[REPORT PREVIEW] The AI Maturity Playbook: Five Pillars of Enterprise Success

The AI Maturity
Playbook:
Five Pillars of
Enterprise Success
DECEMBER 11, 2018
BY Susan Etlinger
RESEARCH REPORT
PREVIEW VERSION

1

TABLE OF CONTENTS
2 EXECUTIVE SUMMARY
3 FOUR MACRO TRENDS AFFECTING AI SUCCESS
3 How We Interact: From Screens to Senses
4 How We Decide: From Business Rules to Probabilities
5 How We Innovate: From Data Analytics to Data Science to Data Engineering
6 How We Lead: From Expertise-Driven to Data-Driven
7 A MATURITY MODEL FOR ARTIFICIAL INTELLIGENCE
9 Strategy: From Optimization to Business Model Innovation
10 Data Science: From Specialty to Scale
12 Product and Service Development: From Reactive to Anticipatory
14 Organization and Culture: From Hierarchical to Dynamic
14 Ethics and Governance: From “The Wild West” to “Enterprise-Ready”
18 BUILDING YOUR PLAYBOOK
20 ENDNOTES
21 ABOUT US
22 METHODOLOGY
22 ACKNOWLEDGEMENTS
1

2

Every day we see new stories — and new hype — about the
capabilities and sheer momentum of artificial intelligence. Use
cases abound: AI is as good as experts at detecting eye disease.
It is changing everything from banking to fashion to education.
Investment continues at a healthy clip; research firm IDC expects
cognitive and AI spending to reach $52.2 billion by 2021.1
DARPA
alone expects to invest $2B in AI over the next five years.2
But as we’ve seen with all significant technology shifts — from the
Gutenberg Bible to electricity to the Internet — AI upends many of
the assumptions, processes, and even cultures of the organizations
and societies that implement it. Some of these changes are
temporary, as we learn more and as the technologies mature,
and some are more fundamental and longer lasting, but the clear
message is that we are still in the earliest stages of the shift to
intelligent systems.3
This report lays out a maturity model for AI adoption in the
enterprise. It outlines four macro shifts that define the impact of AI
in organizations and society and four stages of AI maturity based
on how organizations approach business strategy, data science,
product and service design, organization and culture, and ethics
and governance. As with any significant technology shift, we’ll know
AI has reached maturity when it no longer looks like magic but has
become an integral part of our professional and personal lives.
2

3
FOUR MACRO TRENDS
AFFECTING AI SUCCESS
Beyond use cases, investment levels, and the
technology landscape, the natural questions
business leaders tend to ask about AI are:
“Who’s doing it well?” “What does maturity
look like?” and “How do we move from one
stage of maturity to the next?” The answers lie
as much in external as internal factors. From
an external standpoint, meaning factors that
leaders cannot meaningfully influence, there
are four fundamental shifts at play:
1. How we interact: from screens to senses;
2. How we decide: from business rules to
probabilities;
3. How we innovate: from data analytics to
data science to data engineering; and
4. How we lead: from expertise-driven to
data-driven.
1. HOW WE INTERACT:
FROM SCREENS TO SENSES
One of the fundamental shifts that AI enables
in the digital world is the shift from browser-
or app-based interactions to interactions
that make use of our senses — how humans
most naturally relate to the world. Sensory
interaction is significantly different from screen
interaction; it frees people to communicate
in whatever style they prefer but can be more
challenging to design for.4
Today, vision, hearing/language, and touch-
related technologies are commonplace; we
think little of talking to a plastic cylinder on
the dining table or touching screens to make
them respond to our commands, but that
was not always the case. The next frontier are
technologies that aim to digitize our sense
of smell and taste, which are actively being
researched.5
Figure 1, below, describes some of
the technologies in use or being explored today.
Figure 1. From Screen to Sense-Based Interaction Models
VISION
Interpreting
the objects and
attributes of visual
objects and images:
photos, drawings,
video
HEARING
Interpreting
sounds or text and
translating them
into speech, text, or
images, or from one
language to another
TOUCH
Pinch-and-zoom,
gestural interfaces
(such as in AR/VR)
that translate move-
ment into meaning
or commands
SMELL
The ability to
translate smells into
digital chemical
information, and
vice-versa
TASTE
The ability to
translate taste into
digital chemical
information, and
vice-versa
Computer Vision Speech Recognition
Natural Language
Understanding
Gesture-Based
Communication
Touch-Based Inputs
Digitizing Olfaction Digitizing Taste
Related Science/Technology

4
AI has also expanded the possibilities for how
people interact with applications. Designed
thoughtfully, notifications, nudges, and
other User Interface (UI) features render
it unnecessary for people to open an app,
make an update, and then close the app,
which causes developers and UI designers to
reconsider the flow of information between
apps and the people who use them.
Says Melissa Boxer, VP Product Management,
Oracle Adaptive Intelligent Apps, “AI has
forced us to rethink the user experience. We
can't use apps in the traditional way, so we’re
starting to build in a new user experience
paradigm; for example, auto-filling fields in
machine learning-driven drop-down menus.
With recommended actions, the user interface
can now surface information in a way that is
meaningful. When should a user have to open
up an application? Why should I have to open
an app to input my expense report? So the
paradigm shift is around when to notify or
connect with the person. It’s driven us down an
interesting path.”
2. HOW WE DECIDE: FROM BUSINESS
RULES TO PROBABILITIES
Unlike rules-based systems that work by
means of “if/then” statements (for example,
if checking account balance drops below
$500, send an alert to the customer), artificial
intelligence is based on statistics rather than
rules. As a result, it is inherently probabilistic
(an outcome has an 84% chance of occurring)
rather than deterministic (if X, then Y).
Although the math is usually hidden from us,
we see probabilities in action every day; in
the recommendation engines we use (people
who like Agatha Christie also tend to like G. K.
Chesterton), the social feeds we see on Twitter
and Facebook (If I comment on a post by
my friend Emily, I probably want to see more
content from her), and so forth. While rules are
critical for some types of applications where A
always leads to B, AI, with its reliance on data
and ability to learn, always involves a degree
of uncertainty.
One example is in fraud detection. If someone
lives in California but suddenly her credit card
processes transactions in Berlin, it could be
fraudulent, or she might simply be traveling.
In this situation, she can avoid having her
credit card declined by filing a travel plan
with her issuing bank, but the onus is on her
as a customer. But if the bank used AI (and
many are doing this for just this purpose), an
algorithm might learn from her past behavior.
Maybe she’s gone to Berlin the past three
Octobers. Maybe she used that card to buy
a plane ticket to Germany in the past 30
days. Maybe she logged in with two-factor
4

5
authentication from Germany. Like humans, an
algorithm has to take into account a multitude
of implicit and explicit signals in order to come
to a conclusion or make a decision.
Of course, some probabilistic scenarios are
fairly inconsequential. A customer is unlikely
to sue Netflix for recommending a movie
he doesn’t like, for example.6
But some
probabilistic scenarios have life-and-death
consequences.
Probabilistic Decision-Making in the
Global Intelligence Community
Intelligence analysts are used to thinking in
probabilities – in their case, the likelihood of
events happening. They might evaluate, for
example, events that have financial implications:
currency valuations, changes in interest rate,
and so forth. But to make a recommendation
for action based on probabilities, they need
justifications to build their case.
Says Jana Eggers, CEO, Nara Logics, “Our
customers have seen our AI algorithm indicate
a higher probability than was expected for an
adverse event. This happens when multiple
streams of information can tell different
stories. For example, banks, politicians, and
corporations do not act in concert, and the
confluence of their actions lead to different
event probabilities than can be predicted with
any stream alone.”
“Our platform helps our customers conduct
scenario planning, and our transparent
AI helps build a case for reacting to a
probabilistic event.” But, says Eggers, “The
intelligence community needs reliable
confidence levels so they can determine what
action, if any, to take.”
3. HOW WE INNOVATE: FROM DATA
ANALYTICS TO DATA SCIENCE TO DATA
ENGINEERING
One of the most striking shifts as AI matures
is an evolution from data analytics to data
science to what will eventually look more like
data engineering — a future state in which
we are not only able to collect, analyze, and
learn from data, but in which self-learning
algorithms, fed by clean and plentiful data,
continuously inform our systems, products, and
services. This is what Omar Tawakol, Founder
and CEO of Voicera, calls “a compounding
data advantage” — the data equivalent of a
compound interest rate in which value accrues
and is amplified over time.
Before we can begin to expect anything
approaching a compounding data advantage,
however, there are a number of practical
challenges to consider. As in any major
technology shift, the most salient constraints
are mature tools and people with the right
expertise; in this case, AI toolkits and data
scientists. And, because the tools and
processes are still very new, data scientists
must spend more time on the basics — namely,
data preparation and data organization — than
on more strategic tasks. This will change over
time as the industry and technology matures
and as data science becomes more integrated
into the engineering process.
AI will be a forcing
function for a more
data-centric
organization.”
— Omar Tawakol, Voicera

6
4. HOW WE LEAD: FROM EXPERTISE-
DRIVEN TO DATA-DRIVEN
Organizations that are successfully
implementing AI — no matter their size or
industry — tend to agree on one thing: We
are moving away from purely expertise-driven
to more data-driven decision-making. This
doesn’t obviate the need for expertise — it’s
not a people versus machine scenario — but
it does mean that business leaders need to
become more comfortable with data-driven
decision-making.
“The one thing everybody has to come
to grips with,” says Steve Stine, SVP AT&T
Communications Transformation, “is that data
will be foundational to all the things you will
do. There will be more data available, so if
you’re not doing something with it you've got a
blind spot that may be catastrophic. You've got
to think of this as a sequential process in which
you have to invest. This doesn’t necessarily
mean you need your own data science team,
but you do need to learn more about what
you are trying to do and the outcomes you’re
looking for.”
Voicera’s Omar Tawakol agrees: “Marketers
need to get comfortable with driving strategy
based on what data is telling them. They
need to become data-driven versus HiPPO-
[Highest-Paid Person’s Opinion] driven.”
Tawakol maintains that paying attention to
what the data tells you is critical, whether it’s
a major decision or a tiny product feature.
“We've had features where we thought they’d
increase usage,” he says, “and they ended up
decreasing usage. Whenever we get into a
debate, we always leverage the data.”
6

7
It’s important to recognize that, despite
the investments and the hype, enterprise
adoption of AI is still nascent. As of January
2018, Gartner found that only “one in 25 CIOs
described themselves as having artificial
intelligence in action in their organizations.”7
Furthermore, many of the challenges related
to AI success—such as access to skilled
employees and mature tools—are common
across the industry. Based on interviews with a
range of industry experts, Altimeter has found
that organizations using AI technologies tend
to fall into four stages of maturity (see Figure 2):
• Phase 1: Exploring. The organization
is exploring AI, engaging with experts,
considering use cases, but has not as yet
committed significant time or resources,
either with external or internal experts, to
map processes, make data accessible or
fund AI-related initiatives.
• Phase 2: Experimenting. The organization
is actively experimenting with AI for a
range of use cases, using either internal
(employee) or external (service) resources,
but these are generally seen as discrete
projects rather than scalable and persistent
implementations.
• Phase 3: Formalizing. AI is becoming a
formal part of corporate strategy, and
data is now a core competency across the
organization. Implementations are moving
beyond optimization to more customer-
and market-focused strategies.
• Phase 4: Integrating. AI is part of the
fabric of the company, embedded into
processes, products and services across
the organization, and is delivering value
across the business.
A MATURITY MODEL FOR
ARTIFICIAL INTELLIGENCE

8
Phase 1:
EXPLORING.
Phase 2:
EXPERIMENTING.
Phase 3:
FORMALIZING.
Phase 4:
INTEGRATING.
Strategy Objectives are
undefined and no
resource or budget
has yet been
allocated.
Data is siloed, not in
accessible, useful
form; analytics are
largely descriptive
and retrospective.
Business cases
present, but no
development is
underway as yet.
Seen as promising
but unproven. Not
yet seen as a priority
at the C-level.
Emerging under-
standing of AI
governance issues
but no principles or
processes present.
Discrete proofs of
concept focus on
cost reduction,
productivity
improvement, and/or
Robotic Process
Automation (RPA).
Organization has
committed to
data strategy and
is moving from
descriptive into
predictive analytics.
Organization has
begun to use APIs
and internal or
external resources
to perform proofs of
concept and pilots.
Organization may
have a Chief Data
Officer, but data
science and AI
projects are discrete
rather than critical
elements of an
enduring product,
service, or business
Organization has
identified and com-
municated ethical
principles for AI and
is implementing pol-
icies and processes
to support them.
An expected part of
strategic planning,
focused on custom-
er experience.
Data strategy is
becoming a core
competency but AI is
not yet scaled across
the organization.
AI is becoming
a critical part of
product and service
development.
Clear understanding
of and optimized
relationship with
data science and AI
resources, whether
they are external
services and platforms,
a larger ecosystem,
or a combination.
AI ethics and
governance
processes are
formalized through-
out the business.
Integral to agile
business and a
critical component of
digital transformation
and competitive
advantage.
Organization benefits
from a compounding
data advantage.
AI is a core
development
competency across
the organization.
Organization has a
learning organization
mindset; design
thinking and
experimentation are
valued in the culture.
AI ethics/governance
is embedded in
corporate practice
and customer
experience and is
part of performance
evaluations and
incentive programs.
Data Science
Product
& Service
Development
Organization
& Culture
Ethics &
Governance
Figure 2. Four Stages of Artificial Intelligence Maturity

[REPORT PREVIEW] The AI Maturity Playbook: Five Pillars of Enterprise Success

[REPORT PREVIEW] The AI Maturity Playbook: Five Pillars of Enterprise Success

More Related Content

What's hot (20)

Similar to [REPORT PREVIEW] The AI Maturity Playbook: Five Pillars of Enterprise Success (20)

More from Altimeter, a Prophet Company (20)

Recently uploaded (20)

[REPORT PREVIEW] The AI Maturity Playbook: Five Pillars of Enterprise Success