17

In January 2007, when Steve Jobs paced the stage and introduced the iPhone to the world, not a single observer reacted by saying, “Well, it’s curtains for the taxi industry.” Yet fast forward to 2018 and that appears to be precisely the case. Over the last decade, smartphones evolved from being simply a smarter phone to an indispensable platform for tools that are disrupting or fundamentally altering all manner of industries. Even Andy Grove, who famously quipped that “only the paranoid survive,” would have to admit that you would’ve been pretty darn paranoid to have foreseen how far and wide the smartphone would reach into some very traditional industries.

The recent developments in AI and machine learning have convinced us that this innovation is on par with the great, transformative technologies of the past: electricity, cars, plastics, the microchip, the internet, and the smartphone. From economic history, we know how these general-purpose technologies diffuse and transform. We also realize how hard it is to forecast when, where, and how the most disruptive changes will take place. At the same time, we have learned what to look for, how to be ahead of the curve, and when a new technology is likely to transition from something interesting to something transformative.

When should AI be a critical agenda item for your organization’s leadership team? While ROI calculations can influence operational changes, strategic decisions pose dilemmas and force leaders to grapple with uncertainty. Adopting AI in one part of the organization might require changes in another part. For intraorganizational effects, adoption and other decisions require the authority of someone who oversees the entire business, namely, the CEO.

So when is AI likely to fall into this category? When does a fall in the cost of prediction matter enough that it will change strategy? And what dilemma is a CEO likely to face if this should happen?

How AI Can Change Business Strategy

In chapter 2, we conjectured that once the dial on the prediction machine had been turned up enough, companies such as Amazon would be so confident about what particular customers want that their business model could change. They would move from a shopping-then-shipping model to shipping-then-shopping, sending items to customers in anticipation of their wants. This scenario neatly illustrates three ingredients that together could cause investment in that AI tool to rise to the level of being a strategic rather than operational decision.

First, a strategic dilemma or trade-off must exist. For Amazon, the quandary is that shipping-then-shopping may generate more sales but simultaneously produce more goods consumers want to return. When the cost of returning items is too high, then the ROI for shipping-then-shopping is lower than the ROI for the traditional approach of shopping-then-shipping. This explains why, in the absence of some technological change, Amazon’s business model remains shopping-then-shipping rather than the other way around, just like almost every other retailer.

Second, the problem can be resolved by reducing uncertainty. For Amazon, it is about consumer demand. If you can accurately forecast what people will purchase, especially if delivered to their doorsteps, then you reduce the likelihood of returns and increase sales. Uncertainty reduction hits both the benefit and the cost sides of the dilemma.

This type of demand management is not new. It’s one reason that physical stores exist. Physical stores cannot forecast individual customer demand, but they can forecast the likely demand from a group of customers. By pooling together the customers who visit a location, physical stores hedge demand uncertainty among individual customers. Moving to a shipping-then-shopping model based on individual homes requires more information about individual customer demand, which can overcome the competitive advantage physical stores have.

Third, companies require a prediction machine that can reduce uncertainty enough to change the balance in the strategic dilemma. For Amazon, a very accurate model of customer demand may make the shipping-then-shopping business model worthwhile. Here, the benefits of increased sales outweigh the costs of returns.

Now, if Amazon were to implement this model, it would make further changes in its business. These would include, for example, investments to reduce the cost of securing packages left for pickup and transportation services to handle returns. Although the customer-friendly delivery market is competitive, product return services are a much less-well-developed market. Amazon itself might establish an infrastructure of trucks that visit neighborhoods daily for deliveries and returns, thus vertically integrating into the daily product return business. Effectively, Amazon could move the boundary of its business right up to your front porch.

This boundary shifting is already occurring. One example is the German e-commerce venture Otto.1 A major barrier to consumer purchases over the internet rather than in a store is uncertain delivery times. If consumers have a poor delivery experience, they are unlikely to return to a site. Otto found that when deliveries were delayed (that is, took longer than a few days), returns shot upward. Consumers would inevitably find the product at a store in the meantime and purchase it there. Even when Otto had sales, returns added to its costs.

How do you reduce the time to deliver products to consumers? Anticipate what they are likely to order and have it in stock at a distribution center nearby. But such inventory management is itself costly. Instead, what you want is to hold only the inventory you are likely to need. You want a better prediction of consumer demand. Otto, with a database of 3 billion past transactions and hundreds of other variables (including search terms and demographics), was able to create a prediction machine to handle the forecast. It can now predict with 90 percent accuracy what products it will sell within a month. Relying on those forecasts, it revamped its logistics. Its inventory declined by 20 percent, and annual returns dropped by 2 million items. Prediction improved logistics, which in turn reduced costs and increased consumer satisfaction.

Once again, we can see the three ingredients of strategic importance. Otto had a dilemma (how to improve delivery times without expensive inventory holdings), uncertainty drove the dilemma (in this case, overall customer demand in a location), and by resolving that uncertainty (e.g., forecasting local demand better), it could set up a new way of organizing logistics, requiring new warehouse locations, local shipping, and customer delivery guarantees. It could not have accomplished all this without using a prediction machine to resolve that key uncertainty.

Sweet Home Alabama?

For a prediction machine to change your strategy, someone has to create one that is useful to you in particular. Doing so depends on several things outside your organization’s control.

Let’s look at the factors that might make prediction technology available to your business. To do this, we are going to travel to the cornfields of Iowa in the 1930s. There, some pioneering farmers introduced a new form of corn that they created through extensive crossbreeding for the better part of two decades. This hybrid corn was more specialized than ordinary commercial corn. It required crossing two inbred lines of corn to improve properties such as drought resistance and local environment-specific yields. The hybrid corn was a critical change because not only did it promise dramatically higher yields, but the farmer became dependent on others for the special seeds. The new seeds needed to be tailored to local conditions to yield their full benefits.

As shown in figure 17-1, Alabama farmers appeared to be laggards compared to those in Iowa. But when Harvard economist Zvi Griliches looked closely at the numbers, he found that the twenty-year lag between Iowa and Alabama widespread adoption was not because Alabama famers were slow, but rather because the ROI for hybrid corn for Alabama farms did not justify its adoption in the 1930s.2 Alabama farms were smaller, with thin profit margins compared to those north and west. By contrast, Iowa farmers could apply a successful seed across their larger farms and reap larger benefits to justify the higher seed costs. A big farm meant experimentation with new hybrid varieties was easier because the farmer had to set aside only a small portion of the property until the new varieties proved effective.3 The Iowa farmers’ risks were lower, and they had healthier margins to act as a buffer. Once enough farmers in an area adopted the new seeds, seed markets became thicker with more buyers and sellers and the cost of selling the seeds fell, so the risks of adoption were reduced further still. Eventually, corn farmers across the United States (and worldwide) adopted hybrid seeds as the costs fell and the perceived risks diminished.

In the AI world, Google is Iowa. It has more than a thousand AI tool development projects underway across every category of its business, from search to ads to maps to translation.4 Other tech giants worldwide have joined Google. The reason is fairly obvious: Google, Meta (Facebook), Baidu, Alibaba, Salesforce, and others are already in the tools business. They have clearly defined tasks that extend throughout their enterprises, and in each, AI can sometimes dramatically improve a predictive element.

Those enormous corporations have big profit margins, so they can afford to experiment. They can take a part of the “land” and devote it to many new AI varieties. They can reap huge rewards from successful experiments by applying them across a wide range of products operating at large scale.

For many other businesses, the path to AI is less clear. Unlike Google, many have not made two decades’ worth of investments in digitizing all aspects of their workflow and also do not have a clear notion of what they want to predict. But once a company sets well-defined strategies, it can develop those ingredients, laying the groundwork for effective AI.

When the conditions were right, all corn farmers in Wisconsin, Kentucky, Texas, and Alabama eventually followed their Iowa peers in adopting hybrid corn. The demand-side benefits were high enough, and the supply-side costs had fallen. Similarly, the costs and risks associated with AI will fall over time, so many businesses not at the forefront of developing digital tools will adopt it. In doing so, the demand side will drive them: the opportunity to resolve fundamental dilemmas in their business models by reducing uncertainty.

Complementing Baseball Players

Billy Beane’s Moneyball strategy—using statistical prediction to overcome the biases of human baseball scouts and improve prognostication—was an example of using prediction to reduce uncertainty and improve the performance of the Oakland Athletics. It was also a strategic change that required altering the organization’s implicit and explicit hierarchy.

Better prediction changed who the team hired on the field, but the operation of the baseball team itself did not change. The players that the prediction machine selected played much the same way as the players it replaced, with perhaps a few more walks thrown in. And the scouts continued to have a role in player selection.5

The more fundamental change occurred in who the team hired off the field and the resulting restructuring of the organizational chart. Most important, the team hired people who could tell the machines what to predict and then use those predictions to determine which players to acquire (most notably, Paul DePodesta, as well as others whose contributions were combined in the “Peter Brand” character played by Jonah Hill in the movie). The team also created a new job function, called a “sabermetric analyst.” A sabermetric analyst develops measures for the rewards that the team would receive from signing different players. Sabermetric analysts are baseball’s reward function engineers. Now, most teams have at least one such analyst, and the role has appeared, under different names, in other sports.

Better prediction created a new high-level position on the org chart. The research scientists, data scientists, and vice presidents of analytics are listed as key roles in the online front office directories. The Houston Astros even have a separate decision sciences unit headed by former NASA engineer Sig Mejdal. The strategic change also means a switch in who the team employs to pick the players. These analytics experts have mathematical skills, but the finest of them understand best what to tell the prediction machine to do. They provide judgment.

Returning to the simple economics that underlies all the arguments in this book, prediction and judgment are complements; as the use of prediction increases, the value of judgment rises. Teams are increasingly bringing in new senior advisers who sometimes may not have firsthand experience playing the game and—true to stereotype—may not be an obvious fit in the jock world of professional sports. However, even nerds recruited into this setting require a deep understanding of the game because using prediction machines in sports management means an increase in the value of people who have the judgment to determine payoffs and, therefore, the judgment to use predictions in decisions.

Strategic Choice Requires New Judgment

The change in the organization of baseball team management highlights another key issue for the C-suite in implementing strategic choices with regard to AI. Before sabermetrics, baseball scouts’ judgment was limited to the pros and cons of individual players. But using quantitative measures made it possible to predict how groups of players would perform together. Judgment shifted from thinking about the payoff of a particular player to thinking about the payoff to a particular team. Better prediction now enables the manager to make decisions that are closer to the organization’s objectives: determining the best team rather than the best individual players.

To make the most of prediction machines, you need to rethink the reward functions throughout your organization to better align with your true goals. This task is not easy. Beyond recruiting, the marketing of the team needs to change, perhaps to de-emphasize individual performance. Similarly, the coaches have to understand the reasons for individual players’ recruitment and the implications for team composition in each game. Finally, even the players need to understand how their roles might change depending on whether their opponents have similarly adopted new prediction tools.

Advantages You May Already Have

Strategy is also about capturing value—that is, who will capture the value that better prediction creates?

Business executives often claim to us that because prediction machines need data, data itself is a strategic asset. That is, if you have many years of data on, say, yogurt sales, then in order to predict yogurt sales using a prediction machine, someone will need that data. Hence, it is valuable to its owner. It is like having a repository of oil.

That presumption belies an important issue—like oil, data has different grades. We have highlighted three types of data—training, input, and feedback data. Training data is used to build a prediction machine. Input data is used to power it to produce predictions. Feedback data is used to improve it. Only the two latter types are needed for future use. Training data is used at the beginning to train an algorithm, but once the prediction machine is running, it is not useful anymore. It is as if you have burned it. Your past data on yogurt sales has little value once you have a prediction machine built on it.6 In other words, it may be valuable today, but it is unlikely to be a source of sustained value. To do that you either need to generate new data—for input or feedback—or you need another advantage. We will explore the advantages of generating new data in the next chapter and focus on other advantages right now.

Dan Bricklin, the spreadsheet inventor, created enormous value, but he is not a rich person. Where did the spreadsheet value go? On the wealth rankings, imitators such as Lotus 1-2-3 founder Mitch Kapor or Microsoft’s Bill Gates certainly far outstripped Bricklin, but even they were appropriating a small fraction of the spreadsheet’s value. Instead, the value went to users, to the businesses that deployed spreadsheets to make billions of better decisions. No matter what Lotus or Microsoft did, their users owned the decisions that the spreadsheets were improving.

Because they operate at the decision level, the same is true for prediction machines. Imagine applications of AI that would greatly assist in inventory management for a supermarket chain. Knowing when yogurt is going to sell helps you know when you should stock it and minimizes the amount of unsold yogurt to discard. An AI innovator who offers prediction machines for yogurt demand could do well, but would have to deal with a supermarket chain in order to create any value. Only the supermarket chain can take the action that stocks yogurt or not. And without that action, the prediction machine for yogurt demand has no value.

Many businesses will continue to own their actions with or without AI. They will have an advantage in capturing some of the value that arises from adopting AI. This advantage does not mean that the companies that own the actions will capture all the value.

Before selling their spreadsheet, Bricklin and his partner, Bob Frankston, wondered whether they should keep it. They could then sell their modeling skills and, as a result, capture the value created by their insights. They abandoned this plan—likely for good reason—but in AI, this strategy might work. AI providers may try to disrupt traditional players.

Autonomous vehicles are an example, to some degree. While some traditional carmakers are aggressively investing in their own capabilities, others are hoping to partner with those outside the industry (such as Alphabet’s Waymo) rather than develop those capabilities in-house. In other cases, large technology companies are initiating projects with traditional carmakers. For example, Baidu, operator of China’s largest search engine, is leading a large and diversified open autonomous driving initiative, Project Apollo, with several dozen partners, including Daimler and Ford. In addition, Tencent Holdings, owner of WeChat, which has almost a billion monthly active-user accounts, is leading an autonomous driving alliance that includes prominent incumbents, such as Beijing Automotive Group. Chen Juhong, a vice president of Tencent, remarked, “Tencent hopes to make an all-out effort to reinforce the development of AI technologies used in autonomous driving…. We want to be a ‘connector’ to help accelerate cooperation, innovation and industry convergence.”7 Reflecting on the competitive pressures driving collaboration, Beijing Automotive chairman Xu Heyi said, “In this new era, only those who connect with other companies to build the next generation of cars will survive, while those who shut themselves up in a room making vehicles will die.”8 Relatively new entrants (such as Tesla) are competing with incumbents by directly deploying AI in new cars that tightly integrate software and hardware. Companies like Uber are using AI to develop autonomy with the hope of taking even the driving decisions out of consumers’ hands. In that industry, the race for value capture does not respect traditional business boundaries. Instead, it challenges the ownership of actions that might otherwise have been an advantage.

The Simple Economics of AI Strategy

The changes we’ve highlighted depend on two different aspects of AI impact at the core of our economic framework.

First, as in the shipping-then-shopping thought experiment, prediction machines reduce uncertainty. As AI advances, we’ll use prediction machines to reduce uncertainty more broadly. Hence, strategic dilemmas driven by uncertainty will evolve with AI. As the cost of AI falls, prediction machines will resolve a wider variety of strategic dilemmas.

Second, AI will increase the value of the complements to prediction. A baseball analyst’s judgment, a grocery retailer’s actions, and—as we will show in chapter 19—a prediction machine’s data become so important that you may need to change your strategy to take advantage of what it has to offer.

KEY POINTS

• C-suite leadership must not fully delegate AI strategy to their IT department because powerful AI tools may go beyond enhancing the productivity of tasks performed in the service of executing against the organization’s strategy and instead lead to changing the strategy itself. AI can lead to strategic change if three factors are present: (1) there is a core trade-off in the business model (e.g., shop-then-ship versus ship-then-shop); (2) the trade-off is influenced by uncertainty (e.g., higher sales from ship-then-shop are outweighed by higher costs from returned items due to uncertainty about what customers will buy); and (3) an AI tool that reduces uncertainty tips the scales of the trade-off so that the optimal strategy changes from one side of the trade to the other (e.g., an AI that reduces uncertainty by predicting what a customer will buy tips the scale such that the returns from a ship-then-shop model outweigh those from the traditional model).
• Another reason C-suite leadership is required for AI strategy is that the implementation of AI tools in one part of the business may also affect other parts. In the Amazon thought experiment, a side effect of transitioning to a ship-then-shop model was vertical integration into the returned items collection business, perhaps with a fleet of trucks that did weekly pickups throughout the neighborhood. In other words, powerful AI tools may result in significant redesign of workflows and the boundary of the firm.
• Prediction machines will increase the value of complements, including judgment, actions, and data. The increasing value of judgment may lead to changes in organizational hierarchy—there may be higher returns to putting different roles or different people in positions of power. In addition, prediction machines enable managers to move beyond optimizing individual components to optimizing higher-level goals and thus make decisions closer to the objectives of the organization. Owning the actions affected by prediction can be a source of competitive advantage that allows traditional businesses to capture some of the value from AI. However, in some cases, where powerful AI tools provide a significant competitive advantage, new entrants may vertically integrate into owning the action and leverage their AI as a basis for competition.