solutions or to bring in
partnership management; otherwise, com- particular assets, such as unique sources of
peting or redundant partnerships could data or advanced solutions.
inadvertently be set up in various parts of
3. Locating the AA unit: Yet another import-
the organization, and intellectual property
ant decision is where to locate the AA unit.
could be at risk.
AA is most effective when it is cross-func-
2. To outsource or not to outsource: An- tional, accessible enterprise-wide, and inte-
other decision is whether AA talent should grated with the business. Various levels and
be partially outsourced, and if so, how. functions can host it, but the final location
Should outsourcing be limited to low-level should have enough visibility and access to
data analytics activities? Or should the the C-suite to break through inertia and en-
company establish several tactical partner- able transformation. It is helpful if the unit
ships for selected tasks? Or would a stra- has an enterprise-wide view, given its trans-

Building an effective analytics organization 3

 formational potential for all functions. projects. AA projects can end up last on
their list of priorities. Including AA within
The AA unit is often most effective when
marketing or operations, meanwhile, can
it is a sub-unit of business intelligence—as
limit its potential to transform the remain-
long as this area has an enterprise-wide
ing parts of the organization.
perspective—or of strategy or digital.
Some companies locate their AA units in Staffing the AA center of excellence
IT, but this arrangement can be challeng- Sixty percent of top-performing companies
ing. IT staff—who are used to managing in AA have a “center of gravity” for their an-
longer-term projects that are often dis- alytics efforts, according to our survey. They
connected from the business—may not be typically include a specific set of roles, skills,
prepared to manage short-term, agile AA

Exhibit 1

An organizational blueprint of the advanced analytics COE.

Advanced Analytics COE
Head of Advanced
Analytics COE

COE support office (recruiting,

talent management, etc.)

Demand Analytics Analytics Analytics Data Training

management development support platform governance

HR Program Deployment DWH/DM Metadata

managers managers specialist design

Oil and gas Data Operations Data Data

scientists support architects compliance
manager

... Solution Data Data Data

architects quality operations governance
managers manager

Business Help desk

analysts

Developers

Data
acquisition
managers

Source: McKinsey analysis

4 McKinsey on Payments August 2018

Lessons learned from COE failures
Companies that have rolled out full-scale COEs during an AA transformation have encountered some pitfalls. Some of
the most common include:
1. A failure to focus on business value. Successful roll-out of a COE requires a clear vision as to where the COE’s
biggest business impact can be captured. The COE should be linked to the organization’s overall business strategy
and should commit to achieving a measurable impact during a defined time frame, with its goals clearly prioritized.

2. Over- and under-thinking technology. Some companies attempt to replicate an entire data history when building
their AA COE, resulting in data “scope creep.” It is vital to sufficiently think through data integrity/architecture; failing
to do so may result in missing data and missing data connections. Some companies try to do too much at once by
replacing their hardware, software, and analytics stack simultaneously rather than tackling one at a time. These com-
panies may buy the “best of breed” in each category but then find that none of them “talks” to each other. Instead,
companies should build systems and functionality as needed—especially since technologies tend to become obso-
lete within just a couple of years. New innovations can be integrated later if the system is built gradually.

3. Taking more than 18 months to deliver value. The COE’s benefits should begin to come online well before the
entire roll-out is complete. If the COE does not deliver benefits sooner, it is often because it depends too heavily on
insights-delivery FTEs instead of automation. The delivery of insights should be staged to capture value sooner.

4. Insufficient skill-building and change management. Top management and the internal team must be 100 percent
committed to the COE if it is to succeed. Internal stakeholders must be engaged in development or accountable for
delivery. The effort cannot focus exclusively on technology instead of the process and people; in particular, the or-
ganization must build the requisite front-line skills needed for an effective AA COE. Companies should hew closely to
the business case and avoid the functional scope creep that can occur when mid-flight changes not included in the
business case suddenly become priorities.

5. Operating analytics as an island. One large US insurance company interviewed by McKinsey hired a sizeable num-
ber of data scientists and launched more than 50 pilot projects to test its new capabilities. Despite a real commitment
and considerable investment, the analytics team was isolated from the rest of the company, with no connection to
the overall business strategy—a critical mistake. Not surprisingly, this company’s ad hoc analytics projects had no
real impact.

At the other end of the spectrum, successful AA-driven companies are building centralized AA capabilities and then
creating end-to-end agile teams (“use case factories”) that integrate profiles from IT, sales, marketing, finance, and
other functions. This approach ensures that use cases are immediately integrated into business processes and thus
create value.

and capabilities within the COE (Exhibit 1), the COE and business units. The translators
including data scientists (“quants”), data usually have a combination of business, an-
engineers, workflow integrators, data archi- alytics, and technology skills and are found
tects, delivery managers, visualization ana- in the business partner role in data analytics
lysts, and, most critically, translators from leadership.
the business who act as a bridge between

Building an effective analytics organization 5

 Many COE roles are filled with highly spe- mine how those models will be put into pro-
cialized analytical resources recruited from duction; and learn from the results.
advanced degree programs in computer
The COE can be built in about 18 months, typ-
science or math. But these individuals must
ically in incremental steps. It may start with
also be able to translate sophisticated models
five to ten data professionals, including data
into simple, visual decision support tools for
engineers, data scientists, and translators. In
front-line employees.
its end-state, it likely will require significantly
They also need to have a collaborative mind- more. The number of translators needed will
set, given the interdependencies among data, vary by business unit but is generally about 10
systems, and models. With translators bridg- percent of business unit staff. Most companies
ing any communication gaps, team members source their translators from “client” business
from analytics and the business work to- units and then train them, since these employ-
gether in two- to three-month agile “sprints” ees will have deep knowledge of the processes
as they identify problems; find out whether that AA is trying to optimize. These individuals
relevant data exists and, if not, whether that are usually analytical, critical thinkers who are
data can be acquired; test their models; deter- well respected in the company.

Exhibit 2

Beyond the COE, employees at all levels need to be trained.

Senior leaders Managers
Business leaders of a Business users (i.e. mid-level
transformation or a use case managers) of the outputs of analytics
implementation models and insights

Board

Top team

Analytics COE BU 1 BU 2

Data Data
Design
science engineering

Analytics specialists Analytics translators

Design and build analytics models Collaborate with the business to
Iterate/refine models based on understand the need, and can guide
business and user feedback analytics design to meet this need

Source: McKinsey analysis

6 McKinsey on Payments August 2018

To illustrate how the various key skills and roles come together in the COE, here is an example description of these
roles’ working together to fulfill a business request:
• The translator and business owner identify and prioritize the business request.

• The data scientist works with the translator to develop an analytics use case, including an algorithm and analyses to test.

• A data engineer from the COE works with the relevant business division to understand the data requirements of the
use case and to identify data sources.

• The data engineer works with IT/the business to ensure data availability, identify gaps, and develop ETL (extract, trans-
form, load) to load data into analytics sandbox.

• A data scientist programs the algorithm and analyzes the data in the sandbox to generate insights.

• A visualization analyst develops reports and dashboards for business users.

• A COE workflow integrator works with the business owner to develop a prototype for models and tools.

• The COE ensures that key business and IT stakeholders test the prototype tools and solutions.

• A delivery manager pilots the prototype and dashboard and works to obtain a go/no-go decision.

• The delivery manager and COE workflow integrator work with IT to scale the prototype to the enterprise level.

• The COE delivery team and translator work with the business and IT to ensure adoption and ongoing model maintenance.

In this process, feedback would be gathered between steps nine and ten.

While the COE and some of its roles may In McKinsey’s survey, 58 percent of respon-
emerge gradually, it is best to have the data, dents at top-performing companies say that
platform, and career paths needed for an AA their organization has deep functional ex-
transformation in place from the beginning. pertise across data science, data engineering,
If the platform is still under development, data architecture, and analytics transforma-
adding more people may only make that de- tion. Top-performing organizations have four
velopment more complicated. And without times as many analytics professionals and
a clear career path, attracting this scarce one and a half times more functional experts
talent will be difficult. As much as possible, than other companies.
roles should be clearly delineated to prevent
These companies also retain three times
squandering valuable talent on functions for
more talent—primarily by creating strong ca-
which they are over-qualified, which can un-
reer development opportunities. People with
dermine retention.
superior analytics talent usually have many
Career development and strategic potential opportunities and thus need to
partnerships see a clear career path and opportunities for
Gaining an edge in analytics requires attract- growth within a company if they are to join or
ing, retaining, and sourcing the right talent. stay with it. Several career tracks should be

Building an effective analytics organization 7

 available, as some analytics staff may wish to Beyond the COE: training employees for
pursue a more technical profile, others may cultural change
move into translator or integrator roles with As detailed in “Hidden figures: The quiet
the business, and some will likely move into discipline of managing people using data,”
managerial positions. on page 36, an AA transformation requires a
profound cultural change, as the entire or-
In all cases, these individuals tend to stay
ganization must change the way it operates.
motivated if they are learning on the job and
Employees need to learn to trust in AA, to
from one another. Achieving this goal requires
understand what they can ask of it, and to
a minimum scale for each analytics group.
know that AA can answer far more complex
Having only one or two data scientists in each
questions than traditional analytics ever
function will not help them learn, and they may
could. Outside of the COE, then, employees
have difficulty making themselves understood.
at all levels—senior leaders, managers, ana-
To fill any gaps in talent, 62 percent of survey lytics specialists, and analytics translators—
respondents at top-performing companies need to be trained to be AA-proficient and to
say that they strategically partner with others drive the transformation forward (Exhibit 2,
to gain access to skill, capacity, and inno- page 65).
vation. For example, a large, multinational
A sweeping—but feasible—transformation
retailer developed a strategic partnership
Transforming a company to be AA-driven is
with a start-up incubator that focuses on
a monumental task that should not be un-
identifying cutting-edge technologies—such
dertaken in one fell swoop, but instead incre-
as drones—to transform the retail industry.
mentally, based on use cases. Since AA can
The retailer found that employing a mix of
and will transform a company, the effort to
in-house talent and smart, strategic partner-
cultivate an AA-driven organization is most
ships with other organizations enabled it to
effective when it comes from the top, from
get the best out of both, thus affording access
senior executives. If a company focuses on
to skills, capacity, and innovation on a much
the value of advanced analytics and builds AA
larger scale. Through the incubator, the re-
capabilities as needed—while still having the
tailer formed partnerships with start-ups and
data, platform, and talent strategy in place
venture capital investors. The company also
from the beginning—its AA transformation
created a compelling value proposition for
will succeed.
attracting top analytics talent.

Gloria Macias-Lizaso Miranda is a partner in McKinsey’s Madrid office.

8 McKinsey on Payments August 2018