18

Joshua (one of the authors) recently asked an early-stage machine learning company, “Why are you providing doctors with diagnoses?” The venture was building an AI tool that could tell a doctor whether a particular medical condition was present or not. A simple binary output. A diagnosis. The problem was, to be able to do that, the company had to obtain regulatory approval, which requires costly trials. To manage those trials, it was considering whether to partner with an established pharmaceutical or medical device company.

Joshua’s question was strategic rather than medical. Why did the venture have to provide a diagnosis? Instead, couldn’t it just provide the prediction? That is, the tool could analyze data and then tell the doctor that “there is an 80 percent chance the patient has the condition.” The physician could then explore precisely what was driving that conclusion and make the ultimate diagnosis—that is, the binary “present or not” outcome. The company could let the customer (in this case, the physician) do more.

Joshua suggested that the company focus on prediction rather than diagnosis. The boundary of its business would end with prediction. This obviated the need for regulatory approval, because physicians have many tools for arriving at a diagnostic conclusion. The company did not need to partner early on with established companies. Most critically, it no longer had to research and work out precisely how to translate the prediction into a diagnosis. All it had to deduce was the threshold accuracy required to deliver a valuable prediction. Was it 70, 80, or 99 percent?

Where does your business end and someone else’s begin? Where exactly are the boundaries of your company? This long-term decision requires careful attention at the organization’s very top level. Moreover, new general-purpose innovations often lead to new answers for the boundary question. Certain AI tools are likely to transform the boundaries of your business. Prediction machines will change how businesses think about everything, from their capital equipment to their data and people.

What to Leave In and What to Leave Out

Uncertainty has an impact on a business’s boundaries.1 Economists Silke Forbes and Mara Lederman looked at the organization of the US airline industry around the turn of the millennium.2 Major airlines like United and American handled some routes, while regional partners like American Eagle and SkyWest dealt with others. The partners were independent businesses that had contractual arrangements with the majors. Absent other considerations, the regional airlines typically operated at a lower cost than the majors, saving money on salaries and less beneficial work rules. For instance, some studies showed that senior pilots at the majors received 80 percent higher pay than those at their regional partners.

The puzzle is why majors rather than regional partners handle so many routes, given that partners can deliver the service at lower cost. Forbes and Lederman identified a driving factor—the weather—or, more specifically, uncertainty about the weather. When a weather event is out of the ordinary, it delays flights, which, in the tightly networked and capacity-managed airline industry, can have ripple effects throughout the entire system. When the weather goes sour, major airlines do not want to be hamstrung by partners checking their contracts when they have to make fast changes with uncertain costs. So, for routes where weather-related delays are likely, the majors retain control and operation.

The three ingredients we highlighted in the previous chapter suggest that AI might lead to strategic change. First, lower cost versus more control is a core trade-off. Second, that trade-off is mediated by uncertainty; specifically, the returns to control increase with the level of uncertainty. Major airlines balance lower cost and more control by optimizing the boundaries of where their own activities end and those of their partners begin. If a prediction machine could cut through this uncertainty, then the third ingredient would be present and the balance would shift. Airlines would contract more to their partners.

Businesses engaging in ongoing innovation, especially innovation that involves learning from experience, create a similar pattern. New automobile models are released approximately every five years, and because they involve detailed part specifications and design work, automakers need to know where the parts are coming from before release. Are they making parts themselves or outsourcing them? Throughout the long process of development, an automaker can only know so much about how a new model will perform. Some information can only be gathered after launch, like customer feedback and other long-term performance measurements. This is a key reason why models have annual updates that do not involve major changes in car design but offer improvements to components that work out kinks and improve the product.

Economists Sharon Novak and Scott Stern found that makers of luxury automobiles that manufactured their own parts improved faster from each model year to the next.3 They measured improvements at the customer end, using ratings from Consumer Reports. Having control meant automakers could adapt more readily to customer feedback. By contrast, those that outsourced parts did not show the same improvement. However, the latter received a different benefit; their initial models were of higher quality than the first models of automakers that made their own parts. The brand-new models of automakers that outsourced parts were better right out of the gate because the parts suppliers made better parts. Thus, automakers face the choice of outsourcing or making the parts themselves to reap improvements over time as they control innovation within the life cycle of their product model. Again, a prediction machine that reduces the uncertainty about customer needs could change the strategy.

In each case, the trade-off between short- and long-term performance and routine versus nonroutine events is resolved by a key organizational choice: how much to rely on external suppliers. But the salience of that choice is closely related to uncertainty. How important are weather events that airlines could not plan for up front? How will the vehicle match what customers really want?

Impact of AI: Capital

Let’s assume an AI is available that could reduce this uncertainty, so the third ingredient is in place. Prediction is so cheap that it minimizes uncertainty enough to change the nature of the strategic dilemma. How will this affect what the airlines and automakers do? AI might enable machines to operate in more complex environments. It expands the number of reliable “ifs,” thus lessening a business’s need to own its own capital equipment, for two reasons.

First, more “ifs” means that a business can write contracts to specify what to do if something unusual happens. Suppose that AI allows airlines not only to forecast weather events but to generate predictions for how best to deal with weather-related interruptions. This would increase the returns to major airlines for being more specific in their contracts to deal with contingencies. They can specify a greater number of “ifs” in the contracts. Thus, rather than controlling airline routes through ownership, the major airlines would have the predictive power to more confidently write contracts with independent regional carriers, allowing them to take advantage of those carriers’ lower costs. They would require less capital equipment (such as airplanes), because they could outsource more flights to the smaller regional carriers.

Second, AI-driven prediction—all the way to predicting consumer satisfaction—would enable automakers to more confidently design products up front, thus leading to high consumer satisfaction and performance without the consequent need for extensive mid-model adjustments. Consequently, automakers would be able to select the world’s best parts for their models from independent suppliers, confident that superior prediction up front was eliminating the need for costly contract renegotiations. The automakers would have less need to own factories that provide parts. More generally, prediction gives us many more “ifs” that we can use to clearly specify the “thens.”

This assessment holds the complexity of airline networks and automobile products as fixed. It could well be that up-front prediction gives airlines and automakers the confidence to allow for more complex arrangements and products. It is not clear what the impact on outsourcing would be since better prediction drives more outsourcing, while more complexity tends to reduce it. Which of these factors might dominate is hard to say at this stage. We can say that, while newly feasible complex processes might be done in-house, many of the simpler processes previously completed in-house will be outsourced.

Impact of AI: Labor

Banks rolled out the automatic teller machine (ATM), developed during the 1970s, extensively throughout the 1980s. The potentially labor-saving technology was—as the name implies—designed to automate tellers.

According to the Bureau of Labor Statistics, tellers were not being automated out of a job (see figure 18-1). However, they were automated out of the bank-telling task. Tellers ended up becoming the marketing and customer service agents for bank products beyond the collection and dispensing of cash. The machines handled that, more securely than humans. One reason banks did not want to open more branches was precisely because of the security issue and the human cost of spending time on something as transactional as bank telling. Freed from those constraints, bank branches proliferated (43 percent more in urban areas), in more shapes and sizes, and with them, a staff that was anachronistically called “tellers.”

The introduction of ATMs produced a significant organizational transformation; the new teller required a great deal more subjective judgment. The original teller tasks were, by definition, routine and easily mechanized. But the new tasks of talking to customers about their banking needs, advising them on loans, and working out credit card options were more complicated. In the process, evaluating whether the new tellers were doing a good job became harder.4

When performance measures change from objective (are you keeping the bank queues short?) to subjective (are you selling the right products?), human resource (HR) management becomes more complex. Economists will tell you that job responsibilities have to become less explicit and more relational. You will evaluate and reward employees based on subjective processes, such as performance reviews that take into account the complexity of the tasks and the employees’ strengths and weaknesses. Such processes are tough to implement because reliance on them to create incentives for good performance requires a great deal of trust. After all, a company can more easily decide to deny you that bonus, salary bump, or promotion based on a subjective review than when the performance measures are objective. However, when performance measures are objective in complex environments, critical mistakes can happen, as Wells Fargo’s experience with account managers’ fraud showed us so dramatically.5

The direct implication of this line of economic logic is that AI will shift HR management toward the relational and away from the transactional. The reason is twofold. First, human judgment, where it is valuable, is utilized because it is difficult to program such judgment into a machine. The rewards are either unstable or unknown, or require human experience to implement. Second, to the extent that human judgment becomes more important when machine predictions proliferate, such judgment necessarily involves subjective means of performance evaluation. If objective means are available, chances are that a machine could make such judgment without the need for any HR management. Thus, humans are critical to decision-making where the goals are subjective. For that reason, the management of such people will likely be more relational.

Thus, AI will have an impact on labor that is different from its impact on capital. The importance of judgment means that employee contracts need to be more subjective.

The forces affecting capital equipment also affect labor. If the key outputs of human labor are data, predictions, or actions, then using AI means more outsourced contract labor, just as it means more outsourced equipment and supplies. As with capital, better prediction gives more “ifs” that we can use to clearly specify the “thens” in an outsourcing contract.

However, the more important effect on labor will be the increasing importance of human judgment. Prediction and judgment are complements, so better prediction increases the demand for judgment, meaning that your employees’ main role will be to exercise judgment in decision-making. This, by definition, cannot be well specified in a contract. Here, the prediction machine increases uncertainty in the strategic dilemma because evaluating the quality of judgment is difficult, so contracting out is risky. Counterintuitively, better prediction increases the uncertainty you have over the quality of human work performed: you need to keep your reward function engineers and other judgment-focused workers in-house.

Impact of AI: Data

Another critical strategic issue is the ownership and control of data. Just as the consequences for workers relate to the complementarity between prediction and judgment, the relationship between prediction and data also drives these trade-offs. Data makes prediction better. Here, we consider the trade-offs associated with organizational boundaries. Should you utilize others’ data or own your own? (In the next chapter, we explore issues concerning the strategic importance of investing in data collection.)

For AI startups, owning the data that allows them to learn is particularly crucial. Otherwise, they will be unable to improve their product over time. Machine learning startup Ada Support helps other companies interact with their customers. Ada had the opportunity to integrate its product into the system of a large established chat provider. If this worked, it would be much easier to get traction and establish a large user base. This was a tempting way to go.

The problem, however, was that the established companies would own the feedback data on the interactions. Without that data, Ada would not be able to improve its product based on what actually happened in the field. Ada was emboldened to reconsider this approach and did not integrate until it could ensure that it owned the resulting data. Doing so gave it a pipeline of data now and into the future to draw on for continual learning.

The issue of whether to own or procure data goes well beyond startups. Consider data designed to help advertisers target potential customers. John Wanamaker, who, among others, created the modern structure of advertising in the media, once stated: “Half the money I spend on advertising is wasted; the trouble is, I don’t know which half.”

This is the fundamental issue with advertising. Put an advertisement on a website, everyone who visits that site views the ad, and you pay for each impression. If only a fraction of them are potential customers, then your willingness to pay for each impression will be relatively low. That is a problem for both you as the advertiser and the website trying to make money from ads.

One solution is to focus on building websites that attract people with specific interests—sports, finance, and so on—which have a higher proportion of potential customers for certain types of advertisers. Before the rise of the internet, this was a core feature of advertising, leading to a proliferation of magazines, cable television channels, and newspaper sections for automotive, fashion, real estate, and investing. However, not every media outlet can tailor its content in this way.

Instead, thanks to web browser innovations, primarily the “cookie,” advertisers can track users over time and across websites. They then have the ability to better target their advertising. The cookie records information about website visitors but, most critically, information about the type of sites, including shopping sites, they frequent. Because of this tracking technology, when you visit a site to look for new pants, you may find that a disproportionate share of subsequent ads you see, including on completely unrelated sites, is for pants.

Any website can place cookies, but the cookies are not necessarily of much value to that site. Instead, websites offer cookies for sale to advertising exchanges (or sometimes directly to advertisers) so that they can better target their ads. Websites sell data about their visitors to companies that place advertisements.

Companies buy data because they can’t collect it themselves. Not surprisingly, they buy data that helps them identify high-value customers. They also may buy data that helps them avoid advertising to low-value customers. Both types of data are valuable in that they enable the company to focus its ad spending on high-value customers.6

Many AI leaders, including Google, Meta, and Microsoft, have built or purchased their own advertising networks so that they can own this valuable data. They decided that owning this data is worth the cost of acquiring it. To others, advertising data is less critical, so they trade off the control of that data to avoid incurring the high cost of collecting it themselves; the advertising data thus remains outside the boundaries of these companies.

Selling Predictions

Google, Meta, Microsoft, and a handful of other companies have particularly useful data on consumer preferences online. Rather than only sell data, they go a step further to make predictions for advertisers. For example, Google, through search, YouTube, and its advertising network, has rich data on user needs. It does not sell the data. However, it does, in effect, sell the predictions that the data generates to advertisers as part of a bundled service. If you advertise through Google’s network, your ad is shown to the users that the network predicts are most likely to be influenced by the ad. Advertising through Meta or Microsoft yields similar results. Without direct access to the data, the advertiser buys the prediction.

Unique data is important for creating strategic advantage. If data is not unique, it is hard to build a business around prediction machines. Without data, there is no real pathway to learning, so AI is not core to your strategy. As noted in the example of advertising networks, predictions still might be useful. They allow the advertiser to target the highest-value customer. Thus, better prediction may help an organization, even if the data and predictions are not likely to be sources of strategic advantage.7 Both the data and the prediction are outside the boundaries of the organization, but it can still use prediction.

The main implication here is that data and prediction machines are complements. Thus, procuring or developing an AI will be of limited value unless you have the data to feed it. If that data resides with others, you need a strategy to get it.

If the data resides with an exclusive or monopoly provider, then you may find yourself at risk of having that provider appropriate the entire value of your AI. If the data resides with competitors, there may be no strategy that would make it worthwhile to procure it from them. If the data resides with consumers, it can be exchanged in return for a better product or higher-quality service.

However, in some situations, you and others might have data that can be of mutual value; hence, a data swap may be possible. In other situations, the data may reside with multiple providers, in which case, you might need some more complicated arrangement of purchasing a combination of data and prediction.

Whether you collect your own data and make predictions or buy them from others depends on the importance of prediction machines to your company. If the prediction machine is an input that you can take off the shelf, then you can treat it like most companies treat energy and purchase it from the market, as long as AI is not core to your strategy. In contrast, if prediction machines are to be the center of your company’s strategy, then you need to control the data to improve the machine, so both the data and the prediction machine must be in-house.

At the beginning of this chapter, we suggested that a machine learning startup that aimed to provide medical diagnoses instead sell a prediction. Why would the doctor be willing to buy the prediction rather than the full diagnosis? And why wouldn’t the doctor want to own the prediction machine and data? The answers lie in the relevant trade-offs we’ve discussed. A key part of the doctor’s job is diagnosis, so buying the prediction is not a doctor’s core strategic decision. Doctors continue to do what they did before, with an additional piece of information. If it isn’t a key strategic decision, then they can buy the prediction without needing to own the data or prediction. In contrast, the essence of the startup is AI, and the prediction provides value to customers. So, as long as the startup owns the data and prediction machine, it does not need to own the diagnosis. The boundary between the startup and the doctor is the boundary where the AI ceases to be strategic and instead is simply an input to a different process.

KEY POINTS

• A key strategic choice is determining where your business ends and another business begins—deciding on the boundary of the firm (e.g., airline partnerships, outsourcing automotive part manufacturing). Uncertainty influences this choice. Because prediction machines reduce uncertainty, they can influence the boundary between your organization and others.
• By reducing uncertainty, prediction machines increase the ability to write contracts and thus increase the incentive for companies to contract out both capital equipment and labor that focuses on data, prediction, and action. However, prediction machines decrease the incentive for companies to contract out labor that focuses on judgment. Judgment quality is hard to specify in a contract and difficult to monitor. If judgment could be well specified, then it could be programmed and we wouldn’t need humans to provide it. Since judgment is likely to be the key role for human labor as AI diffuses, in-house employment will rise and contracting out labor will fall.
• AI will increase incentives to own data. Still, contracting out for data may be necessary when the predictions that the data provides are not strategically essential to your organization. In such cases, it may be best to purchase predictions directly rather than purchase data and then generate your own predictions.