Notes About "All-in on AI" by Thomas Davenport and Nitin Mittal

I’ve finished reading the 2023 book “All-In on AI: How Smart Companies Win Big with Artificial Intelligence” by Thomas Davenport and Nitin Mittal, and here go some notes I took.

Before starting, I’ve been reading quite a few books like this during the first half of 2024; here goes the list for those who might be interested:

Now back to our scheduled program and this week’s book. One of the articles written by Thomas Davenport, “Data Scientist: The Sexiest Job of the 21st Century” was showcased on page 273 of the 2022 book “HBR at 100: The Most Influential and Innovative Articles from Harvard Business Review’s First Century” as one of the most important business articles ever published on the Harvard Business Review.

In general, I found this to be a good business book, clearly targeting the C-suite of management people. I recommend it.

Introduction

What’s the purpose of the book?

Our focus in this book is on how large firms that existed well before AI are transforming themselves with the help of that technology.

1. What Does It Mean To Be AI-Fueled?

How many companies are using AI? Not a lot, as it turns out.

The AI-fueled organizations in our analysis comprise less than 1 percent of large companies. It wasn’t easy to find enough to write about in this book, but we were able to discover about thirty.

And for what purpose?

According to our surveys, AI technologies are most commonly applied in making business processes more efficient, improving decisions, and enhancing existing products and services.

Why aren’t there more?

One of the challenges in AI is getting systems into production deployment. Many companies embark on pilots, proofs of concept, or prototypes, but they put few or none of them into production.

Turns out, as usual, the problem is people:

How do highly successful AI-user companies address these issues and get systems deployed? First, they plan on deployment from the beginning, unless there is a problem in the early phases of a project.

Deloitte has referred to the current era as the “Age of With”–meaning that people are working collaboratively with smart machines. And Tom has liked that idea–which he often refers to as “augmentation”–enough to have coauthored two books on the subject.

Executives need their own version of AI upskilling. Most heads of AI and analytics tell us that they still spend a lot of time evangelizing to other managers about the value and purpose of the technology.

However, in one of our AI surveys, only 10 percent of US respondents stated a clear preference for retraining and keeping current employees.

How do companies get better at using AI? Unsurprisingly, through data.

If AI can fuel a company, data fuels AI. Companies that are serious about AI must be serious about data–collecting it, integrating it, storing it, and making it broadly accessible.

In addition to having good data, companies who aspire to transform their businesses with AI must increasingly have some unique or proprietary data.

Part of differentiating your company with AI is finding an existing source of data that hasn’t been fully exploited or getting access to new data types.

Companies that rely heavily on AI don’t just collect data and analyze it when they get around to it. They have a real-time approach whenever possible that allows data-based decisions to be made at the speed of contemporary business.

Here goes one of the main definitions in the book:

One way of summarizing all these attributes is to think of all-in on AI companies as organizational learning machines.

(Emphasis mine)

The other way in which AI-fueled companies become organizational learning machines relates directly to machine learning (at least the supervised form of it, which is by far the most common type in business).

In that way, the continuous training creates continuous learning and a more valuable model that fits the new data. In other words, if the world changes, the company’s predictive models change with it.

The final meaning of the term organizational learning machine focuses on the fact that these companies are consistent, reliable, and indefatigable. Their focus on AI in transforming their business is as relentless as any well-performing machine.

2. The Human Side

The book is clearly talking to higher ups in the C-suite, in which case it is important to remind them that there are humans in the lower ranks.

Leadership, culture, attitudes, and skills are human attributes that affect AI as much as or more than any other aspect of a company.

Surveys of large US organizations have suggested that the percentage saying they have data-driven cultures has even declined in recent years.

Then the book moves on to talk about the DBS Bank of Singapore, the largest bank in Southeast Asia by total assets, and how they’ve successfully moved to a data-driven culture and an AI first approach with the guidance of Piyush Gupta, their CEO.

As these early projects illustrate, one of Gupta’s strategies for AI was simply to start with the technology early and experiment.

Gupta continues to wrestle with the issue of where to store and process data. The bank has largely shifted to private clouds over the past several years, but there was clearly too much data to store it all on premise.

For several years the bank employed participative “hackathons” for senior managers to get them to think and act on digital innovation.

Gupta is committed to getting DBS people to embrace AI, and not to be scared that it might take their jobs. Thus far no one has lost employment at DBS because of AI, though some people have upskilled to change roles.

Some conclusions from this case:

What can we learn about AI leadership from this example? Gupta exhibits several traits that could be generalized to other leaders and organizations. First, it helps a lot to be familiar with information technology.

Also, it’s expensive:

Third, leaders hold the power of the purse. AI exploration is somewhat expensive, and AI development and production deployment is really expensive. AI leaders must invest enough to enable both levels of adoption.

“ROI too early kills the experimentation,” Gupta argues.

Ah, courage once again. CEOs love that word.

AI leadership also involves from seeing the future of a company’s business and having the courage to act to achieve it.

So, how to get started in AI?

For legacy companies like the ones we are describing, one of the greatest challenges to transforming with AI is to create a culture that emphasizes data-driven decisions and actions, and that is enthusiastic about the potential of AI to transform the business.

Many companies have begun data literacy or data fluency programs, in which large numbers of employees–perhaps even all of them–are trained in types of data, how that data can be used in analytics and AI programs, what types of decisions are best made with data, and how data and ways to make sense of it contribute to an organization’s success.

For leaders of analytics and AI functions, evangelism and cultural transformation about AI may be the single most important role they play in bringing about success with the technology.

Leaders of AI projects should solicit and take advantage of support from business leaders.

Internal marketing is very important:

Another aspect of aligning the organization behind AI is to communicate results and publicize successes often.

Oh, and don’t forget to tell people they’re not going to be fired (yet) and instead, setup internal training programs to “upskill” them:

Perhaps the most logistically challenging among the human side of AI issues is educating employees about its capabilities and likely future impact on their jobs.

Despite competing priorities, we believe the time is now to educate employees about AI and its impacts.

Similarly, leaders at DBS Bank in Singapore provided employees with seven digital skills, including digital communications, digital business models, digital technologies, and data-driven thinking. The program is called DigiFY, and it is aimed at upskilling many of the bank’s employees.

DBS also created a group of “translators”–people who are quantitatively oriented, but not data scientists, and who can mediate between business stakeholders and AI developers.

Airbus has partnered with Udacity to train more than a thousand employees in data science and analytics. The company asks both employees and their managers to devote half a day a week to the training.

Most of them would likely attest to the fact that AI technology is the easy part; getting people and organizations mobilized to explore, build, and use it is the challenge.

No shit Sherlock.

3. Strategy

What are the questions companies should ask themselves regarding AI?

Companies should increasingly be asking, “How can AI improve our business?” “What can we do with AI to create new offerings to help us grow?” “How can we make money with AI?”

What are companies actually doing with AI?

There are three major strategic archetypes for what an organization is attempting to accomplish with AI: (…)
Create something new, (…)
Transforming operations, (…)
Influencing customer behavior.

All-in on AI companies employ AI not only to support their existing businesses, but also to facilitate new business creation and entry into new markets.

Our annual “AI in the Enterprise” survey results over several years suggest that most companies use AI to improve existing business processes.

Another strategic use of AI is to create new products and services, or provide significant enhancements to existing ones.

AI can simply be used to transform operations–to make existing and well-defined strategies much more successful.

The hot new thing, however, is using AI to make people buy more of your product:

One of the newest strategic objectives for AI is influencing customer behavior.

This is not really new, but what’s new is the scale and speed at which it happens.

To be fair, this approach of influencing behavior isn’t really new. It was pioneered by Fair, Isaac & Co. (now FICO), which created the first credit score in 1958.

The credit score was one of the first commercial applications of machine learning.

If AI is going to enable new strategies, business models, and customer behaviors, it doesn’t make sense to manage the technology bottom-up.

Remember: make the impossible dialogue possible.

For strategic decisions to be influenced by AI in the appropriate fashion, a few preconditions apply:
Educating senior managers on AI is critical.

4. Technology and Data

In this chapter the book moves into the “how” of using data in AI pipelines.

Different technologies are useful for different use cases, and organizations that adopt AI broadly and deeply have breadth in their use cases and the technologies that are applied to them.

Rule-based systems are often considered obsolete, but they are common in fraud and anti-money laundering systems, and DBS uses them for that purpose. One of their common shortcomings, however, is that they create far too many false positives–as many as 98 percent at DBS.

Unsupervised learning, which is typically used to cluster similar cases with no outcome variable, is less common in business.

Using AutoML, the 84.51º subsidiary of the Kroger Co. is developing a “machine learning machine” that can build and deploy very large numbers of models with relatively little human intervention.

High-achieving organizations with AI, labeled transformers and pathseekers, were more likely (typically by about 25 percent) than the two lower-achieving groups (starters and underachievers) to agree that they have adopted several different AI operational practices that facilitate scaling and ongoing management of AI.

Here’s another use case, in this case the Dutch energy conglomerate, Shell:

Shell partnered with Microsoft to make development tools and methods available.

At some Shell facilities, it took six years to inspect all piping; the drone and the AI system can do it in a few days.

In conclusion:

Data environments for AI-oriented companies have several characteristics: (…)
Most are cloud-based. (…)
The data they use in machine readable. (…)
They are adding team members. Despite some help from AI, wrangling with data is still a labor-intensive activity.

One of the most popular applications for AI in recent years, according to Deloitte’s annual survey of AI activity, is IT itself. AI and automation capabilities can predict and diagnose problems in networks and servers, and automation programs can restore them to health.

We believe that every large organization–and certainly those that are or aspire to be AI first–should designate smart people to follow AI technology trends, try out new technologies, and import them when they seem to fit the organization’s need.

5. Capabilities

How can you evaluate the level of advancement of an organization in the world of AI? The authors propose a “capability maturity model” very similar to the (in)famous CMMI model for process engineering:

Capability maturity models tend to have five levels, and we see no reason to depart from that standard. They also tend to have low capabilities at Level 1 and high ones at Level 5, and we follow that pattern as well.

We might also add a “Level 0” to describe companies that have no AI activity whatsoever, but this is certainly a minority category among large firms in sophisticated economies.

As a result, business leaders ultimately drive the agenda for what Analytics and AI use cases are developed, partnering closely with their dedicated analytics and data teams.

A word about ethics in AI:

A key aspect of developing AI capabilities is ensuring that AI systems are trustworthy and ethical.

However, the status of these AI ethics groups at some vendors–particularly at Google–has been somewhat uncertain and unstable.

Everything is uncertain and unstable with Google, except maybe for Kubernetes and the Go programming language. But I digress. Deloitte to the rescue:

Deloitte’s Trustworthy AI Framework, which was developed to aid clients in developing their own policies, is a good example of such a policy framework.

6. Industry Use Cases

This chapter provides some examples of use cases for AI in various industries:

AI-powered organizations choose use cases that will differentiate them from competitors (at least for a while), advance their business strategies and models, and fit with their business process designs.

Financial services–including banking, insurance, investment management, and trading–have been the industries that are most active in the use of AI. It’s an information-rich sector, rapid and accurate decisions are critical to its success, and its customers need substantial amounts of advice to live more successful financial lives. Financial services organizations also typically have the financial resources to invest in AI.

Seriously? The expression “to live more successful financial lives” is a thing?

In the United States, government and public service organizations got off to a slow start in adopting AI, at least outside the military and intelligence sectors.

I don’t think it’s much better on this side of the Atlantic, to be honest.

7. Becoming AI-Fueled

The final chapter is a call for inspiration and revolution. First of all, the good news: you still have time to become an “AI-first company”.

The good news is that no company was powered by AI a decade ago or so, and for AI-first companies today we can describe several of the paths they took to move in this direction.

Deloitte hasn’t fully completed a transition to be AI fueled, and it’s hardly abandoning its human workforce; it has almost 350'000 employees worldwide.

Finally, some important takeaways for the reader:

There are important lessons that other organizations can learn from these companies’ AI journeys. (…)
Know what you want to accomplish with AI. (…)
Start with analytics. (…)
Reduce “technical debt” and create a modular, flexible IT architecture. (…)
Put some data and AI applications in the cloud. (…)
Think about how to integrate AI with the workflows of your employees and customers. (…)
Marshal some data assets. (…)
Create an AI governance and leadership culture. (…)
Develop and staff centers of excellence in AI. (…)
Be prepared to invest. (…)
Work with an ecosystem. (…)
Build solutions across the entire organization.