Vcii White Paper Value Chain Innovation Promise Ai

VALUE CHAIN INNOVATION:
THE PROMISE OF AI
WHITE PAPER | OCTOBER 2018
STANFORD VALUE CHAIN INNOVATION INITIATIVE
Table of Contents
I. Introduction....................................................................................................................................... 4
What is AI?.................................................................................................................................................................. 5
Current State of AI Adoption in Value Chains................................................................................................................... 6
II. Improvements Along the Value Chain..............................................................................................8–13

Value Chain Phases - Case Study Examples.....................................................................................................................9
III. Insights....................................................................................................................................14–17
Forms of AI-Driven Value Creation................................................................................................................................ 14
Strategic Considerations for Adopting AI....................................................................................................................... 15
The Human Factor..................................................................................................................................................... 15
Technology and Data.................................................................................................................................................. 16
Business Risks of AI................................................................................................................................................... 16
Potential Limitations................................................................................................................................................... 17
Looking Ahead........................................................................................................................................................... 17
IV.Conclusion ...............................................................................................................................18–32
Addendum: Case Studies
Summary of Case Studies............................................................................................................................................. 20
Design Phase Case: Stitch Fix........................................................................................................................................ 21
Source, Make, Store, and Delivery Phases Case: IBM ...................................................................................................... 24
Sell Phase Cases:
Adobe ...................................................................................................................................................................26
Pilot AI...................................................................................................................................................................28
Salesforce .............................................................................................................................................................29
Use Phase Case: OhmConnect....................................................................................................................................... 31
Citations ......................................................................................................................................33–35
Authors
Hau L. Lee, Principal Investigator Lauren Blake
Thoma Professor of Operations, Information and Technology Head of Product, Shiftsmart
Faculty Co-Director, Value Chain Innovation Initiative MBA ‘18, Stanford Graduate School of Business
Stanford Graduate School of Business
Sonali Rammohan
Haim Mendelson, Principal Investigator Evaluation and Learning Lead, Stanford Seed
Kleiner Perkins Caufield & Byers Professor of Electronic Business and Former Director, Value Chain Innovation Initiative
Commerce, and Management Stanford Graduate School of Business
Faculty Co-Director, Value Chain Innovation Initiative
Stanford Graduate School of Business
STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 2

Executive Summary
Artificial intelligence (AI) has been touted as a technology with great promise for value chain improvements. Many papers
have focused on the role of AI in manufacturing, logistics, marketing, and other value chain phases, and about organizations’
future plans to deploy the technology. Yet, survey data show that actual deployments are low, particularly in industries where
information technology is not core to the business. Many papers have focused on what companies are doing with AI and the
types of business benefits they expect or have realized. The focus of this paper is on the how: How companies are designing AI
strategies for their value chains, how they have successfully started using AI, how they have acquired data to train models, and
how they have scaled or intend to scale the technology. Our findings are based on case studies of six companies with varied
backgrounds; proceedings from an executive conference at Stanford Graduate School of Business on May 31, 2018, titled Value
Chain Innovation: The Promise of AI; and secondary industry research.
Our research has shed light on the questions executives are asking themselves to determine where AI will deliver value to their
organizations. We have learned that, while adoption and investment vary across industries, there is potential for AI to improve
many functions, including product design, manufacturing, delivery, retail, and marketing. How companies deploy AI solutions
involves defining a clear strategy around a business problem or opportunity in the value chain, determining an approach to
developing AI solutions, cultivating data and human capital, and carefully managing the many risks AI can pose.
By ensuring that customer value creation is the “North Star” throughout the process, companies can slowly leverage AI as a tool
to generate efficiency, product/process improvement, and even product/process innovation. This paper shares our understanding
of how AI can improve various stages in the value chain, provides insight on how to design business strategies that leverage AI to
create value, and raises important questions about the implications of an AI-driven future value chain.

I. Introduction
There is much hype around autonomous vehicles, as consumers look toward a future where cars and trucks drive themselves.
While full automation of transportation is many years away, car companies such as NIO are working toward high automation,
where a car can use artificial intelligence (AI) to operate without human input or oversight but only under select conditions. NIO
imagines transforming the notion of a car into a robot in which the driver manages driving on surface streets and then becomes
a passenger as the car enters a highway. NIO’s in-car, AI-driven virtual assistant NOMI would manage the passenger’s needs for
shopping, entertainment, and more, giving the passenger his/her time back. Padmasree Warrior, U.S. CEO and Chief Development
Officer, NIO, says this freedom would allow the passenger to be more “productive, playful, and peaceful.”1 One can imagine that
this type of highly autonomous car of the future could lead to the rise of new business models in which cars become extended
living rooms where people engage in various activities and purchase new products and services that haven’t yet been imagined.
As seen through the NIO example, advancements in artificial intelligence and machine learning have the potential to transform
the management and structure of global value chains. New technologies in critical areas such as natural language processing,
sensors, robotics, edge computing, machine and deep learning, and image recognition provide significant opportunities to improve
how products and services are designed, made, delivered, marketed, and used. Industry applications include targeted marketing,
dynamic pricing, product design, manufacturing automation, and supply chain coordination.
Still, it is difficult to distinguish between fact and fiction when it comes to the promise of AI. Companies are reporting significant
optimism about the potential for AI to transform their businesses. According to a 2017 survey of over 3,000 business leaders by
MIT Sloan Management Review and Boston Consulting Group, 85 percent of companies believe AI will allow them to sustain their
competitive advantage.2 Companies also expect to achieve increased revenues and decreased costs by deploying AI technologies.3
Yet, adoption rates have not kept pace with industry enthusiasm. The MIT/BCG survey found that only 23 percent of companies
have adopted an AI technology while only 5 percent of companies have extensively adopted AI.4 When new AI technologies have
been adopted, they have been primarily used in support functions such as customer service rather than in core functions.5
The slow pace of adoption in value chain phases such as design, manufacturing, delivery, and use is a reflection of numerous
barriers. According to executives, AI adoption is most often prevented by a lack of information technology infrastructure, talent,
proven technologies, and financial resources.6 AI experts also report data collection and preparation as significant challenges.7 If
the required data is not collected, disaggregated, or appropriately formatted, AI models cannot be built in the first place.8
Looking ahead, AI presents organizations not only with real challenges, but also with exciting opportunities to refine marketing
and pricing techniques, improve product design, production automation, supply chain coordination, and more. In this paper, we
will review key AI technologies, discuss notable applications and benefits of these technologies in various phases of the value
chain, and discuss their implications. Finally, we will present case studies to explore how value is being generated by AI-driven
innovations. We examine companies’ approaches to data acquisition, human talent management, and potential effects across
the value chain. Case study participants include Adobe, Salesforce, Pilot AI, IBM, OhmConnect, and Stitch Fix—a mix of solution
providers and users, and small and more mature firms from varied industries. Unless otherwise noted, the findings from these
companies come from one-on-one interviews.

What is AI?
Example: Definitions of AI
There is no standard definition of AI, and experts even disagree on the right
criteria to determine whether a technology should be considered AI. This is • AI is a collection of advanced
primarily because the relevant criteria keep evolving as machines succeed technologies that allows machines
on increasingly complex tasks.9 Consider a prominent example: playing the to sense, comprehend, act, and
television show game Jeopardy was considered too difficult for AI until IBM’s learn.”14
solution known as Watson beat human challengers in 2011.10
• “AI is that activity devoted to
Without a concrete definition, it is often difficult to discern between real AI making machines intelligent,
technologies and the growing AI hype. We use the following broad rule of thumb: and intelligence is that quality
AI can be considered any form of machine learning. Machine learning includes a that enables an entity to function
wide range of models that can be trained based on data, learning its own rules
appropriately and with foresight in
instead of receiving the rules from humans. These models are used by machines
its environment.”15
to complete tasks, often making predictions, based on new data.11
• “AI is the theory and development
Machine learning (ML) can take several forms. Two growing areas of machine
of computer systems able to
learning include deep learning and reinforcement learning. Deep learning, which
perform tasks normally requiring
uses complex multi-layered models to make predictions, is inspired by the firing
neurons in the human brain. Deep learning is well-suited for recognizing patterns human intelligence, such as visual
in images, text, and sounds and is behind state-of-the-art technologies for image perception, speech recognition,
classification and natural language processing tasks.12 Reinforcement learning decision-making, and translation
uses prediction models that are continuously retrained based on collecting between languages.”16
and analyzing new large data sets. This retraining is based on improving the
performance based on a reward function. For example, in the context of playing
chess, reinforcement learning would train models to pick the optimal move, play
millions of games of chess to generate data, and track its performance versus a
reward function of winning each game.13

Current State of AI Adoption in Value Chains
FUNCTIONAL AREAS OF AI DEPLOYMENT

Companies are experimenting with and adopting AI technologies across their value chains. Most of their efforts have been
invested in non-core support functions. In one survey of 3,000 professionals, over 25 percent of respondents reported having
adopted AI in each of the following support functions: IT, customer service, marketing, sales, finance, and accounting.17 However,
in contrast to how other new digital technologies have entered the enterprise, companies are also investing in AI for core
functions.18 For example, Stitch Fix, a personal-styling service, e-commerce company, deploys deep learning to understand a
user’s clothing style based on styles they like on Pinterest and also deploys machine learning to match stylists with users.
INDUSTRY ADOPTION LEVELS

As to be expected, the role AI plays in value chains today varies by industry. Figure 1 shows a snapshot of industry adoption
rates and investment plans. We can contrast the high tech and telecom industry and the construction industry as one example.
Over 30 percent of high tech and telecom companies have implemented at least one AI technology, whereas only 15 percent
of construction companies have pursued such advancements.19 High tech and telecom companies are more likely to focus on
using AI to improve their information technology while construction companies focus on using AI to improve operations and
manufacturing, reflecting each industry’s focus.20
Going forward, there is a risk that differing rates of adoption could widen the AI gap between technology-focused industries and
other industries. Over the next three years, AI spending at high tech and telecom companies is expected to grow over ten times
faster than spending at construction companies.21 While this is a significant difference, at a recent Stanford conference on AI, Jim
Sinai, VP of Product Marketing for Salesforce Einstein, remarked that because of the very fact that traditional industries such as
manufacturing, retail, and healthcare have legacy systems, certain types of AI can be deployed relatively quickly and add value.22
Therefore, while technology-centric industries may be investing in AI more heavily than other industries, opportunities still exist for
strong return on investment for targeted applications in more traditional industries.
One example of an industry that is in the middle range of AI adoption and investment growth is retail. Two of our case study
companies are involved in retail—Stitch Fix, which uses AI for functions like marketing, product design, and stylist-customer
matching, and Pilot AI, which sells visual recognition software for cameras used to track customer behavior in retail stores as
one of its applications. Both are examples of how new technology can be applied to more traditional industry. Overall, AI is
increasingly being used by retailers to recommend products, monitor face and hand gestures, employ virtual mirrors to track
shoppers’ movements, and for overall surveillance.23
The automotive and energy industries have seen high AI adoption rates, with moderate projected investment growth. Padmasree
Warrior of NIO U.S. believes that newer car companies are at an advantage with respect to AI technologies because they are
not faced with transforming a combustion engine-centric engineering culture—a challenge at more mature car companies.24 She
believes the biggest opportunities for innovation in the car industry are for electric and autonomous vehicles, both of which are
a focus for NIO. One of our case studies focuses on OhmConnect, a clean energy solutions provider. Since the company is not an
energy generator with heavy assets, it can more easily adopt AI solutions.
High tech is one of the highest adoption sectors, and we have three case studies focused on such firms: Adobe, Salesforce, and
IBM. These companies offer many software-based solutions, making it easier to evolve their AI technology over time. While AI
solutions may be easier to develop for such asset-light companies, other challenges such as change management may exist
in even the most technology-forward firms. Adoption rates may also be influenced by varying industry regulatory structures.
According to Tatiana Mejia, Head of AI Product Marketing and Strategy at Adobe, it is more challenging to deploy their AI software
solutions for professionals in industries such as financial technology and healthcare due to heavy regulation,25 which may explain
the moderate AI adoption in the healthcare sector.

FIGURE 1: SECTORS LEADING IN AI ADOPTION
Sectors leading in AI adoption today also tend to grow their invesment the most
Future AI Demand Trajectory*

Average estimated percent change in AI spending next three years, weighted by firm size**
13
Leading Sectors
12
Financial services
11 High tech and
telecommunications
10
8 Transportation and logistics
7
Travel and tourism Healthcare
6 Professional services Automotive and
Energy and resources assembly
5 Retail
Consumer packaged goods Media and entertainment

4
Education
3
2
Construction
1
Falling Behind
0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32
Current AI Adoption
Percent of firms adopting one or more AI technologies at scale or in a core part of their business, weighted
by firm size**
* Based on the midpoint of the range selected by the survey respondent
**Results are weighted by firm size
Exhibit from “How artificial intelligence can deliver real value to companies,” June 2017, McKinsey Global Institute, www.mckinsey.com.
Copyright 2018 McKinsey & Company. All rights reserved. Reprinted by permission.

II. Improvements Along the Value Chain
A value chain is structured around understanding customer demand and then supplying that demand. Efforts to deploy deep
learning and machine learning technologies have begun to improve demand-and-supply management in many industries. Figure
2 highlights how various AI technologies have improved key value chain functions. We will focus on how AI is generating value in
each of the following key phases: design, source and make, deliver and store, sell, and use. Note that, while “source” and “make”
are distinct activities, as are “sell” and “use,” these stages have been condensed for the sake of simplicity. We will use case
study examples to demonstrate how AI can be used to improve one or more value chain phases. Because AI is increasingly being
deployed beyond support functions into core business processes, we have placed greater emphasis on these core functions rather
than areas such as back-end IT support.
The figure below shows a simplified structure of a value chain. The case studies cited in this figure can be found in the addendum
section.
FIGURE 2 – VALUE CHAINS INVOLVE MANY PHASES FROM PRODUCT/SERVICE DESIGN TO USE
VALUE CHAIN PHASES
DESIGN SOURCE/MAKE DELIVER/STORE SELL USE
Example: Example: Example: Examples: Example:

Stitch Fix IBM IBM Adobe OhmConnect
Pilot AI
Salesforce

Value Chain Phases – Case Study Examples
In this section, we will highlight examples from our case studies illustrating how AI can improve each value chain phase. Many of
the companies interviewed discussed beginning with goals and a strategy, and then determining the role AI can play in meeting
the goals. This process is similar to deployment approaches for other technologies. Cascading down from goals, we examine how
AI solutions were deployed and the results. See the addendum section for the full case studies.
DESIGN PHASE – Stitch Fix

Customer feedback clearly has an important role in successful product and service design. In industries with strong digital
relationships with their customers, feedback gained from e-commerce interactions, digital marketing campaigns, and other means
can be valuable in understanding how to improve the product. Stitch Fix is a personal-styling-platform company that curates a
personalized box of clothing to be delivered to customers, who then choose which items to purchase or return. The company has
used AI to predict which clothing items customers will like and purchase and to match stylists with customers.
After seeing gaps in inventory and a need for better clothing choices, Stitch Fix made a strategic decision to begin designing its
own clothes to complement its existing inventory. The company now uses AI to identify popular clothing features and recommend
new combinations of those features to Stitch Fix’s in-house design team. This approach has led to 100 Hybrid Designs-branded
products thus far. Data sources include objective data on color, pattern, etc., and subjective data such as style. Computer
vision is used to analyze photos to extract additional data. From a human talent perspective, the company prides itself on using
human-in-the-loop AI, where people receive input from AI models but make the final decision. For design creation, this means AI
recommends potential designs, and designers edit the styles or discard them entirely. The company has a chief algorithms officer
who reports directly to the CEO, highlighting the importance of data science to the company. Still, there exists a balance between
data scientists and those with retail backgrounds who can do the more subjective analysis of styles. In the figure below, we
highlight how Stitch Fix and other companies are using AI to improve product design.
FIGURE 2A – DESIGN PHASE: AI CAN ENABLE NEW PRODUCT DESIGNS, IMPROVE THE SPEED OF ITERATION, AND
PERSONALIZE PRODUCTS
DESIGN PHASE
Goal Optimize Product Design Optimize Product Design Iterate Product Design Personalize Product
Based on Predicted Based on Desired Criteria Faster
Customer Behavior
Company/ Stitch Fix / In-house26 Airbus / Autodesk27 Boeing / Citrine28 Shiseido / MatchCo29
Tech Provider (acquired by Shiseido in
2017)30
What Was Done Created new clothing Improved the design of Found materials for airplane Sold makeup that matches
designs that are popular airplane partitions to be bodies that can be 3D the customer’s skin tone
with their existing sturdy yet light printed
customers
How It Was Used ML to analyze past Used reinforcement learning Used ML to analyze Used computer vision and
Done customer behavior and to apply patterns for slime Citrine’s proprietary material ML to analyze customer
clothing style characteristics mold and mammal bones to science database for faster photos and questionnaires
to design new clothes make airplane partitions material identification to customize products
Results Approximately 100 clothing 45 percent reduction in New materials identified in Custom foundation can be
styles designed with this airplane partition weight days instead of years delivered to customers in
process31 under 72 hours

SOURCE AND MAKE PHASE – IBM
IBM is a technology company providing software, hardware, cloud services, and cognitive computing to clients all over the world.
Watson is IBM’s AI platform, helping companies with business insights across a wide range of industries and use cases (e.g.,
talent acquisition, supply chain, customer engagement). In IBM’s own supply chain, Watson is now assisting in nearly every
process. For example, a supply assurance manager can ask: “What are the top five components with risk of supply shortage?”
and “Should we pull more of that specific part, or bring it over from another site’s inventory to mitigate the supply risk?” AI is
enabling IBM to have end-to-end visibility and insights from data across systems. The company is on a journey to predict, assess
and mitigate disruptions and risks, and enable a smart, resilient, and agile supply chain.
Today’s supply chains have access to internal, external, structured, and unstructured data. Supply chain professionals need real-
time contextual insights to predict risks and disruptions and take action. Matthias Graefe, IBM’s Director of Digital Supply Chain
Transformation, reports that Watson is now trained to provide insights in real time. This motivates professionals to interact with
the platform and share with Watson the decisions they are taking and why. For example, Watson learned how to identify product
changes shared by the development teams with the supply chain team via complex electronic documents. Watson identified parts
that were becoming obsolete in a couple of weeks and advised planners to procure new parts. The assistant was trained within
weeks based on around 300 annotated sample documents, identifying parts relationships that had not been previously known.
The number of sample documents needed to train the system can be much lower if the data is more structured in nature.
IBM now has a commercial Watson Supply Chain business unit, and Graefe’s team is accelerating the rate at which Watson
learns and provides deep supply chain insights. Transitioning to AI-enabled supply chain processes have yielded important lessons
on human-machine interaction. Initially, users did not pay enough attention to the recommendations provided via a conversation
panel. Then graphical information was coupled with the advice. With the relevant information linked to the question and answer,
users could form their own assessment and decide whether or not to overrule Watson. Watson observes user decisions, learns
from feedback, and provides playbooks on how to handle issues. It has started to prompt employees to look at likely follow-up
questions, acting as moderator in resolution processes. Graefe envisions a future in which Watson will be a personalized advisor
that learns from increasingly digitized data, interacts with users to clarify its understanding, and then engages in exponential
learning. One day, he says, you may be able to ask Watson “What should I work on today?”
FIGURE 2B – SOURCE AND MAKE PHASE: AI CAN IMPROVE OPERATIONAL EFFICIENCY, INCREASE ASSET UPTIME
AND MINIMIZE INPUT PRICE FLUCTUATIONS
SOURCE AND MAKE PHASE
Goal Optimize inventory levels, reduce surplus, Increase Machine Uptime Minimize Input Price Risk
excess, scrap
Company/ IBM / In-house (Watson) MidAmerican Energy Company / Big River Steel (steel
Tech Provider Uptake32 manufacturing startup) /
Noodle.ai33
What Was Done Provided alerts and recommendations to Anticipated wind turbine breakdowns Hedged the difference between
supply chain professionals on obsolete and minimized their occurrence scrap steel input and finished-steel
parts resulting from upcoming product output pricing
changes
How It Was Built an AI-powered assistant tool Used ML on data collected from wind Used ML to estimate prices of
Done for the supply chain team. The tool turbine sensors to predict near-term scrap steel and finished steel from
analyzes structured and unstructured machine failures historical prices, projected demand,
data to answer a wide range of employee and estimated wear and tear on
questions. Example question: Should we their factory (captured by 50,000
order more of a specific part? sensors)
Results Lower inventory levels, reduction of Realized $250,000 in savings in first Increased overall profitability
scrapping cost week; estimates $3.3 million savings of the mill
per year

STORE AND DELIVER – IBM
To capture the short-term impact of AI investments, Graefe’s group keeps track of the number of questions posted and answered,
questions that receive a “thumbs up” rating, and the net promoter score (NPS) among employees.
Graefe see IBM’s Supply Chain team just reaching the starting point of exponential learning. Strong conversational speech
interfaces for the assistant tool encourage more employee interaction with the tool and, as a result, capture better training data. It
makes it easier for employees to provide the “why” behind decisions, which is so critical for Watson to learn with the team.
In Figure 2c, we highlight how IBM and other companies are using AI to improve storage and delivery.
FIGURE 2C – STORE AND DELIVER PHASE: AI CAN REDUCE DELIVERY TIMES, IMPROVE FUEL EFFICIENCY OF
DELIVERY VEHICLES, INCREASE WAREHOUSE EFFICIENCY, AND REDUCE INVENTORY
STORE AND DELIVER PHASE
Goal Avoid Weather-Related Reduce Delivery Times Improve Fuel Increase Warehouse Reduce Inventory
Delays Efficiency Efficiency
Company/ IBM / Watson (in- DoorDash (food Vale (Brazilian Gap / Kindred AI37 Otto (German grocery
Tech Provider house)34 delivery startup) / mining company with delivery service) /
Starship35 railroad network) / Blue Yonder38
GE Digital36
What Was Provided an early Avoided traffic by using Automated train Sped up assembly of Decreased inventory
Done warning for weather robots that drive on braking and e-commerce orders by improving
events (e.g., hurricanes) sidewalks accelerating with robots forecasts and
and recommend coordinating
alternative plans automated ordering
How It Was Used ML to anticipate Used computer vision Used computer Used reinforcement Used deep learning
Done timing of the weather and AI to navigate vision and machine learning and to extrapolate trends
events impacting robots through an learning to optimize deep learning to from historical
their factories and urban environment train maneuvering transition robots transaction data
warehouses based on relying on data from based on sensor from being human-
Weather Company data cameras and ultrasonic data, weather, and operated to working
(acquired by IBM in sensors railroad routes autonomously
2016)
Results Saved money and 15-30 minute delivery 4 percent fuel Plans to expand 20 percent reduction
“continued to delight times savings its six robot pilot in surplus stock
customers” program

SELL PHASE – Pilot AI, Salesforce, and Adobe
With the growth of omni-channel and online retail, and the emergence of digitized physical retail shopping through stores like
Amazon Go, we have seen a large increase in data available to strengthen insights into customers. One example is Pilot AI,
which offers computer deep-learning vision software for compute-constrained cameras. One solution they offer is to analyze video
footage in retail stores to count and identify customers’ demographic characteristics and create heat maps of customer movement.
Given its low cost per store, investment in the Pilot AI solution can be recovered by small increases in in-store sales.
Salesforce is an $88 billion industry leader with an AI solution called Einstein to inform recommendations such as prioritizing
sales leads and estimating marketing performance. Jim Sinai, VP of Product Marketing for Einstein, emphasizes that successful
AI enterprise implementations require new business processes that provide the right data to their software instead of repurposing
existing data. He says “[Most companies are] thinking about it 100 percent the wrong way. What you should think about is how
you architect a workflow where executing that workflow trains the model.” The company foresees that advanced voice capabilities
will make the user experience more conversational, enabling access to data insights in an informal and intuitive way. The
combination of a clear AI strategy, a process-based approach, and a human-centered model is critical to the company’s vision.
Adobe is another high-tech leader using AI to improve the customer experience throughout its software offerings. Adobe Sensei
uses AI and machine learning to power the company’s solutions across its Creative Cloud (e.g., eliminating manual, time
consuming, and repetitive tasks for creative professionals, such as photo and video editing), Experience Cloud (e.g., helping
marketers and advertisers deliver relevant, personalized experiences in real time and make content recommendations to reach
the right customer at the right time), and Document Cloud (e.g., searching and understanding large amounts of content at a deep
level, like the sentiment behind documents, and producing clean, secure, sharable PDFs from paper). According to Tatiana Mejia,
Head of Product Marketing and Strategy, seamlessly integrating AI into Adobe’s solutions is a key factor in Sensei’s success.
Mejia notes, “Most of our customers aren’t aware that they’re leveraging AI.” The company conducts deep AI research in-
house, conducts extensive online training, and has learned that AI talent must be dispersed through the organization due to AI’s
importance and evolution. Success is evaluated based on the extent to which Adobe is addressing its customer pain points.
Figure 2d highlights how Pilot AI, Salesforce, Adobe, and other companies are using AI to improve selling.
FIGURE 2D – SELL PHASE: AI CAN ANALYZE CONSUMER BEHAVIOR, IMPROVE LEAD CONVERSION, PERSONALIZE
MARKETING, AND INCREASE UPSELLING
SELL PHASE
Goal Analyze Store Traffic Improve Conversion Improve Efficiency of Personalize Increase Upsell
Ad Development Marketing
Company/ Large retailers / U.S. Bank / Adobe / In-house Starbucks / In-house40 Square / In-house41
Tech Provider Pilot AI Salesforce39
What Was Tracked in-store Provided wealth Eliminated manual Optimized food, Sold payroll tools and
Done customer behavior for managers with photo editing work beverage offers in their loans to existing small
specific demographic product suggestions mobile app and medium business
profiles for their clients customers
How It Was Used computer vision Used ML to identify New AI-powered Used AI to customize Used ML to predict
Done and deep learning to clients likely to convert feature assigned the app based on new purchases
recognize people in based on their unified metadata word labels customer’s order based on customer’s
video footage from customer database to photos history, along with purchase history and
security cameras and historical data current weather, date, product usage
and time
Results Increased sales 2x increase in Reduced photo search 3x increase in Increased sales
conversion of top- time from 10 minutes response rate of higher margin
ranked leads to approximately 30 with individually products
seconds personalized offers

USE PHASE – OhmConnect
OhmConnect, founded in 2013, works with energy providers to incentivize consumers to save energy. The company obtains
information on customer energy use from online utilities accounts, pays consumers to reduce electricity usage during peak periods
when more “dirty” energy is used, and sells the energy back to utility companies, thereby reducing overall pollution. Using proven
machine learning, OhmConnect predicts how customers will respond to requests to save energy and thus estimates the energy
supply needed. Going forward, the company plans to transition to deep learning to gain a more in-depth understanding of relevant
energy demand and supply. The company recruits software engineering generalists, with an eye on hiring specialists as it grows.
Given the unpredictability of AI advancements and business applications, the company has diversified its analytics efforts so it
can select elements of various strategies as needed over time.
Figure 2e highlights how OhmConnect and other companies are using AI to improve the ways in which customers use products
and services.
FIGURE 2E - USE PHASE: AI CAN INCREASE EFFICIENCY OF CONSUMER PRODUCTS/SERVICES, AUTOMATE CUSTOMER
SERVICE, AND PREDICT MAINTENANCE
USE PHASE
Goal Increase Efficiency Automate Customer Service Anticipate Maintenance
Company/ OhmConnect / In-house UPS / Microsoft42 U.S. Department of Defense / C3 IoT43

Tech Provider
What Was Done Improved the efficacy of their energy Provided detailed delivery updates Received an early warning for fighter
saving campaigns through their chatbot jet maintenance
How It Was Used ML to optimize consumer Used the natural language processing Used ML to anticipate engine
Done segmentation for campaigns based on and ML capabilities embedded in the problems based on sensor data and
household-level energy usage data, Microsoft Bot Framework to manage maintenance logs
aggregated demographics data and conversations
weather forecasts
Results Increased user participation in Managed 200,000 customer Improved asset availability; reduced
campaigns conversations in first eight months maintenance costs

II. Insights
It is clear from our research that, like the Internet of Things, 3D printing, and automated delivery vehicles, AI can be a powerful
tool. If strategically deployed, it has the potential to drive many forms of value creation. In this section, we share the strategies
and key factors to consider in order to assist AI-driven value creation, and we look ahead to the future implications of AI-powered
value chains.
Forms of AI-Driven Value Creation

We have observed examples of AI improving precision and speed in many value chain phases such as marketing, customer
insights, design, manufacturing, delivery, and retail. By improving the precision and speed of these functions, we find that the
following key improvements are possible.
• Process efficiency: AI automation often improves repetitive processes that are not enjoyable or challenging for humans. For
example, Abundant Robotics44 has designed AI powered automated machines to harvest apples, reducing the amount of labor
needed. Since manual apple picking is physically demanding and labor supply is low, farms have an incentive to improve this
process.
• Process enhancement: AI can also enhance existing processes, leading to better outcomes for users. For example, Salesforce
Einstein can prioritize leads for a salesperson and hide those that have a very low likelihood of converting to a sale. This
enhances the sales process, allowing executives to devote high-quality time to high-value leads.
• Product or service innovation: AI can enable the creation of new products and services. As seen with Stitch Fix, AI can power
new product design, using data from multiple sources to predict styles that will resonate with customers.
FIGURE 3 – SUMMARY OF VALUE DRIVERS: AI CREATES VALUE BY DRIVING PROCESS EFFICIENCY, PROCESS
ENHANCEMENT, OR PRODUCT/PROCESS INNOVATION
SUMMARY OF VALUE DRIVERS
Design Source/Make Deliver/Store Sell Use
Optimize product design Increase efficiency Reduce delivery times Analyze store foot traffic Increase efficiency
basd on predicted
customer behavior
Optimize product design Increase machine Improve fuel efficiency Improve conversion Automate customer
based on desired criteria uptime service
Iterate product design Minimize input price risk Increase warehouse Improve conversion Anticipate maintenance
faster efficiency
Personalize product Reduce delivery times Reduce inventory Personalize marketing
Optimize input purchase Avoid weather-related

timing delays
Process Efficiency Process Enhancement Product or Service Innovation

Strategic Considerations for Adopting AI
Organizations must consider many factors when designing an AI strategy: identifying problems and opportunities, what type of AI
to deploy, and how to introduce and eventually scale the application by developing human talent and a strong execution strategy.
In this section we discuss some key questions organizations should consider when designing their AI strategy and determining
how to execute it.
What problem/opportunity have we identified? In order to generate value creation from AI, those interviewed stressed the
importance of first having a clear value chain strategy. Organizations must be very clear about the value chain problem they need
to solve or the opportunity they want to pursue. Once the area of focus is well understood, according to those we interviewed,
organizations should first design ideal workflows and then gradually deploy AI to effectively support and improve these workflows.
“People are solving one global problem–you can take AI to any workflow.”
—Jim Sinai, VP of Marketing, Salesforce Einstein
How should we develop our AI capabilities? AI solutions can be achieved through building proprietary technology or through
leveraging external technology. Some companies rely on a mix of both approaches.45 Salesforce has designed much of its AI
technology in-house, but also augments it with acquisitions. Clean energy broker OhmConnect is designing machine learning
algorithms in-house, but also relies on third party plug-in software and Amazon Web Services.
Should we diversify our technology? Because the pace of technological advancement and research breakthroughs in the field of
AI is brisk, it is not always clear which technology will become widespread. As a result, organizations may consider deploying
multiple AI strategies and models to be prepared for a few to ultimately prove useful.
Should AI be front and center, or invisible to the user? Several of the software providers we interviewed have chosen to make AI
invisible to their customers. Adobe, for example, has dozens of AI models behind its solutions, but doesn’t directly brand its AI
technology within its product-user interface. Adobe’s goal is to provide intelligence assistance and make solutions frictionless to
adopt. Salesforce uses a similar approach and considers AI just another type of software embedded in its solutions. Providers may
need to consider whether branding their AI tools offers specific value in customers’ minds, or whether their customers value the
solution regardless of the technology used.
“We want to make good AI invisible...There are dozens of features powered

by Adobe Sensei inside solutions.”
—Tatiana Mejia, Head of AI Product Marketing and Strategy, Adobe
The Human Factor
Acquiring AI talent
Whether an organization decides to primarily outsource AI solutions or develop them in-house, significant challenges exist
around building talent. Many companies do not have sufficient AI expertise, and hiring top AI talent is extremely competitive.
A report by analytics provider Teradata found that 34 percent of companies surveyed cited a lack of in-house AI expertise as
a roadblock for AI adoption.46 Even if a firm has chosen to outsource AI, it can be difficult to find the right vendors. Twenty-six
percent of companies surveyed by the McKinsey Global Institute believed the market lacked AI products that were relevant for
their business.47

Other domain experts matter
Regardless of the amount and level of AI talent in-house, those interviewed expressed the importance of including domain experts
not necessarily trained in AI on AI-related teams. For example, at Stitch Fix, retail experts are involved in classifying clothing,
which is considered a subjective exercise. This information is then converted into objective data which can be used in algorithms.
While many companies are more focused on deploying AI to determine how much inventory to carry, Stitch Fix’s approach
ensures that human input and subjective data help determine what inventory to carry. Padmasree Warrior of NIO U.S. also
stresses the importance of having non-AI staff with deep subject matter knowledge in a variety of areas and an affinity for learning
about other subjects.
“My biggest challenge as CEO is to have machine learning data scientists,

supply chain, user interface designers and people from every discipline
working together...This keeps me up at night.”48
—Padmasree Warrior, U.S. CEO and Chief Development Officer, NIO
Human-machine collaboration
Regardless of the type of AI solution deployed, ensuring that human judgment is always involved is critical. Problems can arise if
AI algorithms make strictly profit-maximizing decisions. For example, unless a human intervenes, consumers who fall outside of
average foot sizes may be overlooked by shoe retailers using AI to optimize inventory. In applications where work safety is critical,
such as self-driving cars and in the defense industry, experts argue that human judgment should always be closely coupled with
AI. For example, in a future where self-driving cars become commonplace, in certain dangerous conditions it may be important for
the car to warn owners to take over and if they don’t, pull off the road and stop.49
Technology and Data

Strong AI ecosystems are needed to scale: To successfully scale AI, many elements of the AI ecosystem need to be in place,
such as a strong Internet of Things (IoT) infrastructure, back-end IT expertise, top AI talent, and knowledgeable non-technical
employees who embrace AI opportunities and work alongside AI experts. For example, in the home energy market, providers
may envision the real-time matching of supply and demand via AI to achieve an optimally efficient market. This type of real-time
matching may only scale once home IoT devices become more dependable and accurate and can seamlessly report energy usage
and demand.
Data and model training: A significant amount of high quality data is imperative for training various AI models. For many
companies, the necessary data is currently out of reach. Jonathan Su, CEO of Pilot AI, has observed that companies often have
more data than they can actually use. This can be because companies do not have the right data infrastructure or usage rights. As
suggested by Jim Sinai, one solution is to establish business processes to collect this data. For example, IBM captures important
data on supply chain best practices by having their top employees train AI tools directly and by encouraging decision-making
conversations to be recorded to further train models.
Business Risks of AI
Managers are concerned about many risks involved in adopting AI solutions, including regulations, counterfeiting, lack of
transparency, and privacy. Companies applying AI in heavily regulated industries such as healthcare and banking may encounter
roadblocks in accessing data protected under strong privacy rules. Counterfeiting is a concern for companies using AI to design
digital solutions such as advertisements and marketing campaigns. Tatiana Mejia notes that AI technology such as Adobe Sensei
can incorporate digital signatures or watermarking to show a photo’s authenticity. “It’s a responsibility for the companies that
create these [technologies] to think about different ways they can be used.”50

Machine learning solutions provided by some companies can offer explanations for why a solution was determined to be optimal.
Yet AI solutions providers are concerned that model decisions made by deep learning and reinforcement learning systems cannot
always be understood and justified, making decision-making and accountability more opaque to the user. There is the risk that
error rates in AI models may have harmful consequences. If an AI model targets the wrong customer for a marketing campaign 5
percent of the time, there are minimal harmful effects. Yet, if a defense contractor uses AI visual recognition systems to identify
a noncombatant in a battle, being wrong 5 percent of the time is a risk the contractor may not want to take. All of these risks
can be amplified deeper in the supply chain, since supplier networks may not be as well-equipped to manage them compared to
better-resourced buyer firms.51
Companies must also consider the comfort level of consumers for new AI advancements. Even if a store can use facial recognition
software to identify customers as they approach, retailers believe they don’t want to be greeted by name as they walk in. As Jon
Su of Pilot AI notes, “The consumer values privacy, and we as an analytics provider want to make sure we treat the consumer’s
privacy with the utmost care and consideration alongside our customers.”52
Potential Limitations
Questions arise as to whether companies with a heavy physical asset infrastructure or physical products that take time to
produce are at a disadvantagein adopting AI compared to asset-light or virtual product companies. At Stitch Fix, AI is used to
gain insights into demand, make sophisticated product recommendations, and even design clothes. However, the company still
has a long production cycle relative to a software company, and there may be limits to the gains in efficiency or product/process
enhancements generated by AI. This raises interesting questions. Can future developments in AI help address long production
cycles? How can technologies such as 3D printing be combined with AI to enable faster supply chain cycle time?
As with other technologies, change management within an organization introducing AI can be quite difficult. Even with a targeted
strategy and a clear approach to acquiring or developing AI tools and data, if leaders do not manage the transition to AI well,
value creation may never be achieved.
Looking Ahead
Going forward, digitization of the modern economy will cause dramatic shifts in how customers use products and services, how
businesses manage their value chains, and the role of workers. With respect to customers, Padmasree Warrior of NIO U.S. has
already observed a digital transformation driven by the internet and an “app economy.” Warrior now expects that industries will
shift to a physical transformation—a post-app world where physical objects such as cars, drones, and other robots will have
virtual assistants that engage with the user.53 Stanford conference speakers discussed a future where customers enjoy more
natural interactions due to advancements in language processing. Combining AI with IoT technology shows great potential for
fixed physical objects, such as refrigerators, to interact with users, and for moving such objects autonomously. In both cases, the
increasingly rich supply of data will be useful to train models to improve over time.
For businesses managing evolving value chains, there is an expectation that more processes will take place in real time. Bo Zhai
leads emerging technology investments at U.S. Alibaba, the Chinese e-commerce and internet conglomerate. He believes that if
e-commerce firms can use AI to help them find more efficiencies in the back office and supply chain, ordering merchandise may
become even easier for consumers and delivery times will decrease. “It all comes down to shortening the lead times at reduced
cost,” Zhai said.54 Through process transformation, AI has the potential to upend entire industries. Russ Altman, who is professor
of bioengineering, genetics, medicine, and biomedical data science and of computer science at Stanford University and a member
of the standing committee of the Stanford One Hundred Year Study on AI, believes that future AI-based systems will lead to a
reinvention of medicine. AI systems will be able to analyze information collected from sensors placed on individual patients, social
media, medical records, and genomic data to help doctors design health policy and prevention strategies, diagnose illness, plan
treatments, and determine prognoses.55

There also continues the age-old debate of automation as job creator or
Thoughts on the Future of AI job destroyer. While it is unclear what the net effect of AI on jobs will be,
it is clear that AI will lead to new types of human-machine interactions.
“Just as electricity transformed almost Completely automating some worksites, such as a brick-and-mortar
store, would require such a huge redesign and financial investment that
everything 100 years ago, today I actually
companies may be better off using automation for only select functions
have a hard time thinking of an industry that
and retaining humans to perform the rest.
I don’t think AI will transform in the next
several years.”
—Andrew Ng, Stanford Adjunct Professor, IV. Conclusion
Coursera Co-founder, AI Fund Founder56
Given the speed of technological advancements and continued
breakthroughs in AI research, it is unclear which technologies will generate
“The demand for machines to explain the most value and encounter widespread adoption in the near term.
and justify answers in terms humans can Yet, from our case study explorations and conference discussions, it is
understand will grow with our desire to clear that AI can deliver process efficiencies, process enhancements, and
leverage computers in the process of human product/service innovation, leading to significant transformations. Some
decision making, especially in areas where the estimate that the technology already exists to automate nearly half of
reasons matter...The future of AI will need to all global work activity.59 Process efficiencies from automation alone will
influence job functions, departmental structures, supplier relationships,
focus on making sense of the world and being
and much more. As technologies evolve, organizations should keep a
able to answer the Why?”
constant focus on how to create value for customers, with AI being one
—Dave Ferrucci, Creator of IBM’s Watson, tool to deploy toward that goal.
Founder and CEO, Elemental Cognition57
“The future is bright. People can act more like

people.”
—Daragh Sibley, Director of Data Science,
Stitch Fix58

ADDENDUM – CASE STUDIES
We selected a broad range of companies that highlight the use of AI
within one or more phases of the value chain. The following section
summarizes what we learned about how each company has deployed
AI, either within the firm, for customers, or both.

FIGURE 4 - SUMMARY OF CASE STUDIES
SUMMARY OF “PROMISE OF AI” WHITE PAPER CASE STUDIES
Design Source, Make, Sell Use

Store, and Deliver
Stich Fix IBM Adobe Pilot AI Salesforce OhmConnect
AI AI helps Stitch Watson, IBM’s AI Adobe Sensei Pilot AI offers a Salesforce’s OhmConnect uses
Application Fix understand platform, helps accelerates the computer vision Einstein creates ML technologies
Strategy its customers, companies make adoption of AI software platform custom AI models to increase the
make product sense of their data across their with proprietary for each customer. amount of energy
recommendations, to make better software products, technology to make Most users access saved through
and purchase decisions.IBM’s addressing AI model training the insights through segmenting
inventory. Stitch internal supply challenges for more efficient. In various features its users and
Fix uses “humans- chain team built Adobe’s domain retail use cases, the in the existing personalizing
in-the-loop” an AI-powered expertise areas: platform can track software requests to use less
to audit AI’s “assistant” with content, creative, in-store activity energy
recommendations Watson’s APIs that and marketing
answers employee
questions
Data In addition to The assistant tool Sensei utilizes Pilot AI’s platform Einstein takes Because users
Analysis traditional metrics, relies on domain trillions of content analyzes video advantage of all connect their online
data is gathered expertise along and data assets, camera footage, of a customer’s utilities accounts
from fashion with data from from high- which often comes data in Salesforce to the platform,
experts and the ERP system resolution images from existing and eliminates a OhmConnect
computer vision and Watson. Top to customer clicks security cameras significant amount collects detailed
analysis of clothing employees invested that Adobe already of manual data user behavior
images significant time has in its domain prep and household
showing Watson expertise areas electricity
how to answer consumption data
questions
Human Stitch Fix’s large All supply chain Adobe Research Pilot AI has Salesforce helps Software
Talent data science team professionals are develops new PhDs from clients with change engineering
is encouraged to encouraged to technologies math, computer management as generalists work on
work on novel help maintain the and publishes science and they adopt new various AI projects
problems across assistant tool by academic papers. other disciplines processes, which
the business grading its answers Adobe has online developing leading- can be easier
and recording their training to teach edge AI solutions to manage with
conversations other technical smaller groups
employees about AI
Longer-term AI can enable new Solutions that AI solutions to improve customer insights, targeted marketing, Growing insights
Implications product designs improve supply and creative ads can uncover new product ideas and new on how customers
and a more efficient chain precision customer segments, improve customer loyalty, and strengthen use products and
design cycle. and the speed of the overall customer relationship services can lead
With growing decision making to new products
opportunities to can shorten better targeted to
use sensor and production cycles, customer needs,
other data on improve product improving customer
how customers quality, and reduce satisfaction
use products, this costs
information can be
fed into product
design, creating
a more effective
design cycle

DESIGN PHASE CASE: Stitch Fix
Business
Founded in 2011, Stitch Fix is a personal styling platform that uses a unique e-commerce model; stylists curate a personalized
box of clothing items to be delivered to each customer. The customer chooses which items to purchase and which to send back
to Stitch Fix. The success of this publicly traded company, valued at approximately $1.5 billion at the time of their IPO, depends
on predicting which clothing items customers will like and ultimately decide to purchase. AI plays a critical role in making these
predictions, helping the company determine which clothes to keep in inventory and prioritizing clothing items that a stylist should
select for a particular customer.
“We leverage data science to deliver personalization at scale, which we

don’t see anywhere else in retail. Data science is not just part of our culture
—it is our culture and it’s woven into every aspect of our business.”
— Cathy Polinsky, Chief Technical Officer, Stitch Fix60
To fill gaps in inventory and produce better clothing, Stitch Fix made the strategic decision to design its own clothes in addition
to purchasing clothes from existing vendors. The company uses AI to identify popular features of existing clothing items and
recommend new combinations of those features to Stitch Fix’s in-house design team. The company has already created
approximately 100 new clothing items using this approach.
The company also anticipates that AI will become involved in more far-reaching areas of their business over time. Stitch Fix has
already seen success using AI to match stylists to each client and believes more successes are yet to come.
AI technology
Stitch Fix is a sophisticated user of many AI technologies and applies them throughout their value chain. Below are some
prominent examples.
FIGURE 5 – STITCH FIX AI APPLICATIONS
AI Technology Deep Learning Machine Learning Machine Learning
In-house application at Describing the user’s style Matching stylists and Identifying common
Stitch Fix by measuring the similarity users based on any characteristics across
between his or her favorite prior interactions, stated popular clothing items
Pinterest photos and pictures preferences, and inferred and recombining these
of items in inventory characteristics characteristics to create new
items
Source: http://algorithms-tour.stitchfix.com (accessed May 17, 2018)

Much of Stitch Fix’s AI success has been attributed to “human-in-the-loop AI,” where humans receive inputs from AI models
but ultimately make the final decision. The company first used human-in-the-loop AI for curating items based upon AI
recommendations. This approach is also used in creating designs; AI recommends potential designs, and designers edit the
designs or discard them entirely.
Stitch Fix believes they receive the best of both machine and human intelligence with human-in-the-loop AI. With this
approach, computers can excel at running large, complex numerical analyses while humans can excel at evaluating aesthetics
and answering innate questions. The human factor also allows AI to make more risky recommendations since any outrageous
recommendations will be intercepted before they impact customers.While Stitch Fix expects human-in-the-loop AI to evolve over
the next five to ten years, the company expects to maintain sizable roles for both humans and computers.
“Machines and humans have very different talents when it comes to

processing information. This shouldn’t be surprising given their very
different DNA makeup. The abilities of each needs to be appreciated in
order for us to evolve the way we work together.”
— Eric Colson, Chief Algorithms Officer, Stitch Fix61
Data analysis
Stitch Fix brings together diverse data sources for AI, which extend beyond the typical financial metrics and customer feedback.
For example, fashion experts generate objective data (e.g., color, pattern) and subjective data (e.g., style) for each clothing item.
Computer vision analyzes photos of each item to extract additional data. This often includes data that describes the item’s color
(e.g., hue, saturation) and cut (e.g., sleeve length).
With all of this data, the challenge then becomes identifying what data has predictive power. Stitch Fix relies on their data
scientists and fashion experts to identify new variables and then tests whether these variables improve their predictions.
“You’ll find people doing operations research in AI where they’re figuring

out how we decide how much inventory to carry. I think there are
comparatively few of us who are now trying to use it to figure out what
content we should have.”
—Daragh Sibley, Director of Data Science, Stitch Fix
Human talent
Data science is heavily emphasized at Stitch Fix. The algorithms team, which is responsible for AI, has more than 100
employees and is led by Chief Algorithms Officer, Eric Colson, who reports directly to the CEO.62 The company believes this
structure encourages data scientists to develop novel ideas and experiment. When looking for new hires, the company focuses
on the “scientist” part of the data scientist role. These employees draw upon a variety of hard science disciplines and bring new
approaches and algorithms to existing retail problems. At the same time, the company relies heavily on people who have retail
backgrounds to guide how to classify clothing, a more subjective exercise, which can then be converted into objective data.

Results
Stitch Fix is focused on building “long-term enduring” customer relationships by sending items that make customers extremely
satisfied. As a result, the company favors measuring business impact not only through business metrics such as revenue or
gross margin, but also through customer satisfaction. The company collects this feedback through a survey that customers are
encouraged to fill out after receiving a shipment. The survey asks the customer to rank each item for style, fit, quality, and price.
The company observes that traditional business metrics, particularly revenue, are not necessarily aligned with customer
satisfaction. For example, Stitch Fix may send a customer the perfect blouse but not sell the item because the customer already
owns a similar blouse. Customer satisfaction is high because the customer liked the item and sees that Stitch Fix understands his
or her personal style; however, short-term incremental revenue is zero. Over time, Stitch Fix expects that customer satisfaction will
drive higher lifetime value as customers continue requesting Stitch Fix shipments, leading to revenue growth.
“[Machine learning] hits the top and bottom lines in very real ways.”
—Eric Colson, Chief Algorithms Officer, Stitch Fix63
Challenges with expanding AI

Designing new clothing items, such as blouses, is fundamentally a challenging problem. First, there are nearly infinite potential
blouses. Only a small fraction of all potential blouses have been made across all possible configurations of blouses. Stitch Fix has
data on an even smaller fraction of potential blouses. The company doesn’t know which popular blouses might exist but have
not yet been discovered. Second, there is a slower feedback loop with physical products like clothes. The company must wait
for the items to be manufactured, sent out to customers, and rated, which can take up to six months. This severely restricts the
number of designs that Stitch Fix can test, compared to a company with only digital products. The company mitigates this second
challenge by simulating customer responses to new clothing items with backtesting. In backtesting, Stitch Fix uses historical data
to estimate whether a customer would have purchased a new item if it had been available at that point in time.
“When you’re in the world of software, you can move really quickly. When
you’re in the world of physical products, it is fundamentally slower to
experiment and learn about AI.”
— Daragh Sibley, Director of Data Science, Stitch Fix
Insights
Stitch Fix is an example of a company that uses AI to enable machines to perform tasks they are strong in, and humans to
perform roles that involve creativity, subjectivity, real-world experience, or judgment. According to Daragh Sibley, “Humans
are really good at places where we just don’t have very much data.” In particular, humans are needed to understand how AI-
generated designs might combine with one another to contextualize recommendations and create items customers will enjoy.
The company has challenges given the physical nature of the apparel industry compared to the focus of a services company.
The timescale for experimentation and learning is fundamentally slower than at a services company. Another main challenge is
that data is sparse and variable, which underscores the importance of having humans continually involved in evaluating styling
options.

SOURCE, MAKE, STORE, AND DELIVERY PHASES CASE: IBM
Business
IBM has been an AI pioneer, pushing forward fundamental AI technologies with its tool Watson. Mentioned earlier for winning
the game Jeopardy, Watson has evolved into an AI platform that helps companies find and use business insights across a wide
range of use cases, such as talent acquisition, supply chain, customer engagement, and industries like healthcare and financial
services.64 As described by IBM’s CEO, Ginni Rometty, IBM hopes to augment human intelligence with machine intelligence and
“help you and me make better decisions amid cognitive overload.”65
“Guess what percentage of the world’s data is searchable? The answer is

20 percent. The other 80 percent lives with all of us who’ve established
businesses—and my view is that data has got a lot of gold in it.”
— Ginni Rometty, CEO, IBM66
While IBM offers Watson to clients looking to improve business insights, the company is also using the platform to improve its
own internal operations. This case study focuses on Watson applications within IBM’s own supply chain, which manages $30
billion worth of parts each year.67 The company is managing a digital transformation of its supply chain, incorporating more
analytics and AI into daily business practices.
Watson now acts as an assistant to IBM’s supply chain professionals. Employees can send the assistant questions and receive
answers informed by the company’s latest data and domain expertise. Sample questions for the assistant are:
• What are the top five components of supply shortage issues?

• Should we order more of a specific part or bring it over from another site that has it in inventory?
Although the assistant tool provides advice and data, supply chain professionals can override suggestions and remain the final
decision makers. Watson incorporates these human decisions into its learning over time to improve its functionality.
AI technology
IBM’s supply chain team uses Watson as external customers do, building the assistant tool based on existing APIs and training
the underlying models for the nuances of IBM’s supply chain. Matthias Graefe, director of Digital Supply Chain Transformation,
reports that Watson is sophisticated enough to “provide fast data and enhance human machine interaction.” For example, the
assistant learned how to identify product changes relevant to the supply chain team (e.g., new parts that need to be purchased,
existing parts that are now obsolete) from thousand-page specification documents. The assistant was trained based on 300
annotated sample documents, although fewer documents can be used if the data is more structured in nature.
The machine assistant has now started to prompt employees to look at likely follow-up questions, taking more of a moderator
role in resolution processes. Graefe envisions a future in which the way supply chain professionals work will change completely.
Watson will serve in a personalized advisory role in which it learns from more and more digitized data, interacts back and forth
with users to clarify its understanding of problems, and then engages in exponential learning. One day, he says, you may be able
to ask Watson “What should I work on today?”
Data analysis
The supply chain team embraces the domain expertise of its employees when training Watson.
For several months, top employees from various supply chain management areas (e.g., inventory management, quality
engineering) spent 5 to 10 hours a week demonstrating for Watson how to answer different types of questions. These employees
welcomed the opportunity to share their knowledge with the broader team.

Domain expertise is complemented by other data sources. The assistant tool is connected to IBM’s Enterprise Resource Planning
(ERP) system. Besides the data sources already available in Watson, it also draws upon news, social media, and weather
information.
The role of human talent

Software engineering generalists are responsible for building and maintaining the assistant tool.
These engineers receive informal coaching and advice from AI experts at IBM but operate largely independently. Matthias Graefe
believes that the focus on engineering generalists rather than AI experts on his team exemplifies the point that Watson’s models
are easy to train.
The entire supply chain team plays a role in training Watson by providing feedback. Employees grade each answer with a thumbs
up or thumbs down, triggering Watson to adjusts its models. In addition, there is a push for the team to use Watson Workspace, a
collaboration tool similar to Slack, instead of meetings or conference calls to discuss supply chain issues. These conversations are
recorded as text, creating training data for Watson that may be helpful in the future.
Results
To capture the short-term impact of AI investments, the group keeps track of the number of questions answered, the number of
questions that receive a thumbs-up rating, and the net promoter score (NPS) among employees.
There are promising early proof points. Taking advantage of the weather data in Watson, the assistant tool anticipated a hurricane
heading toward an important hardware manufacturing plant in Guadalajara, Mexico. The team adjusted their plans accordingly.
This insight saved the company money and kept clients satisfied.68
Insights
There is room for AI tools to take on more advanced tasks and become truly transformational for IBM’s supply chain team. As
Watson is trained with information relevant for all supply chain teams, such as customs rules for different countries, benefits can
grow. The human interface for the assistant tool will continue to encourage more employee interaction, and, as a result, capture
better training data. This would make it easier for employees to provide the “why” behind decisions, which is critical for Watson
to fully understand the decision made.
As AI tools take on more advanced tasks, IBM expects that the work of a supply chain professional will be divided between
machines and humans. The machines will take care of common activities (e.g., ordering parts) while humans will handle
collaboration (e.g., negotiations with partner suppliers), which requires their nuanced understanding of culture, politics, and risks.

SELL PHASE CASE: Adobe
Business
Adobe has made a strong push with its unified AI platform Adobe Sensei, which makes it easier for Adobe customers to create,
deliver, and optimize digital experiences using its software products. Sensei offers numerous services in the form of pre-trained AI
models that make predictions from the software user’s data, which are relevant for Adobe’s customers–primarily marketers and
designers. These services are embedded in Adobe’s products, powering new software features, and can be accessed by external
developers through an API. Adobe is also starting to open up the Sensei platform, which will allow external data scientists to
create their own models.
Example: AI-powered features across Adobe’s software offerings

• Creative Cloud: Eliminates manual work to edit an object out of a photo
• Analytics Cloud: Sends an alert when an item on a retailer’s website is selling above or below expectations
• Advertising Cloud: Offers more granular audience targeting across channels (i.e., search, display, video marketing)
• Marketing Cloud: Automatically crops an image to show off the advertised item depending on screen size
“Most of our customers aren’t aware that they’re leveraging AI. They just
know that they press this button and they get what they need.”
—Tatiana Mejia, Head of AI Product Marketing and Strategy, Adobe
Technology
Adobe’s domain expertise has shaped the Sensei platform and brought AI to address new problems. The company has chosen to
invest in AI especially relevant to its business—content intelligence, creative intelligence, and experience intelligence—instead of
general purpose AI. For example, it has partnered with Microsoft to create Experienced Data Models, an industry-standard open-
source database schema for digital marketing data. Given Adobe’s focus on applied AI, Sensei leverages open-source tools and
uses proprietary code only when necessary. It is built to accommodate a variety of data types, ranging from high-quality images
for designers to purchase-history data for marketers.
“Sensei for us is a common approach in platform and framework with all

three domains — creative, document, and customer analytics — because
at the end of the day, customers use all these tools in concert. They use
Creative Cloud to create the content, then they use Marketing Cloud to
deliver the content, and Analytics to measure how it’s performing. Having
the full journey allows us to do a Sensei stack that’s very unique because it
goes all the way.”
—Abhay Parasnis, EVP and CTO, Adobe69

Human talent
With over 10 years of experience with AI technologies, Adobe Research serves as an AI center of excellence within the company.
Adobe Research is on the forefront of AI technology, often partnering with academic researchers and publishing scientific papers.
In 2017, 66 of Adobe’s papers were featured at preeminent computer vision and computer graphics conferences.70
Once AI technologies are fully functional, they are transferred from Adobe Research to the Sensei platform and then to the
company’s software products. According to Tatiana Mejia, Sensei’s role is “to accelerate the speed with which our product teams
are able to adopt AI into our software.”
Adobe has learned that AI talent needs to be dispersed throughout the organization due to AI’s importance and rapid evolution.
To fill gaps in AI talent present today, the company has released an extensive online training program for engineers and product
managers. The online training was internally developed with Adobe-specific content and is taken on a part-time basis. In response
to strong interests from other employees, Adobe is exploring another AI online training program for more general audiences next
year.
Results
Adobe evaluates Sensei’s effectiveness by how much the company’s products are addressing customer pain points. This is
informed by both qualitative and quantitative feedback from customers. Exemplifying a clear success, a new AI-powered feature
that assigns metadata word labels to photos reduced the time for a particular digital creative agency to search for relevant photos
from over 10 minutes to approximately 30 seconds.
Insights
There is good potential for AI technologies like Sensei to enhance human creativity and intelligence, as opposed to replacing it.
Given the ease of using Sensei-powered capabilities within Adobe’s products, job satisfaction for entry-level jobs can be improved
due to the reduction in mundane or repetitive work. Adobe’s current offerings are considered a small step toward what the
company believes is ultimately possible. In the future, an intelligent virtual assistant may sense what the user is trying to achieve
and offer suggestions to streamline the process or improve the outcome.

SELL PHASE CASE: Pilot AI
Business
Pilot AI, founded in 2015, offers a computer vision software platform for compute-constrained camera devices. The platform’s
proprietary technology makes AI model training more efficient. Insights are extracted from the platform and entered into a
database that customers can access with a straightforward API. Among the many potential use cases for their platform, Pilot AI
is currently focusing on applications that analyze video footage in retail and smart home settings. Customers are primarily device
manufacturers and end users are the consumers or companies benefiting from the analytics solutions.
• Retail analytics: Analyzes in-store activity based on video footage and data collected from existing security cameras. As one
example, Pilot AI can track how many customers with a certain demographic profile are in a particular section of a store.
• Smart home devices: Provides consumers with real-time alerts on what is happening at their home (e.g., a package arrived)
from a connected camera or doorbell video footage.
Technology
Deep learning and other AI technologies are used for computer vision applications, enabling many forms of analysis of objects
and people. The company’s platform accelerates the training of deep learning models, reducing the time and memory required
for processing. This allows for model training to occur on compute-constrained commodity hardware, such as security cameras,
instead of on more expensive hardware such as desktop computers, or in the cloud. Pilot AI’s software can be installed remotely.
This allows customers to rapidly scale deployments from an initial pilot to every camera used for monitoring.
Data analysis
The Pilot AI platform analyzes video footage from a single camera. CEO Jonathan Su has found that customers often believe they
have more video data than they can actually use. This is often due to the company not having easy access to the video data in
their systems, not having legal rights to use the video data, or the video data being too sensitive.
Results
The value derived from the Pilot AI platform varies by application and customer. For the typical large national retailer, minimal
additional sales from the platform’s insights are required to offset the low per-store cost of the solution. According to Su, if a
retailer “sells an extra set of headphones, then they’ve recouped their cost for the month.”
Future vision
Pilot AI believes that “intelligence platforms” for speech, natural language processing, and vision will emerge. These platforms
will grow symbiotically with other companies in the IoT value chain. The company aspires to be one of the leading intelligence
platforms with widespread use in all IoT devices. In the company’s future vision, IoT device manufacturers will promote “Pilot
inside,” similar to the way PC manufacturers promote “Intel inside” today.
Insights
Pilot AI exemplifies how intelligence platforms can bring powerful yet cost-effective AI technology to their enterprise customers.
This allows enterprises to gain insights from new data sources without having to create innovative AI themselves or, in Pilot
AI’s case, without building the underlying efficient computer vision technology. Questions remain on whether the accessibility of
relevant data will limit the adoption of intelligence platforms. However, this challenge may be mitigated by intelligence platforms
being distributed on IoT devices that both collect the necessary data and run AI models.

SELL PHASE CASE: Salesforce
Business
Salesforce is an $88 billion company71 aiming to democratize access to AI in order to enhance productivity of sales, service, and
marketing professionals. The company’s AI efforts center around the Salesforce Einstein platform, which is embedded in their
software products as well as sold to consumers directly. Einstein provides customers with over one billion AI-driven predictions
each day.72
In the core software product, AI works in the background to inform recommendations, such as prioritizing sales leads or
estimating email marketing performance. The company tries to present these recommendations where they are most actionable
and provide some rationale for the recommendation.
Salesforce emphasizes that successful enterprise AI implementations require new business processes that provide the right data
to their software instead of repurposing existing data. According to Jim Sinai, VP of Marketing for Salesforce Einstein, “[Most
companies are] thinking about it 100 percent the wrong way. What you should think about is how you architect a workflow where
executing that workflow trains the model.”
Technology
With Einstein, Salesforce has built AI that can create custom AI models for each of their customers. These models reflect the
differences in each customer’s data and business model and are updated and retested when customers start keeping track of new
data variables.
“You don’t need a PhD to use Einstein— employees of all skill levels can
easily leverage it to connect with their customers, and they can also
customize it to fit their unique needs with just clicks.”
—Richard Socher, Chief Scientist, Salesforce73
The AI models used by Einstein vary by the type of data. Machine learning is used for structured data (e.g., statistics) while deep
learning is used for unstructured data (e.g., text, photos, videos, audio). Unlike other AI tools that require “big data,” many of
Einstein’s AI models can accommodate a small number of data points and still make useful predictions.
Data analysis
Einstein benefits from all the data in Salesforce’s CRM and other software modules. This data describes the customer journey
through both standard variables and custom user-defined variables. Approximately 80 percent of the data stored is for custom
variables. The different types of data are labeled with tags, called metadata, which allows the Einstein platform to use the data in
models without knowing the actual data contents.74 Einstein prepares this data to be used in AI models, replicating much of the
manual work traditionally done by data scientists.
Results
The results are highly customer dependent. Salesforce Einstein has seen their customers succeed at enhancing the user
experience in order to both increase revenue and decrease costs. Often, these benefits come from optimizing one small part of a
broader business process (e.g., improving email marketing open rates).
Change management often slows or prevents large companies from obtaining these results. Extended discussions among a large
number of stakeholders draws out the planning process. Smaller organization size and institutional learning, especially that gained
from the first AI project, can accelerate this process.

Customer results with Salesforce Einstein
LIDS
The sports merchandise retailer had a more than four-times increase in email open rates for new products (71
percent with Einstein Engagement Scoring compared to 8-15 percent previously).
U.S. BANK
The bank more than doubled their conversion of top-ranked leads with Sales Cloud Einstein, Einstein Analytics, and
Einstein Discovery.
Future vision
Looking forward, Salesforce Einstein will continue to work toward their goal of democratizing AI, but the exact shape of that
future is unclear. Sinai believes that technical progress in one year is overestimated, yet technical progress in 10 years is
underestimated.
Salesforce is excited about improving the user experience through adding advanced voice capabilities to its software. Also, the
company plans to continue to make its software more “conversational,” allowing users to access data-informed insights in a more
informal and intuitive way.
Insights
Salesforce has deployed AI as an engine behind its key marketing, sales, and service offerings. The company emphasizes process
flow design as a key step in designing effective AI models that can be retested over time, enabling the company to rely on process
rather than voluminous data. As important as the technology is in automating routine micro-processes and generating better
customer insights, human factors play a large role in the success of AI deployment for Salesforce. Within the company, multiple
departments collaborate to understand user needs and develop relevant AI solutions. Sinai believes that managing change
while redesigning processes is very important to Salesforce clients. Users also play a role in teaching AI models over time with
interfaces that ask for feedback. The combination of a clear AI strategy and an approach that is both client-based and human-
centered is critical to Salesforce’s vision for expanding the use of AI over time.
“You need to have an AI strategy now. You don’t need to have it

implemented, but you need to have a strategy and engineer your data flows
and business pipelines, being mindful of how AI will improve them.”
—Jim Sinai, VP of Marketing, Salesforce Einstein
“AI is the next platform—all future applications, all future capabilities for
all companies will be built on AI.”
—Marc Benioff, CEO and Co-founder, Salesforce75

SELL PHASE CASE: OhmConnect
Business
Founded in 2013, OhmConnect works with energy providers to encourage consumers to save energy. The company obtains
information on customer demand from online utilities accounts, pays consumers to reduce electricity usage during peak periods,
and then sells the electricity savings back to utility companies. This reduces the amount of energy sourced from polluting, or
“dirty,” power plants that are activated during peak periods. OhmConnect has 300,000 users across the U.S. and Canada
and has saved the energy equivalent of removing from the road over 341,000 cars (CO2 equivalent).76 Users typically receive
payments of $70-$150 per year.77 The company has been using machine learning to estimate energy savings opportunities and
establish performance targets.
OhmConnect is investigating how it can deploy AI to more effectively segment its users and identify patterns in their electricity
usage. Building on this understanding, the company hopes to personalize requests for consumers to use less electricity by
optimizing times of power usage, targeted messaging, and financial incentives. Planned next steps are to develop and test early
versions of AI models.
AI technology
Carson Moore, senior software engineer at OhmConnect, emphasizes the importance of understanding the business goal before
entering into an AI strategy: “You have to be very upfront about what you’re optimizing for.” In the case of OhmConnect, the
company is trying to optimize individual users’ ability to reduce energy consumption.
OhmConnect uses machine learning technologies to predict how customers will respond to the company’s requests to save energy
and thus estimate the energy supply needed. Going forward, the company plans to transition to AI technologies such as deep
learning to gain a more in-depth understanding of relevant energy demand and supply.
The company builds its own models and internal tools for testing and deployment, and is supported by third party plug-in
software and Amazon Web Services. The company has systems in place to test early models, receiving initial results in under two
weeks.
“We’ve built out a number of internal tools to be able to put AI models into
practice without committing lots of engineering time to implementing the
models at full scale. We’re able to get feedback pretty quickly on whether
or not this was a good idea or...whether we should go back to the drawing
board.”
—Carson Moore, Senior Software Engineer, OhmConnect
Data analysis
The company obtains comprehensive data about its users, directly collecting information on user behavior and household
electricity consumption data through online utilities accounts connected to the platform. OhmConnect relies on external load
forecasts to understand grid electricity requirements. The company is working to examine other signals from the data they collect
in order to group users together and better predict their behavior. For example, some customers may be more responsive to the
price offered for an hour of electricity saved, while others may be more responsive to special deals offering a small chance at a
receiving a large sum of money if they turn off their power.

Over time, OhmConnect is gaining a better understanding of usage patterns. The company plans to use data such as weather
forecasts, percentage of homes with air conditioning, and other targeted information to better group users and assess their
electricity needs. In San Francisco, for example, most homes do not have air conditioning due to a mild climate. As a result,
electricity usage is negatively correlated with hot weather in the city—a counterintuitive phenomenon.
The role of human talent

Artificial intelligence at OhmConnect is executed by software engineering generalists. In the future, the company may hire
data science or AI specialists as it becomes harder to increase OhmHours savings with current AI technologies, or as future
technologies become more relevant to the business. Given the unpredictability of AI advancements and business applications
for the company, OhmConnect has diversified its analytics efforts so that one or two people are focused on three or four different
strategies. This way, the team can pick and choose elements of various strategies as needed over time.
Results
OhmConnect focuses on a few key metrics. One is the percentage of normal energy usage that is avoided due to OhmConnect’s
offering. Another is incremental OhmHours—additional hours of electricity that are not consumed by OhmConnect’s users. To
date, hours saved have equated to 100 megawatts of energy. As additional AI technologies are deployed, the incremental impact
will be identified by comparing individual user behavior before and after the AI technologies are introduced.
With the next stage of AI technologies, OhmConnect hopes to respond to real-time electricity requirements. For example, if there
were a sudden increase in electricity demand, the company could automatically turn off all unused appliances in users’ homes.
These plans will heavily depend on the adoption of IoT technology for household appliances and improved reliability of IoT-
connected device data. The fact that future advancements in AI at OhmConnect may depend on improvements in IoT-enabled
devices is a good example of interdependencies within the AI ecosystem.
Ultimately, Moore believes that connecting demand and supply information will add the most value. “What could we do if we
had the most optimal machine learning algorithm that perfectly predicted when people are going to do everything they’re going
to do? And what would happen if users knew exactly what they needed to know and were able to act on it every time?...It’s
the combination of both that actually gets you into a place where we feel like we’re really providing the value we need to be
providing.”
Insights
OhmConnect demonstrates that AI can create new business opportunities. To date, the company has deployed machine learning
to predict customer usage. Going forward, AI can help the company understand nuanced consumer behavior at scale, which can
lead to increased efficacy at reducing electricity consumption and thus the potential for broader geographic reach. The company’s
vision includes leveraging AI to connect demand and supply in real time for optimal consumer savings and pollution reduction. By
leveraging developing multiple AI strategies today, OhmConnect is also positioning itself to take advantage of new AI technologies
as they become relevant for their current operations or inspire new approaches.

Citations
1
May 31, 2018 conference at Stanford Graduate School of Business, “Value Chain Innovation: The Promise of AI,” Padmasree Warrior.
2
Sam Ransbotham, David Kiron, Philipp Gerbert, and Martin Reeves, “Reshaping Business With Artificial Intelligence,” MIT Sloan
Management Review and The Boston Consulting Group, September 2017, p. 1.
3
“State of Artificial Intelligence for Enterprises,” Teradata, 2017, p. 14.
4
Op. Cit., “Reshaping Business With Artificial Intelligence,” p. 5.
5
“Artificial Intelligence: The Next Digital Frontier,” McKinsey Global Institute, June 2017, p. 17.
6
Op. Cit., “State of Artificial Intelligence for Enterprises,” p. 18.
7
Op. Cit., “Reshaping Business With Artificial Intelligence,” p. 8-9.
8
Ibid.
9
Peter Stone et al, “Artificial Intelligence and Life in 2030.” One Hundred Year Study on Artificial Intelligence: Report of the 2015-2016
Study Panel, Stanford University, Stanford, CA, September 2016. http://ai100.stanford.edu/2016-report (accessed September 6, 2016).
10
John Markoff, “Computer Wins on ‘Jeopardy!’: Trivial, It’s Not,” New York Times, February 16, 2011.
11
“Accenture Technology Vision 2018: Intelligence Enterprise Unleashed,” Accenture, 2018, p. 25.
12
Ibid.
13
Ibid.
14
Ibid.
15
Nils J. Nillson, “The Quest for Artificial Intelligence: A History of Ideas and Achievements,” Cambridge University Press, 2010.
16
“Artificial Intelligence,” OED Online, Oxford University Press. https://en.oxforddictionaries.com/definition/artificial_intelligence (accessed
September 17, 2018).
17
“Getting Smarter by the Day: How AI Is Elevating the Performance of Global Companies,” Tata Consultancy Services, 2017, p. 31.
18
Op. Cit., “Artificial Intelligence: The Next Digital Frontier,” p. 17.
19
Ibid, p. 19.
20
Op. Cit., “Reshaping Business With Artificial Intelligence,” p. 4.
21
22
Op. Cit., “Value Chain Innovation: The Promise of AI,” Jim Sinai.
23
Karl Utermohlen, “Four Cases of AI in Retail,” Towards Data Science, April 10, 2018.
24
Op. Cit., “Value Chain Innovation: The Promise of AI,” Padmasree Warrior.
25
Op. Cit., “Value Chain Innovation: The Promise of AI,” Tatiana Mejia.
26
“Algorithms Tour,” Stitch Fix, http://algorithms-tour.stitchfix.com (accessed February 19, 2018).
27
“Airbus: Reimagining the Future of Air Travel,” Autodesk, www.autodesk.com/customer-stories/airbus (accessed February 19, 2018).
28
Sophia Chen, “The AI Company that Helps Boeing Cook New Metals for Jets,” Wired, December 6, 2017.
29
Celia Shatzman, “bareMinerals’s New App Is A Game-Changer For Makeup,” Forbes, June 20, 2017.
30
Polina Marinova, “Shiseido Just Bought a Makeup App That Scans Your Skin and Creates Custom Foundation,” Fortune, January 18,
2017.
31
Diana Bubbs, “How Stitch Fix Is Using Algorithmic Design To Become The Netflix Of Fashion,” Fast Company, June 8, 2017.
32
https://www.uptake.com (accessed May 25, 2018).
33
Rachelle Blair-Frasier, “Q+A: Inside a Smart Steel Mill,” Manufacturing.Net, January 4, 2018.

34
https://www.ibm.com/watson/commerce/resources/order-optimizer-demo/fulfillment.html (accessed June 18, 2018).
35
Melia Robinson, “Tiny self-driving robots have started delivering food on-demand in Silicon Valley — take a look,” Business Insider,
March 24, 2017.
36
“Fasten Your Seatbelts: GE And Vale Are Making The Brazilian Railway More Connected And Smarter!”, GE Reports Brazil, April 22,
2017.
37
Jonathan Vanian, “Futuristic Robots Are Lending Their Hands in Gap’s Warehouse,” Fortune, October 24, 2017.
38
“How Germany’s Otto uses artificial intelligence,” The Economist, April 12, 2017.
39
https://www.salesforce.com/customer-success-stories/us-bank (acessed May 17, 2018).
40
Madeleine Johnson, “Starbucks’ Digital Flywheel Program Will Use Artificial Intelligence,” NASDAQ, July 31, 2017.
41
“Square Has a Profitable New Talent: Reading Your Mind,” Barron’s.
42
https://customers.microsoft.com/en-us/story/ups (accessed May 25, 2018).
43
“DIUx Selects C3 IoT as Strategic AI Software Platform for U.S. Air Force,” C3 IoT, November 1, 2017.
44
https://www.abundantrobotics.com (accessed June 26, 2018).
45
Op. Cit., “State of Artificial Intelligence for Enterprises,” p. 19.
46
Ibid, p. 18.
47
48
Op. Cit., “Value Chain Innovation: The Promise of AI,” Padmasree Warrior.
49
Ibid.
50
Op. Cit., “Value Chain Innovation: The Promise of AI.”
51
Ibid.
52
Ibid.
53
Ibid.
54
Ibid.
55
Ibid.
56
Shana Lynch, “Andrew Ng: Why AI is the New Electricity,” Insights by Stanford Business, March 11, 2017.
57
Nick Hastreiter, “What Impact Will AI Have In The Next 20 Years?,” Huffington Post, September 14, 2017.
58
Op. Cit., “Value Chain Innovation: The Promise of AI,” Daragh Sibley.
59
“A Future that Works: Automation, Employment, and Productivity,” McKinsey Global Institute, January 2017, p. 5.
60
Natalie Gagliordi, “How Stitch Fix Uses Machine Learning to Master the Science of Styling,” ZD Net, May 23, 2018.
61
Eric Colson, “More Human Humans: One Way in Which Our Lives Can Be Made Better By Ceding Tasks to Machines,” MultiThread,
June 16, 2016.
62
https://multithreaded.stitchfix.com/algorithms (accessed May 12, 2018).
63
Op. Cit., “How Stitch Fix Uses Machine Learning.”
64
https://www.ibm.com/watson/about (accessed June 12, 2018).
65
Megan Murphy, “Ginni Rometty on the End of Programming,” Bloomberg Business Week, September 25, 2017.
66
Ibid.
67
Peter Buxbaum, “IBM and the Cognitive Supply Chain,” Global Trade, November 28, 2016.
68
Ibid.

69
Khari Johnson, “How Adobe Used Its Huge Data Banks to Build Sensei, an AI Tool for Creatives,” Venture Beat, March 17, 2017.
70
https://research.adobe.com/adobe-research-papers-at-top-visual-computing-conferences (accessed May 17, 2018).
71
https://finance.yahoo.com/quote/CRM (accessed April 27, 2018).
72
Blair Hanley Frank, ISG, “Salesforce Einstein Now Powers Over 1 Billion AI Predictions Per Day,” Venture Beat, February 28, 2018.
73
Srinivas Tallapragada, “How WEF Young Global Leader and Salesforce Chief Scientist Dr. Richard Socher Views the Future of AI,”
Salesforce Blog, March 23, 2017.
74
Scott Rosenberg, “Inside Salesforce’s Quest to Bring Artificial Intelligence to Everyone,” Wired, August 2, 2017.
75
Ibid.
76
https://www.ohmconnect.com/about-us (accessed May 10, 2018).
77
Op. Cit., “Californians Collect Cash for Saving.”

Vcii White Paper Value Chain Innovation Promise Ai

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Vcii White Paper Value Chain Innovation Promise Ai

Uploaded by

Copyright:

Available Formats

VALUE CHAIN INNOVATION:

Current State of AI Adoption in Value Chains................................................................................................................... 6

II. Improvements Along the Value Chain..............................................................................................8–13

Strategic Considerations for Adopting AI....................................................................................................................... 15

The Human Factor..................................................................................................................................................... 15

Technology and Data.................................................................................................................................................. 16

Business Risks of AI................................................................................................................................................... 16

Design Phase Case: Stitch Fix........................................................................................................................................ 21

Source, Make, Store, and Delivery Phases Case: IBM ...................................................................................................... 24

Sell Phase Cases:

Use Phase Case: OhmConnect....................................................................................................................................... 31

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 2

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 3

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 4

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 5

FUNCTIONAL AREAS OF AI DEPLOYMENT

INDUSTRY ADOPTION LEVELS

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 6

Future AI Demand Trajectory*

8 Transportation and logistics

Consumer packaged goods Media and entertainment

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 7

VALUE CHAIN PHASES

DESIGN SOURCE/MAKE DELIVER/STORE SELL USE

Example: Example: Example: Examples: Example:

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 8

DESIGN PHASE – Stitch Fix

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 9

SOURCE AND MAKE PHASE

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 10

STORE AND DELIVER PHASE

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 11

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 12

Goal Increase Efficiency Automate Customer Service Anticipate Maintenance

Company/ OhmConnect / In-house UPS / Microsoft42 U.S. Department of Defense / C3 IoT43

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 13

Forms of AI-Driven Value Creation

SUMMARY OF VALUE DRIVERS

Design Source/Make Deliver/Store Sell Use

Personalize product Reduce delivery times Reduce inventory Personalize marketing

Optimize input purchase Avoid weather-related

Process Efficiency Process Enhancement Product or Service Innovation

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 14

“We want to make good AI invisible...There are dozens of features powered

The Human Factor

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 15

“My biggest challenge as CEO is to have machine learning data scientists,

Technology and Data

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 16

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 17

“The future is bright. People can act more like

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 18

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 19

SUMMARY OF “PROMISE OF AI” WHITE PAPER CASE STUDIES

Design Source, Make, Sell Use

Stich Fix IBM Adobe Pilot AI Salesforce OhmConnect

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 20

“We leverage data science to deliver personalization at scale, which we

FIGURE 5 – STITCH FIX AI APPLICATIONS

AI Technology Deep Learning Machine Learning Machine Learning

Source: http://algorithms-tour.stitchfix.com (accessed May 17, 2018)

STANFORD GRADUATE SCHOOL OF BUSINESS | WHITE PAPER: THE PROMISE OF AI 21

“Machines and humans have very different talents when it comes to

“You’ll find people doing operations research in AI where they’re figuring