Accenture-Ready-Set-Scale - AI.pdf

READY. SET. SCALE. 1
READY.
SET. SCALE.
A practical primer on scaling
AI for business value

About the authors
Athina Kanioura
Chief Analytics Officer, Accenture
Dr. Athina Kanioura is Accenture’s Chief Analytics Officer and the global
lead of Accenture’s Applied Intelligence practice. An accomplished
innovator and data scientist, Athina helps clients across industries
and geographies leverage and scale Artificial Intelligence, analytics
and automation to generate long-term business value. Athina joined
Accenture in 2005, and from her time as a founding member of
Accenture’s analytics business to her role as Accenture’s Chief Data
Scientist, her focus is always on delivering data-driven insights to
power clients’ businesses. She holds a PhD in Macroeconomics and
Econometrics from the University of Sheffield.
Fernando Lucini
Artificial Intelligence Lead, Accenture UK and Ireland
Fernando Lucini is a Managing Director leading Artificial Intelligence for
Accenture in the United Kingdom and Ireland. Fernando is a passionate
and experienced senior leader with extensive experience in Artificial
Intelligence and Machine Learning. Previously Fernando spent 18+ years
in the enterprise software industry, creating technologies to automate
and understand text, speech and video data and integrating these into
business solutions for various Fortune 100 companies.

Table of Contents
Focus on value: Your business strategy is your AI strategy
Rethink work and get your people ready
Still having trust issues? Build responsibility into your AI
Productionise, and get ready to realise value
Get in touch
01
03
02
04
05

THINK BIG
AND SCALE
WITH INTENTION
Why are so many businesses still struggling to realise the
full value of their AI projects? The answer appears to lie in
the gap between applying AI in pockets versus applying it
at scale. Our latest Accenture Research, AI: Built to Scale,
indicates that Strategic Scalers—organisations that are
scaling AI effectively—are seeing up to three times the
return on their investments compared to non-scalers, or
the ‘Proof of Concept (POC) group’, who are conducting AI
experiments and pilots but achieving a low scaling success
rate and low return on their AI investments. Four out of five
UK executives (versus 75 percent globally) believe they
will be out of business in five years if they can’t crack the
code of scaling AI effectively. Companies need to get their
heads around how to scale AI and fast—as global revenues
from AI for enterprise applications are projected to grow by
more than 50 percent between 2018 and 2025.1
Strategic Scalers achieved
nearly triple the return
from AI investments
compared to their non-
scaling counterparts.
3X

In practice, companies still find
it difficult to make the transition
from thinking about AI as a source
of innovation to a critical source of
business value.
There’s a state of paralysis beyond the pilot.2
Why? Until
now, there hasn’t been a proven blueprint for scaling,
and organisations can fall into some common traps. First,
companies don’t have an AI roadmap or ‘route to live’—
the steps to take their AI project from POC to production,
effectively and expediently. AI is different from ‘traditional’
software implementation projects, which companies
are typically set up to deliver. Changing the status quo
requires agility, openness to trying a new way of working
and the ability to recognise when an idea works—and
when it needs to be scrapped.
Second, the unfamiliar landscape of AI also means
businesses can be tempted to fall back on their time-
honoured behaviours, reinventing the wheel and building
from scratch. Big mistake. There are many proven, low-
cost AI options to buy ‘off the shelf’ and start using straight
away. It is key to leverage what already exists, customise as
The difference in return on AI investments between companies in the Proof of Con-
cept stage and Strategic Scalers equates to an average of US$110 million.
PROOF OF
CONCEPT
32%
STRATEGIC
SCALERS
86%
needed for the organisation and start proving the value of AI
as the first step to successful scaling. But don’t get bogged
down in the technology. Be driven by the business strategy
and vision, and let that dictate the AI approach. Focus on
finding the right way of working that will allow AI to flourish,
diversifying skills and talent beyond the data scientists. And
get the right governance approach in place from the outset,
with outcomes in mind.3
US $110M GAP

The following chapters will help you
think through these critical success
factors and understand how you can
unlock a new wave of exponential
value by scaling AI successfully.

FOCUS ON VALUE
YOUR BUSINESS
STRATEGY IS YOUR
AI STRATEGY
01

Decide what to focus on—and focus.
To scale AI successfully, get your ducks in
a row early. That means:
Understanding what business value means to you
Translating that definition into a business strategy and
Focusing in on AI solutions that explicitly deliver on the most
critical elements of that strategy.
Simple, right? If you have already defined value in your own
unique context, you can harness AI to multiply that value—
not just grow it marginally—charting a course that genuinely
aligns with your business’s strategic priorities and delivers
unprecedented returns. In our research, AI: Built to Scale,
Strategic Scalers understand this imperative, with more than
70 percent linking their AI ambitions explicitly to their overall
business strategy.4
01
02
03
of Strategic Scalers have
a clearly defined strategy
and operating model.
71%

Look to the highest-level priorities.
It seems like more and more applications of AI are
emerging each day. So, how do you determine which
applications are going to deliver value, whatever that
means for your specific context?
Finding true value starts with defining what really matters
to a business and aligning the AI agenda to the highest-
level strategic plans.
Ask yourself: What are the boardroom’s short- and long-
term priorities? How can AI help achieve the objectives of
the C-suite such as organic growth, expansion into a new
industry or development of new products?
For example, your company might look at 300-400
potential applications of AI (‘use cases’) as a starting
point. They can’t all be of equal value—or equally aligned
to the overriding business strategy. A clear framework for
identifying value should be used to qualify and prioritise AI
use cases. To help get to that framework, you should think
about these key questions:
1. What growth levers can you pull to
deliver the most value?
2. What is the potential business value in
monetary terms?
3. Will the use case be a differentiator in
the marketplace, compared with those
of competitors?
4. Will it make our colleagues’ lives
simpler and better?
5. Does the use case improve customer
satisfaction and promote loyalty?

Define value for today—with a vision
for tomorrow.
While you need to look at the short-term return AI
can create for your business, you also need to look at
value—and therefore your high-level priorities—through
a broader lens. Where is your organisation headed in the
‘human-plus-machine’ era? What is the future of your
industry? Will that change how you define value three to
five years from now? A full 88 percent of UK executives
say they won’t achieve their growth objectives without
scaling AI successfully.
AI has the power to disrupt well beyond individual
businesses. It is already blurring traditional industry
boundaries, threatening legacy companies and giving
agile new entrants the chance to make an impact, fast.
Make sure you’re paying attention to what’s disrupting
your industry already, how your world and the world at
large are changing, and adjust your strategy, act boldly
and invest to buy your way into the action. You may find
yourself making different choices when you bring the
macro into play.
For example, self-driving cars are dismantling the
distinction between the car industry and the insurance
industry.5
New questions are emerging—will the car
industry require more safety features on self-driving cars
as they evolve to keep people safe? Will the insurance
industry be pushing for the same or different features
to reduce claims? What about the effects of enhanced
safety on vehicle design, insurance policies, emergency
response, road network design, parking needs, car
dealerships, petrol taxes and lawyers?
This is just one example of AI’s power to redefine
industries. It shines a light on how your organisation
may need to change with the context of the larger-scale
changes around you. We believe we can get under the
surface of disruption—to understand how it is changing
industries and how to capitalise on its opportunities. Our
Disruptability Index can help you determine where your
industry sits and how you can act now.6

Think practical,
not provocative.
Leading AI organisations don’t prioritise
interesting or novel use cases that
aren’t aligned to the direction of
their company, how their industry is
developing or how society is evolving.
Additionally, they tend to think
practically about what will work in their
operating conditions, ensuring that…
• The required data is readily available
or easily sourced
• The use case can be integrated with
existing tools and delivered to the
front line
• The necessary skills and capabilities
are in place and able to execute the
plan
• There are data privacy or security
risk protocols in place to mitigate
unintended consequences
• There’s an established method for
calculating value and ROI
of executives say they know their industry will be
disrupted at some point in the next five years, but
only 20% feel they’re well prepared to address it.7
Decide what value looks like for you—now and
three years down the line. Don’t be so focused on
delivering for today that you aren’t prepared for
the next wave. Understand how AI is changing
your industry and the world—and have a plan to
capitalise on it and stay relevant.
93%
11

Take a portfolio view of your AI projects.
To be successful on your AI journey, think about your AI projects as
a portfolio of things you’re trying to achieve. This means thinking
holistically about where you’re headed and navigating the iterative
nature of AI initiatives while remaining aligned to strategy and value.
Scaling value relies on a formally defined AI roadmap which can
help you deliver faster with more rigour and get to production more
quickly.
The first step of the lifecycle is to create an ‘idea pipeline’—and
populate it with potential AI concepts that are yet to be tested
for feasibility and value for your business. Shape, develop and
investigate those ideas iteratively—but quickly—before a ‘go/no go’
decision. The ideas you generate may vary in terms of their potential
to succeed, so having a holistic view of the collective success of your
AI projects will be vital.
For example, data-driven technologies such as machine learning
have a high risk/reward profile because development proceeds
through a series of experiments. These experiments continually
solve business problems by moving the engineering solution toward
the optimal data pre-processing, learning algorithm and algorithm
parameters for that specific use case.
Stay focused on your
priorities while you learn.
At Accenture, we often find ourselves
advising clients to spend at least 80
percent of their time designing AI that
aligns very directly to their strategic
priorities—and 20 percent or less on
experimenting and learning for pure
capacity building in the absence of a
very clear target outcome. Experience
tells us that when it comes to AI, even
within that 80 percent, there’s still a lot
of experimentation!
12

Many of your experiments may not give you
the outcome you think you needed.
A solution is never guaranteed—and
that’s OK.
Therefore, assimilating AI into your business brings a new type
of project execution risk, with only a portion of your ideas and
experiments expected to go to production. But the good news is
that following an AI roadmap, like the one here, helps qualify ideas
quickly and effectively—so ideas that fail, fail fast and can be shelved
with minimal investment before moving on to something else.
Strategic Scalers have mastered this approach. This group pilots
more initiatives and successfully scales more often than their
counterparts: They reported scaling 114 applications in the past three
years, compared to just 53 for companies in the POC group.8

Underpin your AI strategy
with a data strategy.
Every AI transformation journey starts with data. The
vast majority of AI Strategic Scalers (72%) agree that
a core data foundation is an important success factor
for scaling AI.9
More specifically, they understand the
importance of having a data strategy—a design and
intent that underpin what data is being captured, in
what way, and for what purpose. The data strategy
drives value as much as AI does.
And more data is not always better. In a world where
data is proliferating and data begets more data, it can
be tempting to gather more and more. Having a strong
data strategy ensures you’re curating the right data
to deliver the desired outcome and then capturing its
insights to fuel an AI strategy that delivers that outcome
at speed and scale.
Once the data strategy is set, data can be mined
to generate insights that help refine both the
organisation’s strategy and the AI systems themselves.
To really get the most out of this constant stream
of data-driven insights, you’ll need to explicitly
integrate ‘feedback loops’ into business decisions in
an orchestrated way—for instance, to fine tune your
business strategy and/or make necessary adjustments
to your AI initiatives at the same time. This requires
a new way of working: an agile, iterative approach to
decision making—as well as AI development—with
data at the core.
of Strategic Scalers have a core
data foundation, compared to
just 63% of their non-scaling
counterparts in the POC group.
72%

a. Are you able to reuse your valuable data and
insights or do you recreate the wheel each time?
b. Are you mobilising your data with internal
resources or are you working with partners?
What part of your business (function), if any, is most
AI-mature/ready? Are you already implementing AI
in your organisation?
Are you running it within the business or in the
technology function?
Do your current or planned AI use cases conform
to your ethical framework? Are they aligned to a
responsible purpose?
a. What are the ethical norms and regulatory
considerations in your industry and/or market?
b. Or, are you operating in any markets that have
launched or are preparing to launch their own AI
policies, strategies and initiatives? If so, have you
thought about what you can leverage for your
organisation?
Self-Assessment #1: Value and
AI Strategy.
Now that you have read our empirical insights on
defining value and an AI strategy, here are some
practical questions to help you ‘self-calibrate’ and
determine your readiness for AI at scale.
What are your organisation’s strategic objectives
and vision? How can AI help you achieve your
desired outcomes?
a. What does ‘value’ mean to your company? What
does it look like?
And how do you currently calculate it?
b. Have you thought about what AI use case(s) will
provide most value for you?
c. Are these use cases feasible? Do you have the
necessary elements—data, technology, capabilities,
etc.—within your company, or with partners, to
deliver the use case(s)?
Do you have a data strategy? For example, are you
able to find and use your data in ways that will help
you achieve the objectives above? Is there any data
you’re not using that you wish you could extract value
from?
01
02
03
04

RETHINK WORK AND
GET YOUR
PEOPLE READY
02

Challenge your assumptions about
‘work’ and get your people ready.
AI’s disruptive nature means your old ways of working will
need to change. Those who can successfully integrate AI
into their culture and processes will be able to multiply
value for businesses, employees and customers alike.
Our AI: Built To Scale research confirms the correlation:
Strategic Scalers are more likely than those in the
Proof of Concept group to embed AI ownership and
accountability into teams and ensure employees fully
understand AI and how it relates to their roles.10
UK
executives may understand this imperative better than
most, with 72 percent of all UK respondents giving every
employee working with AI formal training, compared to
67 percent of all respondents globally.

Reimagine how work gets done.
The business of the future looks starkly different when your abilities
can be multiplied with the use of machines. So how do you integrate
AI into the way you work to maximise productivity? How do you
reallocate workloads and reskill/upskill employees? It’s imperative
businesses think about these questions to successfully integrate AI
into their culture and processes.
Word on the street.
As AI has moved into the adoption phase,
urgent calls for reskilling and fears about
job loss have taken centre stage.
But we believe the majority of jobs will
be reconfigured—not lost—as human-
machine collaboration evolves. In fact,
high-performing organisations will
need big and diverse teams to scale AI
successfully, with AI Strategic Scalers
using twice the number of team members
(about 200 people) for their AI projects
compared to those in the POC stage.
Some roles will be totally new. For
example, we have the ‘AI trainer’ who
intervenes and manages AI technologies—
helping systems ‘learn’ tasks and adjust
conclusions—and helps us stay on the
straight-and-narrow when it comes to
ethics. As we learn new skills and delegate
work to AI technologies, we can help the
AI learn as well. It’s a team, just not as we
know it.
18

Benefits across the business.
Leading organisations understand
that the human + machine era will
require them to break down and rebuild
traditional notions of jobs and activities.
Companies from around the world and across
industries are using AI to change the fabric of what
they do and how they do it. Strategic Scalers are more
likely to achieve a range of benefits including:

Start configuring the business
of the future—now.
There are some practical steps you can take to start
configuring business processes and the workforce to
support AI at scale. Here are some key considerations
to help you organise your business and improve agility
in the way your people work.
Move from ‘workforce’ planning to ‘work’
planning
Break down traditional job roles, and look at which
tasks and activities will be automated, which will
require human-machine collaboration, and how this
might impact how people and teams intersect and
interact.11
Look seriously at new skilling
Get a clear view of the knowledge and skills you’ll
need to generate real value from human-machine
collaboration. Look at your leadership, learning and
recruitment programmes, and invest in new ways to
teach new things. For example, we put 60 percent
of the money we save from investments in AI into
our training programmes.12
01
02

Look at the big picture
What entirely new jobs—such as the ‘AI trainer’—can AI
create in the organisation? Are we prepared for those
in the context of new markets, products and customer
experiences? Operational roles may become insight-
driven roles. For instance, when autonomous vehicles
are introduced into society, the long-haul trucker
may be reskilled to perform high-level technical work
as an ‘in-cab systems manager’. He or she might be
monitoring systems and optimising routes—potentially
for multiple vehicles.
As we lean into human + machine collaboration, many
human tasks will be augmented by AI. For example, AI
can provide enhanced views of real-time data to help
support decision making—without the decision making
itself necessarily being offloaded to AI. It’s important to
be clear about the right boundary, or process, for the
organisation when it comes to the split between the
human and the machine—including how that boundary
may shift as the organisation’s AI maturity continues
to change. Successful scaling relies on understanding
how the organisational chart will change with the
upskilling and reskilling of people to be ‘data native’ and
with new ways of delineating jobs and tasks.
03
of workers believe that AI will have a
positive impact on their jobs, and 67%
say it’s important to develop skills to
work with intelligent machines.14
AI may be good for workers:
Your workforce may be more ready than you think to
adjust.
Our research on the Future Workforce says so.13
Now it’s
up to you to take action.
62%

You need the right mix of talent to
move from POC to production.
Establish the right talent mix.
It’s no surprise that you need new kinds of talent to
create AI products and services that deliver value.
But beware of thinking data scientists are the only
ones who matter when it comes to creating a route
to ‘go live’.You also need data integration experts,
business analysts, data engineers, and software
engineers among others—and enough of them, in the
right configuration.
In addition to the technical skills, it’s important your
team is interdisciplinary, bringing industry, business,
design and governance expertise in the right ways
and early on. These areas of knowledge might be
easy to overlook, but they play a crucial role in
creating successful AI applications. Think about
it—we are creating AI to solve real-life problems,
so having diverse specialists with expertise across
numerous fields will help create a better AI solution.
Additionally, AI projects have a habit of generating
new and unexpected demand for knowledge and
skills—and having an interdisciplinary team can
create a buffer, covering for any skills shortfalls you
may encounter on your pathway to scaling AI.15
Strategic scalers build diverse teams to scale AI:
Data integration experts
Platform engineers/application specialists
Data visualisation experts
Business analysts
Data scientists
Business unit leaders
Data modellers
22

Look at your organisational set-up.
Along with establishing who gets the work done, it’s
important to revisit the ‘how’. A priority should be to
set up your organisational structure and governance
strategy and establish the right connections within
the business. Think about the kind of physical set-up
that will help you achieve your business goals and
integrate AI most effectively. For instance, do you need
geographically dispersed business units and AI tools or a
more centralised structure?
A centralised organisational model may be the most
effective, with Strategic Scalers saying they now use
this approach.16
This model uses a centralised cross-
functional AI group (sometimes called a Centre of
Excellence) to deliver projects to the various businesses,
geographic units, and functions. This central group
defines a standard way of working, governance model,
and method of interaction for solution selection and
implementation.
Another variant is ‘hub-and-spoke’, a model that includes
both a centralised cross functional AI group (i.e., the
‘hub’, sometimes called a Centre of Excellence) and
separate autonomous AI teams (i.e., the ‘spokes’) that
sit within business units. The hub provides a common
set of services and knowledge relevant to all business
groups or functions, and sets common guidelines/
operating principles, with some guidance from the
business unit teams in the spokes. The spokes operate in
a decentralised way to implement AI solutions relevant
for their business units/functions.
Finally, a ‘distributed’ model also exists. Highly
autonomous AI teams are housed within each business
unit or function, with a delivery focus specific to that
business unit or function. There are a few common
standards, tools or operating principles imposed, with
potentially varied practices across teams.
Highly distributed ways of working are the sign of
mature technology, tools and usage patterns. Here,
AI practitioners have considerable resources at their
disposal and can be productive working in distributed
but loosely connected groups with established ways of
reusing and providing knowledge and expertise to the
wider business.
Be guided by your business aspirations, and define a way
of working that best supports those goals and your level
of AI maturity.

Mind the gap.
Top challenges for scaling AI (Ranked top 5)
When you cut the distance between the
C-suite and the AI practitioners, you im-
prove your odds of delivering value.
There can be a gap between the CEO’s understanding
of AI—what it can do and how—and what the people
actually implementing the AI believe. The CEO’s
perspective will naturally be influenced by the topline
strategic intent of the company, what her peers are
telling her, and what her long-term aspirations are for
the organisation. The AI leads doing the work might not
always be aligned with the realities and focused goals
of the C-suite—but they need to be! In fact, our research
indicates that leadership’s limited understanding of AI’s
potential can be one of the top challenges companies
face when scaling AI. Strategic Scalers ‘mind the
gap’—they reduce the distance between the goals and
understanding of the C-suite and the practitioners when
it comes to how AI can and should be applied to change
the world, and their world.17
Employee adoption of the solution
Expected ROI from AI investments not adequate
Inability to set up an organisational structure that
supports continuous innovation...
Foundational capabilities (data, technology)
Failure to scale AI Pilots in the past
Shortage of or difficult access to AI skills
Lack of culture for change
Poor understanding of AI potential within top
management
Inability to visualise the final outcome
Lack of clear roadmap
Lack of budget
............................. 53%
........ 52%
...... 48%
............... 47%
............................. 47%
.................... 46%
........................................... 43%
............. 42%
......................... 42%
................................................. 40%
.............................................................. 38%

When it comes time to implement,
look outside your organisation
for help.
Growth of AI Startups in the Market
We are now firmly in the ‘era of implementation’ with
an explosion of investment in AI capabilities coming
from well beyond Silicon Valley.18
These days, there
are myriad tools which are proven, low-cost and
academically rigorous. And there are varied and
flexible ways to get your hands on AI: open source
code, application programming interfaces (APIs), and
small and medium-sized enterprise (SME) vendors to
name just a few. As AI becomes mainstream, solution
price points will also continue to drop.
AI start-ups are multiplying quickly as well. This
means there’s an ever-increasing pool of potential
partners to help you on your AI journey. The figure
below shows the number of AI start-ups in the market
over time. Just look. In less than five years, the
volume of machine learning AI start-ups has more
than quadrupled.19
Now you can reuse, partner or buy to implement and
scale AI capabilities before you even need to consider
building new proprietary technologies in-house. Take
advantage of what’s out there for success at speed
and scale.
Number of Startups

Reuse before you buy. Buy before you build.
So how do you decide when to reuse, buy, partner or build? This is a full
topic in and of itself (we published an article on it), but the simple answer is
almost always reuse, buy or partner to take advantage of the investment other
companies have already made—and get started quickly.20
And ‘buying’ doesn’t need to mean purchasing an enterprise software product—
rather, it’s a more fluid idea about paying for value and can include using a
service, open source code or packages. We’ve helped clients, for example, move
from buying speech-to-text software in its entirety to buying specific APIs that
best solve the problem at hand—and pay for only what is used. By arbitraging use
between the top vendors, we efficiently tackled the specific problem, sped up
execution and avoided being stuck with the limitations of the software.
‘Buying’ can also mean purchasing an entire company with the capabilities
to fulfil your future business needs. For example, in 2018, Apple bought Silk
Labs,21
a small start-up specialised in developing on-device machine learning
(ML) services, and it was reported that the technology could potentially be used
not only to improve Apple’s existing device solutions but also its ML tools for
other applications.22
By acquiring companies that already have the people and
capabilities in a specific area of interest, you can bypass steps and accelerate
your scaling journey.
So you’re resolved to
build—but are you doing
it for the right reasons?
Sometimes, organisations decide to build
AI in-house for the wrong reasons. For
example, inexperienced teams may want
to build from scratch as a way of learning
on the job or trying to demonstrate their
value—or they may believe they are the
only ones who can get it ‘right’ for their
companies. But don’t get caught in this
trap. Build only because it makes business
sense and the output will genuinely make
you stand out.
Key questions to ask yourself to qualify
the opportunity to build AI:
• Have we proven the value withan agile
pilot? Prove the principle at small scale.
• Does someone in the business really
need this, and is it a true differentiator? Be
demand- and value-driven.
• Will they pay for it? Engage the business
to generate investment commitment.
26

Self-Assessment #2:
People and Capabilities.
Now that you have read our empirical insights on building
your talent mix and capabilities, here are some practical
questions to help you ‘self-calibrate’ and determine your
readiness for AI at scale.
How much trapped value could be unlocked by moving
POCs into production on a monthly basis?
Do you understand where the existing analytics tribes
and data scientists are within your organisation?
Are they equipped to deliver against the business
outcomes you’re pursuing with AI or do you need to
hire and create a more diverse talent pool?
Have you professionalised data science in your
organisation?
01
02
03
04
05
Do you have a route to ‘live’? Do you have an
established plan for productionising to ensure
that projects move past POC? Have you defined
who is responsible for creating models and then
operationalising/productionising the models? Are there
gaps? Where are they? (Often the latter is a gap.)
Is your incentive structure set up to encourage the
support of AI ‘productionisation’? Do you have a sense
of the right breakdown between buy, partner, reuse,
build? (Remember to build last, or only when it makes
absolute business sense.)

STILL HAVING
TRUST ISSUES?
03

Still having trust issues? Build
responsibility into your AI.
How do we learn to trust AI? Responsible AI builds trust and
lays the foundation for successful scaling by taking a ‘human
first’ approach—using technology to help people make better
decisions, while keeping them firmly accountable through the
right governance processes and technical steps. Our AI:Built to
Scale research says responsibility is more than a ‘nice to have’—
with AI Strategic Scalers significantly more likely to brief their
employees clearly on how they tackle responsible AI,23
and the
UK markedly above the global average in this respect.

So you see the value in AI… but how do you
trust it?
AI affords tremendous opportunities, from increasing efficiencies
and improving outcomes, to reimagining industries altogether.
Against this backdrop, it’s easy to forget that AI’s decisions also
have a real bearing on people’s lives, raising some big questions
around ethics, trust, legality and responsibility. Enabling
machines to make decisions may expose a business to significant
risks, including reputational, employment/HR, data privacy,
health and safety issues.
Enter: Responsible AI. It’s a topic that’s becoming pervasive in
the media and a real consideration for clients in the public and
private sectors.
What happens when a machine’s decision turns out to be
erroneous or unlawful? The potential fines and sanctions could
threaten the business’s commercial sustainability. And what
about other unintended consequences?24
AI has already shown
it can be biased in ways that weren’t anticipated and can hurt
a brand’s reputation. AmazonSM
, for instance, had to scrap its
AI-based recruiting tool that appeared to show bias against
women.25
And if need be, how does a human know when to
intervene in a process driven by a machine?
In an Accenture global executive survey,
60 percent of respondents believe that
the adoption of AI is necessary to have a
competitive advantage in a digital economy;
however, 45 percent of our clients agree
that not enough is understood about the
unintended consequences of AI.27
30

What are unintended
consequences?
Unintended consequences are the
by-products of applying technology
without accounting for the potential
impact on a variety of unique
communities and stakeholders—
whether they are the intended
audience for the technology or not.
31
We define Responsible AI as the practice of designing,
building and deploying AI in a manner that empowers
employees and businesses and fairly impacts customers
and society—allowing companies to engender trust and
scale AI with confidence.
As AI increasingly permeates healthcare and government, the stakes
get higher and potential unintended consequences could escalate. For
instance, we will begin to trust AI tools to help doctors assess patients and
determine if fast action is needed. ‘Explainable AI’ will be crucial to the
longevity of these technologies.26

Design trust into how you operate AI from
the start. And from the top.
However, it’s not just about establishing the
appropriate governance structures. It’s also
important to translate those ethical and legal
frameworks into statistical concepts that can be
unambiguously represented in software. In other
words, there need to be clear and prescribed
actions to support your ethical AI principles.
The Board of Directors needs to know what obligations it owes to
its shareholders, employees and society at large, to ensure AI is
deployed without unintended consequences.
The CEO might be asking, how can I be assured we have thought
through AI’s possible brand and PR risks? Meanwhile, the Chief
Risk Officer and Chief Information Security Officer need to be
thinking: If we deploy AI, how can we do it in a way that complies
with data protection regulations? Creating a robust ethical
underpinning for AI allows you to ‘design out’ legal and ethical
concerns to the extent that it is possible.
Putting the human
at the Centre.
At Accenture, Responsible AI is more
than just technology alone—it requires
an interdisciplinary, innovation-friendly
approach with humans at the core. The
approach is rooted in our principles of
TRUST:
• Trustworthy
Safe, honest and diverse
• Reliable
Enabling better decisions
• Understandable
Interpretable and transparent
• Secure
Strong data privacy and security
• Teachable
Human-centric in design
32

So, where to begin?
First, ensure considerations for AI are built into your core values
and robust compliance processes. Then, you will need to
implement specific technical guidelines to make sure that the AI
systems are safe, transparent and accountable to protect your
employees, clients, civilians, and other organisations.
Next, identify new and changing roles, and put the right training
in place for technology specialists and your diverse team of
experts to understand their new roles and remit. It’s also critical
to articulate a responsible business mission and conduct
brand and public risk assessments to ensure your AI initiatives
remain anchored to your core values, ethical guardrails, and
accountability structure.
All of these elements are part of an innovation-friendly blueprint
for Responsible AI that you can apply across functions and
projects—allowing you to understand and manage the ethical
implications of everything you do.

Police patrols versus fire wardens.
There are two main ways to approach AI governance: the police
patrol and the fire warden.28
Police patrols are the effect of AI
governance rules applied from the top down. In this model,
organisations monitor people and detect violations of the rules
in order to enforce compliance. It seems like a straightforward
option, but it tends to stifle innovation. Teams see AI governance
as a barrier, rather than their shared responsibility.
The fire warden model is different. It embeds skills within teams to
help them escalate issues that need attention—much like training
fire wardens to raise the alarm and then carry out the necessary
safety actions.
In general, we favour the latter approach for AI governance
because it supports innovation and the agile development that is
crucial to the competitiveness of today’s businesses—and it can
evolve more easily alongside fast-changing AI technology.
In an Accenture global executive survey,
88 percent of respondents indicated that
they do not have confidence in AI-based
decisions and outputs.30
And 60 percent of
respondents reported that they still require
a human override of an AI system at least
once a month.
34

That said, you should review the outputs of your AI systems as
often as every week and have processes in place for overriding
questionable decisions.29
And as you begin to scale AI more broadly, it’s unlikely legislation
will be able to keep up just yet. A robust ethical framework will be
essential to deal with issues too complex or fast-changing to be
addressed adequately by legislation.
To generate trust from the outset, have a view on
Responsible AI and a blueprint for how to achieve
it—only then can you scale with confidence.

Put ethics at the core of AI develop-
ment to build and retain trust.
Design in ethical frameworks when you’re planning AI. We
programme algorithms to give us exactly what we have
asked for, so we shouldn’t be surprised when they do.
And the problem is that simple algorithms treat all data as
immutable, even data about our preferences, income and
life situation. What can happen then, is that algorithms can
trap people in their origins, history or a stereotype. These
‘bad feedback loops’ can lead to negative impacts on
society.
The issues mentioned are not inherent to machine
learning algorithms themselves. Instead, issues arise from
the way they interact with society and the unintended
consequences that can result from those interactions. As
such, putting the ethical implications at the heart of the
development of each new algorithm is vital.
One way to ensure this is by embracing public health
models of governance, which treat issues as indicative of
underlying drivers, rather than problems to be solved per
se. Another would be to ensure algorithms can be adapted
more readily to newer or better data, in ways that do not
exaggerate historical patterns. We see this every day in
the way that AI at Spotify™ or Amazon quickly adapts
recommendations to our latest searches.
Finally, targeted research identifying individual problems
and solutions is critical to the success of any effort to
create more ethical AI. We need to see more resources—
and more senior leadership attention—directed at
making sure algorithms do not have negative impacts
on individuals or society. Just as data privacy and cyber
security have moved from departmental to board-level
issues, responsible governance of AI must be quickly
elevated in importance by all organisations that use it.

Trade your black box for a glass box.
With a recent upsurge in machine learning, especially deep learning,
it can feel as if AI is working in a ‘black box’. But when you’re on the
receiving end of an AI-generated decision, how do you know how it was
made or how to challenge it? For example, if AI is deciding whether or
not to grant you a loan, how did it make that decision, and how would you
dispute any of its conclusions?
In light of such issues, perhaps the greatest barrier to AI adoption among
consumers will be building—and sustaining—trust in AI. This is where
transparency and explainable AI come into play.31
Be open with consumers about when they are interacting with an AI
capability. Be able to show how decisions are made. Create strong
frameworks for human decision override and review—especially where
it most impacts the consumer’s experience. And edit AI’s outcomes
frequently to bring it out of the black box.
We will only begin to understand and trust AI when
‘explainability’ takes it out of the ‘black box’ and
demystifies its decisions.
37
To meet the requirements around
compliance, accountability and
transparency coupled with the adoption
of AI models, financial institutions will be
required to explain the workings of their
AI models. For example, the Explainable
AI (XAI) research project run by the
Defense Advanced Research Project
Agency (DARPA) is developing a suite of
ML techniques that can produce more
explainable models—enabling humans to
interpret, trust and manage AI systems.32

Self-Assessment #3:
AI Governance and Responsible AI.
Now that you have read our empirical insights on AI
governance and Responsible AI, here are some practical
questions to help you ‘self-calibrate’ and determine your
readiness for AI at scale.
What’s your company definition of fairness and bias?
How do you translate that definition and your code of
ethics into implementation standards (turning talk to
action)?
Do you have an ethical data usage manifesto? Do your
customers know if/that you’re ethically responsible
with their data and yours? Do you have regulation and
compliance capabilities in place to manage data and
AI technology?
What is your process for mitigating biases in the data
you use to power AI initiatives?
01
02
03
04
05
Have you begun or are you planning to begin embedding
mechanisms for achieving fairness, transparency, and
accountability into your design, development, deployment
and monitoring of AI systems? Who is accountable for the
decisions made by AI systems?
Have you updated your risk frameworks to incorporate
contingency plans for incorrect outcomes associated with
a range of inputs, including the use of inaccurate historic
data and lineage? Is there a formal process in place to mitigate
unintended consequences of AI decisions?

PRODUCTIONISE
AND GET READY
TO REALISE VALUE
04

Scaling value is about understanding how to move from pilot to
production; getting your data strategy in place to drive real-time
strategic actions; and establishing the right talent mix, operating
model and governance framework. Those who succeed will reap
the rewards. And those who fail may find their businesses fall
by the wayside (84 percent of UK executives believe they will be
out of business in five years if they cannot scale AI effectively—
compared to 75 percent globally).33
AI is no longer a ‘nice to have’ or a set of cool tools to impress
management. AI and data strategies are becoming the very core
of business, and all the while it’s becoming easier and cheaper to
get your hands on the technology. The time to act is now.

41
There’s a lot to think about—and a strong business case to get started
quickly. In this primer, we have asked some questions and provided some
insights on what it takes to scale AI effectively and move beyond proofs of
concept to production. But how does it all come together in practice—and
what concrete steps can you take to realise value quickly?
The final chapter of this primer is our AI Roadmap, a start-to-end model
we use with our clients to help them realise and multiply value from
their AI projects. It details an AI use case’s route to live, which includes
defining value and formulating a solid AI strategy; bringing together the
right AI capabilities; thinking about the optimal talent mix; and getting the
appropriate governance and ethical parameters in place. But it doesn’t
just end there. It lays the path for how to multiply value from the use case
through continuous engineering, optimisation and the extension of the
feature to new use cases.
We invite you to go through our roadmap and evaluate how you’re
approaching your AI projects. Stop at each checkpoint and ask yourself
the flagged questions to make sure you’re setting yourself up for
success—with your data, your people, your infrastructure and your
organisation at large. Whether you’ve been in proofs of concept or are
already starting to scale AI, be assured that there are concrete steps you
can take to realise even more value from your AI initiatives.
So are you ready? Because now’s the time to get set—and scale.
Recap: Are you built
for AI at scale?
Have you defined what ‘value’ means to
your business and which AI use cases
to prioritise to deliver on that defined
value? Do you have a data strategy in
place, knowing data is a critical enabler
of your AI technologies?
1. Do your AI teams understand the
business goals you are trying to
achieve with AI—and are they set up
to move to production (and avoid
being trapped in POCs)?
2. Do you have ethical frameworks and
a way to course correct should the AI
make erroneous, biased or unlawful
decisions?

AI ROADMAP - The Journey to Live
CHECKPOINT #3
Are there any adjustments
you need to make to your
operating model to optimize
how these specialists can
work together?
CHECKPOINT #6B
Are you able to re-use the
feature you’ve just created to
deliver value against other
priority use cases? If not, what
tweaks can you make to expand
the use of the feature?
Delivery
Sprint 1 - N
DELIVER EPICS
USER STORIES
Iterate epics and user stories in
detail and begin development
to create the Product.
Understand the
customer need.
Establish the use case
and business benefit.
Establish your
data strategy.
Curate the right data to deliver
your desired outcome. Then
determine how to bring that data
together to fuel your AI strategy.
Generate
senior
stakeholder
buy-in
towards new
journey.
Analyse the impact.
Review the strategy
and use case, and
provide scope and
timelines.
Create a
data foundation.
Make sure your
Design Authority
reviews and approves
the use case and
feature, as well as an
agile development /
release schedule.
Update product
roadmap
to include the
new feature.
Identify your
support functions.
Who in the organization
needs to get involved (i.e.
risk, legal, impacted
systems)?
Build devops
environments
and pipeline.
Have high-level solution architects
review journeys in alignment to
business strategy sign off.
Copyright © 2019 Accenture. All rights reserved
CHECKPOINT #2
Is the data you’re currently using
going to be able to deliver the
expected outcome for your use
case? What adjustments are
needed? Thinking ahead, can
your feature be expanded to
support additional use cases?
CHECKPOINT #4
Are you seeing opportunities to engage
more vendors or partners before moving
into production? Do you have the right
team and feedback loops to continuously
improve production?
Include design and technical
spikes. Concentrate your
resources on designing and
implementing new features.
Build a
product
backlog
of epics
and stories.
Refine sprint plan.
Finetune and complete
user stories, visual design
and wireframes. Organize
scrum ceremony.
Support
the engaged
functions.
Train and enable
business and customers.
Ensure the business and
customers are prepared to
work and engage with the
new feature.
Obtain approvals
from support functions
(i.e. risk, legal, security).
Sprint 0
2-3 Weeks
PLAN CREATE
EPICS USER STORIES
Rapidly iterate high-level
design for the releases, related
features and epics.
CHECKPOINT #1
Have you defined your
data and AI strategies?
Do you know what value
you expect to achieve?
START
Define and analyse
Non-Functional
Requirements (NFR).
Obtain solution design
approvals from architect,
design working group, etc.
Go live in
production.
Multiply value
by supporting
additional use
cases with your
reusable feature.
CONTINUOUS
ENGINEERING
6B
1
3 4
CHECKPOINT #5
Have you updated your risk frameworks
to incorporate contingency plans for
incorrect outcomes? Who is accountable
for the decisions made by AI systems?
CHECKPOINT #6A
Are you realizing value as expected
or projected? How are you
measuring it? Are there optimizations
you need to make to maximize?
5
6A
OUT OF SPRINT
ACTIVITIES
Delivery
Sprint 1 - N
DELIVER EPICS
USER STORIES
Iterate epics and user stories in
detail and begin development
to create the Product.
OUT OF SPRINT
ACTIVITIES
VALUE + STRATEGY
GOVERNANCE
VALUE REALIZATION
PEOPLE + CAPABILITIES
2
AI roadmap - The journey to live.
To scale effectively, run an unbreakable thread that traces the critical path to production through
all of these highly connected elements. Only then can you amplify value.

GET IN
TOUCH
05

Join the conversation.
Authors.
Contact the authors to find out more about how to scale AI no
matter where you are on your AI journey—and see how value can be
multiplied for your business.
@AccentureAI
www.linkedin.com/company/accentureai
www.accenture.com/appliedintelligence
Athina Kanioura
Chief Analytics Officer, Applied Intelligence
Global Lead, Accenture
Fernando Lucini
Managing Director, Artificial Intelligence Lead,
Accenture UK and Ireland

References.
1
Accenture, “AI: Built to Scale”, November, 14, 2019. https://www.
accenture.com/us-en/insights/artificial-intelligence/ai-investments
2
Accenture, “Overcoming data paralysis when preparing for
AI”,February 4, 2019. https://www.accenture.com/us-en/insights/
digital/preparing-data-ai
3
Accenture, “An AI governance approach to support innovation”,
February 8, 2019. https://www.accenture.com/us-en/insights/
artificial-intelligence/ai-governance-support-innovation
4
5
Accenture, “Insuring autonomous vehicles”, https://www.accenture.
com/us-en/insight-autonomous-vehicles-opportunity-insurers
6
Accenture, “Disruption need not be an enigma”, February 26,
2018. https://www.accenture.com/us-en/insight-leading-new-
disruptability-index
7
Accenture, “Disruption need not be an enigma”, February 26,
2018. https://www.accenture.com/us-en/insight-leading-new-
disruptability-index
8
Accenture, “AI: Built to Scale”, November, 14, 2019.
https://www.accenture.com/us-en/insights/artificial-intelligence/ai-
investments
9
10
11
Accenture, “Empower the human+ worker”, February 7, 2019. https://
www.accenture.com/us-en/insights/technology/future-of-work
12
Accenture, “Reworking the revolution”, https://www.accenture.com/
gb-en/company-reworking-the-revolution-future-workforce
13
Accenture, “Reworking the revolution”, https://www.accenture.com/
gb-en/company-reworking-the-revolution-future-workforce
14
Accenture, “The big disconnect: AI, leaders and the workforce”, July
12, 2018. https://www.accenture.com/us-en/insights/future-workforce/
big-disconnect-ai-leaders-workforce
15
Accenture, “New, unexpected skills you need for AI success”,
August 22, 2019. https://www.accenture.com/us-en/insights/artificial-
intelligence/new-unexpected-skills-ai-success
16
17
18
Dr. Kai-fu Lee, AI Superpowers: China, Sillicon Valley and the New
World Order. Houghton Mifflin Harcourt, 2018.
19
Venture Scanner, “Artificial intelligence startup highlights Q2 2018”.
https://www.venturescanner.com/2018/08/30/artificial-intelligence-
startup-highlights-q2-2018/

20
Accenture, “What’s the best AI strategy: Build, buy, partner?,”
May 6, 2019. https://www.accenture.com/us-en/insights/digital/
best-ai-strategy-build-buy-partner
21
The Verge, “Apple reportedly buys AI startup with privacy-
conscious approach”, November 21, 2018. https://www.theverge.
com/2018/11/21/18106192/apple-privacy-ai-silk-labs-acquisition
22
Venture Beat, “Apple reportedly bought Silk Labs, a maker of
privacy-focused AI for devices”, November 21, 2018. https://
venturebeat.com/2018/11/21/apple-reportedly-bought-silk-labs-a-
maker-of-privacy-focused-ai-for-devices/
23
Accenture, “AI: Built to Scale”, November, 14, 2019. https://
www.accenture.com/us-en/insights/artificial-intelligence/ai-
investments
24
Accenture, “How to stop AI from reinforcing biases”, August
2, 2019. https://www.accenture.com/us-en/insights/artificial-
intelligence/stop-ai-reinforcing-biases
25
Reuters, “Amazon scraps secret AI recruiting tool that showed
bias against women”, October 9, 2018. https://www.reuters.com/
article/us-amazon-com-jobs-automation-insight/amazon-scraps-
secret-ai-recruiting-tool-that-showed-bias-against-women-
idUSKCN1MK08G
26
Accenture, “Explainable AI”, September 21, 2018. https://www.
accenture.com/us-en/insights/artificial-intelligence/explainable-ai
27
Accenture, “Critical mass: Managing AI’s unstoppable progress”,
September 17, 2018. https://www.accenture.com/us-en/insights/
digital/critical-mass-managing-ai-unstoppable-progress
28
Accenture, “An AI governance approach to support innovation”,
February 8, 2019. https://www.accenture.com/us-en/insights/
artificial-intelligence/ai-governance-support-innovation
29
30
31
Accenture, “Explainable AI”, September 21, 2018. https://www.
accenture.com/us-en/insights/artificial-intelligence/explainable-ai
32
DARPA, “Explainable Artificial Intelligence (XAI)”. https://www.
darpa.mil/program/explainable-artificial-intelligence
33
References.

About Accenture. About Applied Intelligence.
Accenture is a leading global professional services
company, providing a broad range of services and
solutions in strategy, consulting, digital, technology
and operations. Combining unmatched experience and
specialized skills across more than 40 industries and all
business functions — underpinned by the world’s largest
delivery network — Accenture works at the intersection
of business and technology to help clients improve
their performance and create sustainable value for their
stakeholders. With 505,000 people serving clients in more
than 120 countries, Accenture drives innovation to improve
the way the world works and lives.
Visit us at accenture.com
Applied Intelligence is Accenture’s approach to scaling
AI for our clients. We embed AI-powered data, analytics
and automation capabilities into business workflows to
accelerate time to value. Our expertise in defining end-
to-end strategy, combined with deep data infrastructure
capabilities, cognitive services and industralised
accelerators help smooth clients’ path to AI adoption,
extending human capabilities and supporting clients in
scaling AI responsibly. Recognised as a leader by industry
analysts, we collaborate with a powerful global alliance,
innovation and delivery network to help clients deploy and
scale AI within any market and industry.
Follow @AccentureAI
Visit accenture.com/appliedintelligence

Accenture-Ready-Set-Scale - AI.pdf

More Related Content

What's hot (20)

Similar to Accenture-Ready-Set-Scale - AI.pdf (20)

Recently uploaded (20)

Accenture-Ready-Set-Scale - AI.pdf