I had the chance to meet Karim Lakhani on a virtual Harvard Business School panel a few weeks ago. He is one of the authors of "Competing in the Age of AI," along with Marco Iansiti. I picked up a copy of their book and have to say it is hands down the best book I've read on AI and business strategy. It's the first book that deeply analyzes how the existence of AI technology re-shapes strategic thinking, and how companies should respond. And it addresses some of the practical challenges of doing so. What follows is my Q&A with the authors.
1. The most recent wave of AI technology is a few years old now. When did you realize that businesses weren't thinking about it the right way and that something like this book needed to exist?
We began our research journey for the book about 8 years ago when we started a new course at HBS on Digital Innovation and Transformation. As we dug deeper into companies that were leading the digital revolution we realized that not only were these companies utilizing network effects to increase the underlying utility of their products and services to their users but also pioneering the use of data science and machine learning to both increase value creation and value capture. In addition these companies were also organized very differently in terms of their operating model - they had worked pretty hard to bust silos and to build a data and analytics infrastructure in a way that would overcome bottlenecks in value delivery to their customers. We noticed a real divide in what the AI-native companies were doing and the rest of the incumbents in the economy. The AI-native companies had transformed their operating models while the incumbents were mainly stuck in piloting a few good ideas with no real implementation. So although the hype around AI has increased quite a bit over the past few years - real adoption and transformation for much of the economy is still pretty far away. We wrote this book to enable leaders to see the changes in both business and operating models that can be enabled with AI and make it strategic and practical at the same time without resorting to either magical thinking or hoping for AI unicorns to drop-in from the skies to help companies.
2. You write about the need for companies to build an "AI Factory." What are the tactics that get employees and managers comfortable with the idea that AI makes many of the business decisions?
The fundamental idea behind the AI Factory is to industrialize a company’s approach to data, analytics and artificial intelligence. The AI Factory does for analytics what industrialization did for manufacturing over a hundred years ago. Data is processed in a systematic, standardized fashion, catalogued and centralized, cleansed, tagged, normalized and integrated, and exposed through open interfaces, known as “APIs,” which are available to power new business applications. The data platform forms the core of the firm, with an organization consisting of engineers and managers overseeing it and harvesting its power. The technology underlying the AI Factory also shapes the organizational architecture and processes.
AI Factories drive three types outputs: 1-Predictions - about some future action or state of a context; 2 - Pattern Recognition - find patterns in large amounts of data that are hard to find by humans; 3 - Automation - drive a series of actions based on responses from interactions in the environment. Most of us interact and encounter AI Factories everyday. For example, your search on Google and the related ads shown to you happen in an automated fashion through an AI Factory. Movie, music and book recommendations are happening through AI factories at Netflix, Spotify and Amazon. At Ant Financial, new bank account openings and approvals, loan applications and insurance offerings are being driven by AI Factories.
Our advice to managers and employees is to first acknowledge the inevitability of AI Factories playing an important role inside of companies. We are going to be living in a world of data and building an AI Factory is going to be essential to deal with the deluge. Second, we point out that the consumer tech giants have changed the expectations we have about our interactions with AI. Their customers, B2C or B2B or even B2B2C now have expectations of superpowers in terms of services and product features. Not having AI Factories will lead to a competitive disadvantage. Third we think that these types of investments are important to drive organizational transformation. The future really is one where all companies have AI factories at their core with workers designing and working in conjunction with AI systems.
3. My favorite part of the book was the discussion about network effects and the new concept you introduced of "learning effects." Can you give an example of learning effects for the newsletter readers?
Some call it “learning by doing”. The basic idea behind the learning effect is that AI systems improve in their capabilities as they accumulate more and more data. A simple example is how Google can predict what you are searching for and it continually gets better over time. The accumulating data Google has simply enables them to keep improving the search experience. The same thing happens in GMail with its ability to predict the completion of your sentences. Learning effects improve product quality but also enable the company to increase the scope of its offerings to its customers as well. Ant Financial’s data on transactions among consumers and merchants enables it to come up with new offerings for both sides of its platform. Uber Eats has unprecedented data on the micro-geography of taste and menu options, and they can use it to inform restaurants about changing preferences and about how to optimize their online menus to meet new needs.
4. Do you think learning effects have a cumulative advantage to them - the longer they are in place, the more difficult they are to dislodge? And if so, does that mean companies that start late are at a disadvantage?
We are just beginning to understand how learning effects work and the underlying value of data that companies have. We have begun a research program at lish.harvard.edu to help kickstart our thinking and analysis. WE have already discovered that the answer is complex and, basically, it depends a lot on the kinf of data you are accumulating, on the technical properties the algorithms being trained, and on the impact of the algorithms on the business. The cumulative advantage provided by specific data is certainly one hypothesis to investigate further. But we all need good empirical evidence to help us figure it out!
5. Much of the book focuses on practical pieces of implementing these changes, and one of the big things you point out is the need for employees to be designing and managing these systems, more than operating them day to day. How does that change the skills taught in business schools?
We have been thinking and acting about this a lot. The age of AI is as important for current students as it is for working executives. We started the Harvard Business Analytics Program https://analytics.hbs.edu/ as a way to consolidate and offer a curated set of learning combining the best of HBS, Harvard Paulson School of Engineering and Applied Sciences and Harvard’s Statistics department on this topic. Our sense is that leaders today need to understand the technical stack (statistics, systems and AI/ML approaches) and be able to apply it to business settings including strategy, operations, marketing and leadership. We have courses in all of these subjects at the executive level and also at the MBA level. As you can imagine, demand is through the roof.
6. As companies collect and consumer more data to run these AI models, do think data privacy rules will be a source of differentiation in certain markets? To date it seems like consumers have been willing to trade privacy for free stuff almost all the time. Is that changing? And if so, will companies adapt to serve different market segments around data privacy?
This is a very important topic. Privacy as an “economic good” is not well understood by individuals, companies, regulators and academics as well. One of our colleagues, Leslie John https://www.hbs.edu/faculty/Pages/profile.aspx?facId=589473 is doing fascinating research in this area. Preferences between data sharing and convenience are changing fast. Companies are learning to use “less” data - or throw away data after initial use - or do learning on device. At the same time companies are differentiating on privacy - look at Apple versus Google. So we believe that this too is a dynamic situation with preferences, regulation and technology changing rapidly. Lots of opportunities to innovate and for startups to make a huge difference.
7. You write that companies need to rebuild their entire digital foundation to best take advantage of AI, but much of the conventional wisdom the past few years has been to start with a small project. How do you see this debate shaking out over time? Do companies need to be more aggressive in adopting AI?
Well we all have to learn to crawl before we can walk! Nothing against small pilots but the worry we have is that these pilots just stay as pilots and there is no overarching strategy to implement and scale and transform. AI needs to be part of the broader digital transformation agenda for the C-suite and it should not be relegated to the basement as a nice to have. Rather it needs to be figured out and moved to the core of the business. So yes companies definitely need to be much more aggressive in adoption but also transformation. The current COVID crisis has made this an even greater imperative.
8. One of your five rules for thriving in the new AI age is that "capabilities are increasingly horizontal and universal." In my industry - early stage VC - we tend to look for companies that are focused on a very narrow target market. Do you think this will change for us? And will this new world lean to more winner-take-all markets, or not?
Universality is an important aspect of the Age of AI. What this means is that the skills and capabilities to build and scale an AI Factory and develop a business and operating model for companies that are organized around platforms and takes advantage of learning and network effects are going to be universal and needed broadly. For example, although the factory making hamburgers for a burger chain is going to be very different from a factory making cars for autonomous driving, the AI Factory is going to be similar. This means that, strategies for running a company in the Age of AI are going to be similar whether you are in retail, travel, or health care. Companies will need to be aware of the universality of both business and operating models and learn to compete in this way. It’s a massive change away from traditional industry specific capabitlities as the critical source of competitive advantage.
9. You include a chapter on the ethics of the new AI age, which I think is extremely important. This technology has legal, political, and ethical implications we haven't thought much about as a society. What worries you the most about the ethics of AI?
This was the hardest chapter for us to write and in fact one of the most important ones. We strongly believe that the question of AI ethics can’t be outsourced to the legal department or the philosophy department but is of central concern to executives and technologists inside of companies. This view is similar to what we have learned about other business issues like “quality.” It's too late If you wait to find defects in your products and services after they have been produced. Quality has to be designed into the process and products. Similarly the questions around ethics, bias and transparency need to be top most and designed-in from the beginning and not at the end. Companies really need to figure this out and not wait for the inevitable disasters before they get moving.
Thanks for reading, and enjoy the rest of your weekend!
@robmay
