Artificial Intelligence: Business Strategies and Applications
Module 6 – AI Strategy
Quick Reference Guide
Learning Outcomes
• Examine the negative and positive consequences of AI technology due to the hype
surrounding it
• Explore several successful applications of AI and the sectors where AI can and will
be the most useful
• Explore a couple of AI failures as well to understand the limitations of the technology
and challenges that come with AI implementation
• Explain value generated industry AI-driven applications in terms of the 4Cs
framework
• Explain the business strategy and technology approach of an AI startup through the
lens of identified frameworks
• Describe the potential success of an AI startup based on its business strategy and
technology approach
How is AI different from traditional computer programming approaches?
• Expectations of AI range paradoxically from unrealistically high to ridiculously scary,
and there is a lot of hype and fear around the technology, mostly in the news media
and science fiction.
• Because of the mathematical intensity of many AI techniques, especially artificial
neural networks, the computational power of computers really matters. The
continuation of Moore's Law over the past 30 years has only increased the raw
capability of AI even more.
• The successful application of AI is often a complex alchemy of approaches.
Different problems require different approaches and algorithms determined through
trial-and-error.
• Data is the fuel for AI learning machines. With AI systems, we are trying to solve a
problem where we don’t know the exact relationship. We can guess at the list of
potential input variables, but we don’t know how much they should be weighted or
combined to predict an output. We need data to train the AI system and develop a
model for this relationship between inputs and outputs.
• AI systems tend of lack certainty and predictability. Most of the techniques of AI are
dealing in probabilities. Dealing with uncertainties is a distinctive advantage of AI,
but the lack of explainability is particularly challenging in regulated or public policy
areas.
Artificial Intelligence: Business Strategies and Applications
Page 1 of 10
o Emergent Behavior: the algorithm may and should take novel approaches
not imagined by the programmers
• Many AI models are non-deterministic. In computer science, a nondeterministic
algorithm is one that, even for the same input, can exhibit different behaviors on
different runs.
AI-Solvable Problems
• There are a set of well-defined problem categories that have proven to be very
solvable with existing AI technology. Some business sectors and functions are
particularly ripe for adopting AI.
• The set of AI-solvable problems include optimization, search, and prediction. Great
progress has been made in perception, understanding, and complex movement and
manipulation.
• Computer perception, especially computer vision and hearing, has made great
strides in recent years.
o Speech-to-text is already about as good as humans
o Google API is comparable to human closed-captions has word-error-rate less
than 5%
o Companies have been using a process called fingerprinting to identify songs
o Self-driving can look beyond the visible spectrum using LIDAR and radar to
create a 3-D model for its environment
o Facial-recognition algorithms are now as good as the best forensic
examiners
o Perspective, a project by Jigsaw (Alphabet Google’s think/do tank) is an API
to identify toxic speech, which is anything using abusive language, threats, or
harassment
o AI can perform rudimentary reviewing, and even drafting, of some legal
documents, saving time while avoiding errors and omissions
o Word2Vec was used to sift through over 3 million scientific paper abstracts
for connections humans had missed, and was able to find predictions for
possible new thermoelectric materials
o Using AI, robots can now perceive their environment and navigate around
people and other obstacles while accomplishing their task
o Kiva robots reshaped the way Amazon set up its warehouses, allowing for
smaller, more densely-packed facilities closer to urban centers
Ripe Sectors and Functions – Optimization
• Optimization: doing the same things better, cheaper, and faster
Artificial Intelligence: Business Strategies and Applications
Page 2 of 10
• In agriculture, AI is used to harvest more delicate crops. Strawberry picking robots
like the one developed by HarvestCroo use computer vision and advanced robotics
to identify and pick only the ripe berries without damaging the plant.
• AI is using soil sensor data and images from robots in the field, as well as drones and
even satellites from above, to monitor the health of plants and precisely deploy
pesticides and nutrients.
• The X lab at Alphabet (Google’s parent company) has a project called Mineral that
uses AI-powered robots in the fields to monitor plants and soil conditions.
• AI is being used extensively in supply chain and logistics, from Kiva robots and
Stretch by Boston Dynamics, a robot that can unload boxes at near human speed.
• Manufacturing is quickly adopting AI to optimize production, minimize waste and
errors, and help the process of quality control. Machine learning can check for flaws
in manufactured products.
• Machine Learning is being used to detect fraud, assess credit and insurance risk,
and even make investment decisions.
Ripe Sectors and Functions –Transformation
• In the legal space….
o Though AI is more likely to aid than replace attorneys, it is already being used
to review contracts, find relevant documents in the discovery process, and
conduct legal research. More recently AI is being used to draft contracts and
opinions, predict legal outcomes, and even recommend judicial decisions
about sentencing or bail.
o However, AI is not ready to replace human judgment, and the risk of
embedded bias in data that fuels AI is especially concerning in this context.
o Companies like CS Disco, which went public recently, provide AI-powered
discovery services to law firms across the US.
• In healthcare…
o In addition to improving overall efficiency, and simply avoiding medical
mistakes, AI is transforming healthcare in many ways, but especially
diagnosis and drug discovery.
o One startup, Paige.ai is combining biopsy slide image data with clinical notes
to improve this prediction. Recently, the FDA cleared their AI model to spot
prostate cancer in tissue slides. AI can even use low-quality images to identify
early signs of disease.
o AI is also being used to develop new medical tests. Grail used machine
learning to develop a blood test to detect cancer early. Their model was
trained on finding correlations between DNA fragments in the bloodstream
and early markers of 50 cancers and can determine the location of the tumor
with near 90% accuracy.
Artificial Intelligence: Business Strategies and Applications
Page 3 of 10
o AI can help reduce the very costly process of drug discovery as well as help
discover drugs that might otherwise be missed by human scientists. A major
contributor to the cost of discovering a new drug is sifting through the
massive number of possible candidates.
o Atomwise is using computer vision to look at the shape of molecules to infer
the likely function or behavior compared to similarly shaped ones.
o Exscientia, recently announced the first molecule designed by AI to enter a
phase one clinical trial. The exploratory research phase took less than 12
months, a fraction of the time typically required to discover a new candidate
molecule.
o Google’s DeepMind recently developed an AI called AlphaFold2 which solved
the 50- year-old problem of “protein folding”.
• In the workplace…
o Startup companies like Seattle-based Textio are helping employers find more
diverse candidates and to be more inclusive in their communications in
general. Textio uses AI to recast job descriptions to remove unconscious
bias.
o Using AI as notetaker during virtual meetings can help teams to be more
effective, while capturing institutional knowledge.
o AI is even being used to measure and improve organizational effectiveness.
o Across all applications of AI in the workplace, it is essential the employers
fully disclose what employee data is being used and take care to avoid bias,
especially when machine learning is being used for making important
decisions like hiring or promotions.
Ripe Sectors and Functions – Reinvention
• These sectors are not just using AI to do the same things better, they are changing
their value proposition to customers, inventing new products and services, and
completely changing their economics.
• In transportation…
o Fully autonomous vehicles on the ground or in the air will have a profound
impact on this sector.
o Waymo is focused on developing fully autonomous vehicles, referred to as
level 5 autonomy. With over 20 million miles on the road its autonomous
driving technology is arguably the most experienced. Waymo cars use a
complex array of sensors including LIDAR, radar, and cameras to feed its AI
models. Recently they have started testing in San Francisco and other cities.
o One of the biggest challenges with an assisted driving approach is that it’s
hard to keep humans ready to re-engage when needed. Humans become
complacent and distracted when not focused on an activity.
Artificial Intelligence: Business Strategies and Applications
Page 4 of 10
o For the near term, there will be many corner cases in autonomous driving
requiring driver intervention.
o An alternative to having a human driver in the vehicle is remote control. Long-
haul trucking companies may prefer this approach to achieve greater
automation more quickly in the transportation of goods.
o Einride is a startup based in Sweden that allows remote controls of driverless
trucks.
o Phantom, a startup based in Los Angeles, allows drivers to operate forklifts
and construction equipment remotely assisted by Machine Learning. If the
network has latency or other challenges, AI can step in to operate the vehicle.
• In retail…
o Computer vision is being used by Amazon Go, an experimental convenience
store that allows people to shop without needing to checkout.
o Built upon AI models, Search has already revolutionized the retail process
online, as well as offline. Search companies like Google are getting more and
more precise in answering queries using AI. Rather than simply matching key
words and returning a long list of possibly relevant websites, Google search
uses machine learning to try to understand the meaning of the query and
answer it directly if possible. As you type, it even suggests the query based
on common requests, personalized to your history.
• In the creative industry…
o OpenAI created Jukebox to demonstrate the power of GPT-3. The model
was trained on raw audio, rather than words. It can generate original music,
including rudimentary singing, in a variety of genres and artistic styles.
o There are lots of generative text tools, some built on GPT-3. With a short
prompt, an AI writer can create a story, article, or report, but don’t expect
perfection.
o AI can be used to accompany a performer as well. Google launched an AI
experiment called Duet that would play along with someone on the piano in a
call-and-response manner.
o Google has since launched its own AI music generation tool, Magenta, which
learns how to compose in the style of a given artist by training on their works.
o LA-based startup, Endel, uses AI to generate personalized music, they call
“soundscapes,” that sync to your desired mood -- and even your heartbeat --
to allow you to focus or relax.
AI Business Application Failures
• The first kind of failure is when AI is just not ready for the task at hand. AI sometimes
fails due to a lack of good training data or a well-defined objective. But many times,
AI fails because it requires operating conditions to be too perfect compared to
Artificial Intelligence: Business Strategies and Applications
Page 5 of 10
reality or an understanding that comes from a human lifetime. Certain tasks also
demand dexterity and agility that robots lack.
o Ex) One area that AI is just not ready yet is the task of grasping random
objects.
• AI can fail when it cannot deal with corner cases. Many AI applications work under
controlled conditions but fail as soon as those conditions change. Since these
failures are unexpected, they can lead to tragic accidents. It’s when we start relying
on AI that humans can become vulnerable to these corner case failures.
o Ex) A self-driving car killed a woman walking her bike across the street at
night in Arizona in March 2018. It is the first known autonomous-vehicle-
related pedestrian-death on a public road.
• Lastly, some AI failures are due to poor deployment. The most common version of
this is when uncorrected bias in the training data yields inaccurate results. Another
type of poor deployment is when humans inaccurately interpret or use the results of
AI, sometimes attributing more intelligence or meaning than is justified.
o Ex) AI systems have been used by police and courts to predict the likelihood
that convicts will reoffend. Research by ProPublica showed that the system
was biased against black prisoners, disproportionately flagging them as being
more likely to reoffend than white prisoners (45% to 24% respectively).
These predictions may have led to longer sentences than justified.
Competitive Advantage
• Gartner’s Hype Cycle
• First-Mover Advantage can be fairly strong as you develop expertise in the area
and by the time the technology is really mature, you are ready to implement it
quickly.
• If your competitors all follow the hype and adopt say deep neural nets, you may be
left behind. On the flipside, you may just wait until others in your industry figure out
how to use deep neural nets the best and copy them.
• AI Infrastructure, Algorithms, Data, and Domain Expertise are other potential
sources of competitive advantage.
Artificial Intelligence: Business Strategies and Applications
Page 6 of 10
• Building a strong AI infrastructure can be a strong source of advantage itself; not
just the computing infrastructure, but also the organizational infrastructure of
building and placing AI teams within the firm.
• With all the open-source tools and rapid development, experts say that a new
algorithm gives you about 6 months, so new algorithms don’t give you a very strong
competitive advantage.
• Data, on the other hand can be a very strong source.
• We can see in many examples that domain expertise may be a strong source of
competitive advantage.
Data Strategy
• More sophisticated algorithms, especially large neural networks can better take
advantage of large training data sets than their predecessors.
• Virtuous Data Cycle: to build a prototype, you need data to train the model, but
where do you get the data if you haven’t built the prototype yet? Once you get an
initial data set, train a neural network and build a product. When people buy your
product, they’ll use it, and you can generate more data. This leads to better models
and a better product, completing the cycle.
• It’s fairly easy to copy the hardware and the algorithms used to train the bot, but
nearly impossible to build a large data set of a specific photos from scratch.
The 4C’s Framework
• Carefully defining the value proposition is crucial for commercial success and the
value proposition has important implications for the technical implementation.
• It’s always good to have some kind of a structure when considering strategy, and the
4C framework highlights possible ways in which the firm can extract profit from the
‘environment’ defined by the demand structure, competitors, and other industry
players.
• The 4C’s are Create, Compete, Concentrate, and Capture.
• Create: you create a new market as a first mover, either by creating an entirely new
product category or by adapting an existing product to a context or environment
where it wasn’t feasible before.
o Ex) RS metrics collected images from retail parking lots to get a good idea
for how each store is doing and predict retailer revenues and stock prices
and sold this data to investors who then used it to trade stocks.
o Ex) AI used for drug discovery. The relationships between input and output
are so complex that state of the art techniques like deep neural networks are
needed to uncover them.
Artificial Intelligence: Business Strategies and Applications
Page 7 of 10
• Compete: you use some sort of new tech to improve an existing offering to gain
traction against your competitors.
o Vertical Differentiation: increase quality, and everybody agrees the resulting
product is better
o Horizontal Differentiation: different consumers have different tastes; some
people may like the new product better, some might not
▪ Ex) McCormick, built an AI system trained on decades worth of data
about spices and flavors to come up with new flavor combinations
▪ Ex) Companion Labs uses computer vision to perform the painstaking
repetitive exercises needed to train a dog
o AI is very useful in customizing products to individual needs, eBay has been
using algorithms for years, matching sellers and buyers, social media
platforms use algorithms to curate feeds to the user’s interests
▪ Ex) JP Morgan Chase was one of the first companies to hire an AI
startup to write the ads because they outperform copywriters in
getting clicks on the ads
▪ Ex) A startup called Tailor Brands uses algorithms to generate brand
logos and offers this service to small businesses as a replacement for
human logo designers
• Concentrate: a fundamental change to your cost structure, either by economies of
scale or by shifting your cost base from variable to fixed.
o Ex) Self-driving trucks are about reducing costs, or in other words
concentrating the costs. Instead of hiring a truck driver for every single truck
you can replace them with a software which has to be developed once
o Ex) Motion.ai is a startup that was acquired by HubSpot the inbound
marketing company with the clear intention of deploying the chatbots for
reducing human costs
o Cost saving applications often couple AI with IOT technologies. Using
sensors to monitor systems and to track objects is a great way to generate
data that can be fed into an AI powered tool to optimize many areas from the
supply chain to internal operations.
• Capture: use technology to capture more of the area under the demand curve,
mainly through personalized or dynamic pricing.
o Dynamic Pricing: some products are perishable, they become worthless
after some time, so manage pricing over time by capturing as much value as
possible through variable prices. At its core, the system forecasts demand
and optimizes price at any given moment.
o Personalized Pricing: a fully dynamic and personalized pricing system
utilizes personal information and offers a potentially different price to every
person at every moment in time
o Obstacles: Taking into account competitive pricing is challenging, not only
from the algorithmic perspective, but also just getting competitive pricing
Artificial Intelligence: Business Strategies and Applications
Page 8 of 10
information. People tend to be fairly accepting of dynamic pricing, but not
personalized pricing.
Startups
• Startups have the luxury of being very focused and nimble. But time is their enemy
since competition is forever at their heels and resources are limited.
• It’s important to have clarity about what real-world problem you’re solving. This
usually requires domain, as well as sufficient technical understanding of what AI can
achieve.
• Startups need a well thought out and scalable approach to data to make sure they
can deliver on the promise of AI. Quality data is the lifeblood of machine learning. In
most cases a startup will need to partner for access to this training data.
• There are powerful platforms that startups can leverage to test and even launch
their AI solutions. Using these cloud-based tools will allow the startup to seamlessly
scale with its business.
• Find ways to add value quickly in the early stages of the implementation. This can be
in the form of a technical proof of concept or a set of business case studies. Even
better yet, adding value to customers out of the gate can provide your startup with
some early revenue.
• It’s also important to have a plan to attract and retain the right talent. Also,
depending on the solution, don’t be afraid to bring on people in related fields like
psychology and linguistics, since diversity of thinking, as well as people on the team,
tend to yield more robust AI models.
• Finally, it’s important to establish ethical boundaries early.
Established Companies
• On data strategy, large companies have a substantial advantage in that they likely
own the data, or at least have access to it from their customers, suppliers, and
partners. Nonetheless, the stakes of mishandling this data is dramatically higher.
• Much of the same implementation guidelines for startups applies for AI projects
within big companies. In fact, many companies tend to think of AI projects like
internal startup ventures.
• Having a data strategy function helps to overcome organizational resistance to
sharing data and setting priorities across AI and other data-hungry projects will
require strategic oversight.
• Attracting AI talent is a different kind of problem for large companies, since they
usually can’t offer the promise of big equity payouts that startups can. One way to
overcome this challenge is to pool machine learning and data science talent
together into a central organization.
• Large companies should establish ethics-related positions or review boards as a
part of their AI initiatives.
Artificial Intelligence: Business Strategies and Applications
Page 9 of 10
• Here are a few ways from Accenture that large companies can pivot themselves
around AI. First, redesign the IT stack for the age of AI. Second, create an adaptable
IT architecture. Rather than large, monolithic systems, deploy microservices and
serverless architectures to allow for more agile application development. And third,
design systems for humans, by humans. As much as possible allow less technical
people to contribute to the process of developing and deploying AI solutions.
Artificial Intelligence: Business Strategies and Applications
Page 10 of 10
