AI Readiness: Five Areas Business Must Prepare for Success in Artificial Intelligence

By Jessica Groopman
Edited by Jaimy Szymanski
Includes input from 27 industry leaders
September 2018
Five Areas Businesses Must Prepare for Success in Artificial Intelligence
AI Readiness

2
Executive Recommendations 3
Introduction 4
Definition of Artificial Intelligence
The Struggle for Readiness 5
The Five Areas of AI Readiness 6
Strategy 7
AI Transformation is an Extension of Digital Transformation
Strategic Approaches to AI
Lay the Foundation for AI Governance
Measuring AI’s Success
Key Questions to Ask
People 13
The AI Mindset
Identify Key Personae and Ready Each Group Accordingly
Address AI’s Limitations & Cultural Stigma
Best Practices
Data 22
Assess Enterprise Data Strategy
Ready the AI Data Feedback Loop
Leverage AI for Ongoing Enterprise Learning and Knowledge Management
Best Practices
Infrastructure 31
Assess Architecture Needs and Evaluation Criteria
Prepare Infrastructure
AI Software Solutions
AI Hardware
Best Practices
Ethics 39
Organization & Resources
Bias In, Bias Out
Transparency, Consent, and Data Privacy
Best Practices
Next Steps Towards Artificial Intelligence 48
About the Author 49
About Kaleido Insights 49
Research Methodology 50
Ecosystem Inputs 50
Acknowledgements 50
Endnotes 51
Table of Contents

3
Executive Recommendations
Although this research weaves best practices and considerations throughout each respective area of
readiness, companies interviewed surfaced a number of overarching areas of advice, based on their
AI implementation experiences.
Investment in AI = Investment in People: Early successes in the space show that the sum of
human + AI is greater than either alone, thus business preparedness and investment in AI requires
proportionate preparedness and investment in people.
Start small, pilot problems, fail forward. Businesses can realize immediate incremental value with
narrow pilot efforts that are focused on solving concrete problems; increase value by learning from
mistakes.
Data and AI preparedness must align with broader data and business strategy. Starting small
doesn’t mean thinking small, instead, every AI initiative must begin by assessing data, people, and
processes to support and align with broader business objectives, impacts and governance structures.
Lockstep coordination between technical and business subject matter experts must support every
phase of AI lifecycle, from design, research, and deployment, to management and optimization.
Trust and transparency must infuse enterprise AI. From evangelizing the limitations of AI, to
demonstrating the benefits, clear communication and user support fosters adoption and curbs fears.
Systems and process documentation and explainability are critical to mitigate risks of AI’s complexity
and autonomy.
Maturity won’t be defined by any single application. Advanced AI deployments will be marked by
the ability to infuse both user-facing services and interactions with back-end or enterprise process and
supply chain optimization.
Machine learning must translate to enterprise learning. Business leaders must create a culture of
learning and adaptability, by supporting people with skills development, workflows, and infrastructure
to learn from interactions and productize domain knowledge.

4
Introduction
The fundamental business question of “what do we do with all of this data?” is colliding with our endless
fascination for technological biomimicry. Artificial Intelligence (AI) is an unavoidable concept in the digital
age, if not for its pervasive application, then by virtue of the fact that it provides a technological solution to
information overload and the quest for context. Recent research finds some 80 percent of businesses plan to
invest in AI in the next three years.1
Even if the techniques and concepts underlying AI are more than 50 years old, the recent renaissance in
commercial environments is a result of a rather sudden convergence of three key trends: colossal data
generation; better algorithms; and faster processing hardware. AI is already yielding business impacts
across functions, including product development, process and workflow optimization, recommendation and
personalization at scale, risk mitigation, and cybersecurity, among scores of others. Enterprise interest in AI has
been compounded by its promises for efficiency in the form of speed, accuracy, agility, and access to insights
embedded in “dark data”—a reference to the estimated 80% of unused enterprise data.2
Meanwhile, the technology giants of the world are leading the charge, not only in the amount of data they
possess to train models, the frameworks and libraries they’re open-sourcing, or the firmware and hardware for
computing such colossal data, but in how AI-driven services are impacting consumer expectations. Every day
we hear new stories of AI used for innovation, for ill, for intrigue; new techniques emerge; and data volumes
and complexities grow.
Innovators and change agents see the hype of AI everywhere but struggle to know where to begin.
This Kaleido Insights report examines the need for AI readiness and introduces a framework for organizational
preparedness—not only of data and infrastructure, but of people and processes needed to deploy effective and
sustainable machine and deep learning programs. What follows is the culmination of extensive primary and
secondary research to surface the barriers, enablers, and best practices of leading organizations as they go
about deploying AI in business processes and services.
Definition of Artificial Intelligence
As often the case in emerging technologies, different market constituencies define terms differently. Artificial
intelligence (AI) evokes a special challenge (and emotionalism) as it is almost always compared to human
intelligence— a domain which also struggles to be defined.
Kaleido Insights defines AI as an umbrella term for the variety of tools and methods used to mimic cognitive
functions across three areas: perception/vision, speech/language; and learning/analysis. A machine’s ability to
“cognate” is supported by multiple approaches—machine learning, deep learning, natural language processing
(NLP), computer vision, and other existing and emerging techniques—multiples of which can be used at the
same time for a given use case. Kaleido Insights acknowledges discrete differences among techniques, but
for simplicity, this report will use “AI” interchangeably for applications involving machine learning and other
techniques mentioned above.

5
Although the race towards AI is on, the vast majority of organizations struggle to know where to begin.
An estimated 85% of organizations want to deploy AI, but have not.3
But, the gap between aspiration and
execution is wide, not only because data is a mess; the same study found less than 30% of enterprises
surveyed had any kind of AI strategy in place. Barriers to readiness, never mind deployment, span the
following areas:
FIGURE 1 BUSINESS, MARKET,AND SOCIETAL BARRIERS TO AI ADOPTION
Beyond the significant emotional, societal, and competitive challenges lie an array of barriers to practical AI
deployment, some of which are mentioned in Figure 1. Now— in what are no doubt the most nascent days of
commercial AI deployment— is the time for organizations to grapple with the business, economic, cultural, and
technical questions AI engenders.
Our research finds the most common stumbling blocks to deployment lie in data, technology, and talent—each
of which vary based on the nature of the business. More often than not, infrastructure is lacking to execute on
data; technology barriers around processing and integration stunt deployments; and people at all levels are
resistant to change. Achieving the benefits of AI requires companies deeply understand more than its utility; it
requires preparedness across five critical areas.
The Struggle for Readiness
Kaleido Insights, “AI Readiness: Five Areas Businesses Must Prepare for Success in Artificial Intelligence” September, 2018.

6
AI readiness isn’t merely a state of preparation, but a willingness and facility for implementation. This is a
particularly difficult endeavor in the context of AI, given we barely understand the implications of deploying
information systems to behave like biological systems.
FIGURE 2 THE FIVE AREAS OF AI READINESS
Today, most of enterprise discourse and activity around AI preparation is in the context of data preparedness.
Even if data is the paramount and most practical area “to ready” in order to deploy, it must be superseded with
intent, with human assessment for impact. Our research finds AI readiness results from practical assessment
across five areas: strategy, people, data, infrastructure, and ethics.
The Five Essential Areas of AI Readiness
Kaleido Insights, “AI Readiness: Five Areas Businesses Must Prepare for Success in Artificial Intelligence” September, 2018.

7
For most organizations, the trouble isn’t usually how to get started with AI, but knowing where. Not only can
machine learning and other AI techniques be applied across virtually every single industry, business function,
and workflow, it is subject to immense hype, over-inflation, and promise. From chatbots to facial recognition,
from automated reports to predictive maintenance, from procurement management to simulations for
decision-making, AI’s applications are vast. But while identifying use cases may intrigue, the starting point for
any sustainable strategy is never the technology. The starting point begins with identifying the problems and
objectives and setting goals.
AI transformation is an extension of digital transformation
In corporate environments, objectives tend to start with high-level mission statements and parse into specific
departmental, functional, product, team, and individual objectives in support. How can the organization
improve its product experience, its employee retention, lead generation, its partner channels, or accelerate
R&D? In this way, AI becomes an extension, not a bolt-on to existing digital transformation efforts.
This distinction is an important one, as companies risk following the siren hype of “building an AI strategy”
rather than evaluating how existing programs could function better and letting problems guide solutions.
While, obviously, there would be no AI without data, many companies struggle to understand the differences
and similarities.
1: Strategy
INTENDED AUDIENCE READINESS ELEMENTS
• Executives • Digital Transformation vs. AI Transformation
• Innovation Strategists • Strategic Approaches
• AI Leaders • Foundations for Governance
• Measuring Early Success

8
FIGURE 3 BRIDGING DIGITALTRANSFORMATION AND AI TRANSFORMATION
To date, digital transformation has been marked by digitization of information—a sort of “phase zero” for
bringing an organization’s programs and processes online. Born of the age of social media, and (often)
driven by marketing, digital transformation has evolved from external (customer-facing) programs to internal
and cross-functional. AI, on the other hand, is driven entirely by data. It is born of analytics, increasingly a
discipline within every department. Digital transformation has brought about a proliferation of points systems;
AI transformation benefits from systems integration. In both cases, people and culture aren’t just essential for
adoption, they are essential for activating insights and customer success in the digital age. At the end of the
day, thinking strategically about AI becomes synonymous with thinking strategically about data.
Strategic approaches to AI
Companies tend to design strategies based on current capabilities around data, where they sit in the value
chain, who is designing the strategy, and what practical limitations exist. Depending on who is driving the
program—business or technical leader—starting points can vary. Business leaders often look for areas of process
automation, visibility, or cost savings, while those more technically minded tend to first think bandwidth and
access to training data. Virtually every company we interviewed emphasized organizational structure as an
essential enabler of success—for short-term wins and over time.
Common Points of Origin for Enterprise AI
Bottom-up: Single spokes or business functions select a few workflows to which applying machine
learning will yield immediate benefits, then assess broader applications as early initiatives grow.
Top-down: Executive-driven efforts to transform the entire business into a fully automated hub in
which all interactions inform optimization and greater automation across every function, and related
products or workflows.
Cross-functional: Dedicated groups set up specifically to foster digital transformation efforts,
efficiencies, shared tools, and activate data and emerging technologies to enhance business
objectives are another point of origin for AI ideation and experimentation.
Kaleido Insights, “AI Readiness: Five Essential Areas Businesses Must Prepare for Success in Artificial Intelligence” September, 2018.

9
Bottom-up approaches emerge from vertical business functions and solve one problem at a time
The most common testing beds for AI projects are business functions or vertical LOBs, namely their data
analyst teams as they are most accustomed to dealing with lots of data. In some cases, these are experimental
efforts driven by internal “change agents” or data-savvy employees. In other cases, these efforts are the fruits of
mandates and investments from leadership to modernize or extract more data-driven insights, often pushing
specific business objectives (cost reductions, higher conversions, etc.). These groups have the greatest context
for where AI techniques could be applied for maximum immediate impact.
Most companies interviewed recommend a bottom-up approach to support understanding and achieve
relatively faster results. This is helpful for procuring more resources to expand efforts, as most organizations
are funded at a project level. Many companies also point to the value of starting small because there are
plenty of development resources available at low or no cost. Image recognition, captioning, natural language
processing (NLP), clustering, recommenders, and text processing are reasonable starting places for enterprises
given the variety of open source frameworks, APIs, and other industry tools and libraries available to all.4
“Start with big thinking, but work on one brick at a time,” recommends Toby Cappello, VP of worldwide
expert delivery services at IBM who recommends beginning with projects that leverage both the business
perspective with the technical development perspective. “Use the business perspective to break down big
ideas into smaller initiatives that have substantial business value; leverage the development perspective to
tool each one and aggregate learnings from the technical side.” For example, if applying AI to a help desk
environment, start by offering service agents a recommendation system; use their feedback to improve the
model, aggregate learnings, and build trust. Then, look at tackling a consumer-facing application for customer
support.
Top-down approaches are driven by central organizational hubs
Hubs are responsible for designing corporate strategies and resource allocation. They are more commonly
seen in tech companies or legacy enterprises aggressively investing in becoming a tech company. In these
early days of AI, relatively fewer AI projects are mandated through top-down approaches. Instead, corporate
hubs are the enablers for broader education, investment, governance, strategic metrics, and communications.
While broader top-down, multi-disciplinary strategies are important for longer-term product innovation and
coordination across functions, AI is unique in that tangible value can be achieved even in point solutions. “The
nice thing about this machine learning vs. other tech like enterprise resource planning systems or even IoT is
that you can pluck off a small part of a relatively low-risk business process and immediately drive efficiencies,
eliminate tedium, or improve accuracies,” shares Eric Berridge, managing partner at Bluewolf, an IBM-owned
cloud consultancy. While these efforts can prove value and indeed empower individual LOBs, they risk hitting a
proverbial value ceiling if not integrated.
Cross-functional programs are another common catalyst for enterprise AI programs
Unlike hub or spoke, the core role of these groups is to connect the two, specifically to incubate new concepts,
drive innovations and business objectives more efficiently, reduce technology and vendor redundancy, and
increase collaboration, interdepartmental workflows, etc. As AI programs evolve (from very narrow efforts in
single spokes, to more enterprise-wide learning and optimization), these liaisons are critical.
Over time, more advanced organizations work toward an integrated and shared platform that provides AI-
related services across the organization. Facebook’s FBLearner Flow, an enterprise-wide machine learning
platform for Facebook’s optimization, is in an infamous example of what this looks like in a technology platform

10
company.5
In other industries, it starts with people infrastructure. Capital One has a Digital Practice which
supports all LOBs by providing dedicated expertise in data science, mobile, APIs, and beyond to collaborate
with product leads on new solutions.
Wells Fargo recently developed an Enterprise AI Solutions group, whose charter is to identify both
opportunities (for new) and integration (of existing) AI applications across the entire company. “AI was starting
to emerge all over the business and we realized that we needed a dedicated group (beyond the Innovation
Group) to optimize Wells Fargo’s use of AI,” shares David Newman, SVP and strategic planning manager for
Innovation at Wells Fargo. This group:
• Takes inventory of all AI tools used
• Identifies areas of duplication and opportunities to standardize
• Surfaces best practices for idea-sharing and faster deployments
• Develops new partnerships and integrations with its technology organization
Starting small doesn’t mean thinking small: it means laying the foundation
for AI governance
While starting with smaller initiatives can yield relatively quicker wins, that doesn’t mean they are any less
strategic. While enterprises may have some governance processes in place to support digital, even single AI
initiatives introduce new considerations around workflows, onboarding, collaboration, program design, and
ongoing management. The broader subject of AI governance will be covered in a future report, but for the
purposes of readiness, companies we interviewed repeatedly articulate the need to conduct due diligence on
both people and processes associated with any AI initiative, no matter how small.
From a people perspective, user journey analysis is important for ensuring humans are comfortable using AI
in context, and to make sure AI is the best tool for solving the problem in question. To make AI work for the
business, you have to put it the path of existing work. The program has to be deployed in such a way that
doesn’t just invite feedback, but implements it in a manner that shows people how and why this way is better.
“It’s about using the interface to surface insights employees didn’t have, or had to spend searching for before,”
shares Gregg Spratto, VP of Operations with Autodesk. Einat Haftel, VP product management at Informatica
articulates why these early steps are strategic. “Lowering the learning curve and speeding up productivity with
data widens access, and the more people that can become experts in data management, the more value to the
organization.” Informatica includes the following in their onboarding process with customers:
• Human support: sitting down directly with users to gain feedback on what is and is not important to
show; where they can easily (or not so easily) make changes, and how
• Iteration to reduce false positives; in early phases of deployment, this is essential for both model
optimization (supervised learning involving users) as well as improving trust
• Tailor UX/UI: Interfaces and dashboards, displaying or ranking confidence levels for outputs surfaced
(e.g. 78% confidence these are product SKUs)
• Make requested changes to User Interface (UI) or user experience (UX)—doesn’t mean you have to
change the core model
From a process perspective, assess how AI will impact existing governance structures such as triage, queues,
feedback loops, employees’ access to data, etc. Assessing this early on informs where breakdowns are
occurring in the process and if automation would help or hurt.

11
For instance, if applying machine learning to routing particularly irate customers in a service context, you
wouldn’t want a chatbot to route them to the end of the normal queue, effectively putting them to the back of
another line.
As is the case with digital transformation efforts, top-down and ground-up efforts combined create the best
results. Initiatives that try to take a bolted-on, very technical, or altogether new approach will face greater
challenges and steeper barriers to adoption. The art here is aligning AI workflows with existing workflows and
behaviors to drive trust and adoption.
Measuring AI’s success must go beyond measuring financial impacts
It is important that early efforts yield tangible value, which naturally leads to the question of measurement.
Although small prototypes are useful for demonstrating ‘the art of the possible” to decision-makers in concrete
monetary impacts— e.g. cost reductions, lifts in conversion, pipeline optimization—companies we interviewed
reinforced the need to begin by measuring more than ROI.
“Even though we started with traditional cost metrics, we quickly realized we had to expand how we measured
value,” explains Gregg Spratto, VP of Operations with Autodesk. “Success should really be measured by the
ability to move an initiative from RD concept to deployment in production and scale in the real world.” This
means measuring the impact on people. Begin with metrics that assess value, often in the form of productivity
or sentiment, to user or function. For example:
• Time saved, spent (for agents, customers, analysts, etc.)
• Accuracy gained
• # successful outputs (e.g. cases resolved, conversions, transactions, referrals)
• Identification of fake, garbage, or fraudulent inputs (e.g. referrals, threats)
• Improved sentiment (e.g. shopping, sales, support, invoicing process)
• Improved Net Promoter Score (NPS)
• Feedback from customers or partners
• Impact on other existing metrics
Whether building or buying, businesses should beware of vendor promises involving metrics, warn some
vendors in the space. When you hear about other companies touting “97 percent improvements,” know there
are two general buckets for measuring AI model performance: General metrics such as ‘of the millions of
samples or questions asked; how many could be addressed given the ontology of the system; vs. Specific
metrics such as ‘for a specific case, e.g. customer request for an address change, 97 percent of the time AI can
resolve that one request.”
As efforts evolve and programs mature, measurement itself must too. As AI constantly learns, “lift” can be a
moving goalpost, and optimization in one area, (e.g. risk reduction) can impact another area (e.g. how to shift
from reactive risk analysis to proactive risk mitigation).
“You don’t want to automate processes that were crappy to begin
with. Good candidates for automation can sometimes be good
candidates for policy change.”
Greg Spratto, VP of Operations with Autodesk

12
Regardless of application, CX is true north
Just because AI is applicable across business functions does not mean businesses should de-prioritize
customer experience (CX) within strategic planning efforts. Impact on CX—ease, security, agility, access, speed,
meaningful personalization, improvement across user journey—is the paramount assessment of value for
any AI program, B2B or B2C. “No matter which lens you’re assessing readiness— of data, of infrastructure, of
employees-— the ultimate question must always be how will this impact the end-to-end customer experience?”
shares Gianni Giacomelli, Chief Innovation Officer at Genpact. Regardless of consumer, employee, or business
partner, humans are the architects and consumers of AI in every application.
Strategy: Key Readiness Questions
How will AI initiatives support existing digital transformation efforts and data strategy?
Where will design, development, and management for AI initiatives sit in the organization?
What are adjacent use cases to which we can apply learnings, metrics, or similar techniques?
How does our internal culture facilitate and enable innovation?
How will organizational structures support governance and scale of AI deployments
and collaboration?
What is the level of awareness we have in data inventory, pipeline, and integration?
How will ongoing efforts be driven from the ground-up but empowered top-down?

13
This concludes the first of five essential areas businesses must prepare for artificial
intelligence. While Strategy sets the course of an organization’s AI journey, it
is only a plan on a page without anchoring readiness across PEOPLE, DATA,
INFRASTRUCTURE, and ETHICS. Download the full research report to discover how
to prepare your organization for AI across each of these areas.
PURCHASE THE FULL RESEARCH REPORT
Report Purchase Includes:
• 54-page Research Report AI Readiness: Five Areas Businesses Must Prepare
for Success in Artificial Intelligence, including real-world examples, pragmatic
recommendations and best practices, frameworks to activate, and endnote
resources, sourced from more than 27 research interviews
• Twelve (12) high-resolution graphics and frameworks visualizing research findings
• Two (2) Downloads per purchase
• One (1) Complimentary Call with Lead Analyst, report overview discussion
To learn more about this report, Kaleido Insights, or our ongoing AI coverage:
• Visit the report page here
• Contact us directly with questions or package inquiries
• Find us at an upcoming event
• Look out for more AI coverage and research by subscribing to our newsletter
Download the Full Research Report

14
RESEARCH METHODOLOGY
This research was developed through extensive primary and secondary qualitative research methods. We interviewed 27
market influencers, vendors, and adopters between September 2017 – June 2018. We also conducted countless briefings
and discussions with industry innovators in the artificial intelligence, big data, cloud and related software and hardware
markets. Input or mention in this document does not represent a complete endorsement of the report by the individuals or
the companies listed herein.
ECOSYSTEM INPUTS
• Gregg Spratto, VP Operations at Autodesk
• Ashish Bansal, Senior Director Data Science Merchant Products Lead at Capital One
• David Newman, SVP and Strategic Planning Manager for Innovation at Wells Fargo
• Jeetu Patel, Chief Product Officer at Box
• Clayton Clouse, Senior Data Scientist at FedEx
• Tatiana Mejia, Group Product Marketing Leader at Adobe
• Gianni Giacomelli, Chief Innovation Officer at GenPact
• Toby Cappello, Watson and Cloud Platform, VP WW Expert Delivery Services at IBM
• Einat Haftel, VP of Product Management at Informatica
• Jeff Kavanaugh, Senior Partner at Infosys
• Helena Carre, EMEA Omnichannel Analytics Lead, Kimberly-Clark
• Julien Sauvage, Senior Director of Product Marketing, Einstein at Salesforce.com
• Kumar Srivastava, AI Fintech Expert, Stealth-mode AI start-up
• Jay Klein, Chief Technology Officer Sahar Dolev-Blitental, Director of Marketing at Voyager Labs
• Brian Schwarz, VP Product Management at Pure Storage
• Ian Collins, CEO at Wysdom.ai
• Tom Kraljevic, VP Engineering with H20.ai
• Mary Beth Ainsworth, AI and Language Strategist at SAS
• Robin Hauser, Documentary Film-maker, and Director of “Bias”
• Eric Berridge, Managing Partner at Bluewolf
• Veena Gundavelli, CEO at Emagia
• Girish Mutreja, CEO Founder at Neeve Research
• Jem Davies, VP ARM Fellow GM of Machine Learning Rhonda Dirvin, Director of IoT and Embedded Systems at ARM
• Srikanth Velamakanni, Group Chief Executive Executive Vice-Chairman at Fractal Analytics
• Alan Anderson, Director of Enterprise Solutions at IPSoft
• Ted Shelton, Chief Customer Officer at Catalytic
• Martin Stoddard, Principal Director of Webscale Services at Accenture
With additional inputs from Lyft, Western Union, Discover, Visa, Jet.com, Uber, Kia Motors, Microsoft, and Oracle
ABOUT KALEIDO INSIGHTS
Kaleido Insights is a research and advisory firm focused on the impacts of disruptive technologies on
humans, organizations, and ecosystems. Our industry analysts provide business leaders with clarity
amidst a fragmented technology landscape. Kaleido advisory relationships, webinars, speeches, and
workshops are grounded in research rigor, impact analysis, and decades of combined expertise.
Innovators are realizing that implementing each new technology isn’t enough, especially as business
models are disrupted. Keeping up is becoming more dif cult. Our mission is to enable organizations
to decipher foresee, and act on technological disruption with agility, based on our rigorous original
research, trends analysis, events, and pragmatic recommendations.
If you’re interested in building a relationship with our analysts, we’d love to hear from you. Please
email info@kaleidoinsights.com to start a conversation, or visit www.kaleidoinsights.com to learn
more about our offerings.

AI Readiness: Five Areas Business Must Prepare for Success in Artificial Intelligence

More Related Content

What's hot (20)

Similar to AI Readiness: Five Areas Business Must Prepare for Success in Artificial Intelligence (20)

Recently uploaded (20)

AI Readiness: Five Areas Business Must Prepare for Success in Artificial Intelligence