By the early 1990s, the Sears catalog—the phone book–sized tome that once showed American families the latest in clothing and home appliances—was hemorrhaging money. Analysts had argued for years that Sears should use its resources to enhance its brick-and-mortar stores, rather than keep the mail-order division afloat.

So why did Sears continue making the catalog? The short answer: legacy.

The catalog had once been responsible for a huge portion of Sears’s revenue. Consequently, the catalog division had been granted more and more power every year, until the company’s leadership could not overrule them, even when it knew it should.

“Then the question is, ultimately, why did they have more power?” says Michael Powell, an associate professor of strategy at the Kellogg School. “Why didn’t Sears just take it away?”

Powell set out to investigate odd power dynamics like this in a new paper with fellow Kellogg strategy professor Niko Matouschek, and Jin Li of the London School of Economics. By modeling how business relationships evolve, the authors find that handing over immense power to certain employees may not be irrational on a company’s part. Indeed, it stems from higher-ups keeping their promises to reward employees for work well done.

Over a long-enough time horizon, the researchers found, extreme power dynamics are inevitable. The study concludes that a manager with the firm’s best interests in mind who can’t simply dole out financial incentives will eventually give subordinates either full discretion or zero discretion over their domain—any other path would make both the company and its employees worse off. This same dynamic plays out at every level, from the CEO and her direct reports to a store manager and her staff.

Power Dynamics and Employee Incentives

To understand the team’s research methods, consider this hypothetical:

You are a senior manager at a tech company, trying to plot your firm’s next move. Should you venture into manufacturing printers? Or smartphones? Or just stick to laptops, your current bread and butter?

As a leader, you want to choose the project that will be best for the firm. The problem is, you do not know which one that is. So you ask a well-informed subordinate to make a recommendation about which project to choose.

“The person in the C-suite might not have good sense for all these possibilities,” Powell says. “But the person on the ground, they have good sense for it.”

This subordinate knows full well which particular project is best for the company. But his choice is not so simple—after all, a different project may be best for him personally. (Perhaps this project is more likely to ensure the employee’s own job security, or is easier, or more fun.)

“The best way to reward people is to backload their rewards as much as possible.”

In the end, the employee’s recommendation comes back to you, the manager. You can either accept the recommendation, or reject it.

Clearly, you need to provide an incentive for the employee to recommend the project that is truly best for the company. But you are short on cash. Or perhaps your company has rules limiting what you can offer as a monetary bonus. So, you make him a promise: he will get more autonomy to choose the projects he likes later on if—and only if—he makes a decision that benefits the firm in the immediate future.

“We reward people by making it so that their future within the firm is going to look a little brighter,” says Powell.

On the other hand, if the employee’s recommendation turns out not to benefit the company, then you promise to take away his power in the future, by ignoring his recommendations.

The authors wrote this decision scenario in mathematical language. How, they wondered, would the manager–employee relationship evolve over time? Could it explain why Sears’ executives had given the heads of the catalog division carte blanche?

Keeping Promises to Employees

Their model allows them to think through the consequences of any recommendation on strategy by the employee. For instance, the employee could recommend the company’s optimal project every time, or selfishly recommend his own favorite project, despite the consequences.

If power is managed optimally, the relationship always begins the same way: the manager rewards or punishes the employee based on how frequently the project he picks ends up being good for the company, and the employee’s future power prospects fluctuate up and down accordingly.

But over time, those accumulated rewards or punishments reach the point where something unexpected happens.

“The relationship evolves into what is referred to as a steady state,” Powell says. That is, depending on how the employee’s recommendation pans out, he will eventually end up being promised full autonomy or zero autonomy to select which projects he and the firm will implement. When the manager delivers on that promise, she will be left either rubber-stamping or rejecting his every recommendation.

This counterintuitive finding derives from complicated math, but the underlying intuition is simple: in order to keep incentivizing the employee, the firm has to promise to give (or take away) more and more power each time the scenario is repeated, until it reaches a point when complete autonomy is on the table.

And, as the case of the Sears catalog shows, such power dynamics in organizations are not necessarily a good thing for the organization in the long run. In theory, an all-powerful employee’s preferences could align with the company’s best interests—but odds are, that will often not be the case.

There is one way a manager could avoid getting locked into this power relationship. The paper assumes that the manager is doing what is best for the firm. But, in principle, a manager could act differently—she could renege on past promises, for example, to avoid a complete transfer of power.

However, this comes at an expense.

“It turns out, that’s not what you want to do,” Powell explains. Without strong, reliable incentives, the employee no longer has reason to consider the firm’s welfare or even stay on at the firm. As a result, the firm suffers, bringing the manager down with it.

The study gives some insight into how to incentivize employees. Once a manager delivers on her long-awaited promises, she will no longer have any leverage over the employee. So, the longer the manager can delay having to hand over absolute autonomy, the longer she will have an employee who still cares about doing what is right for the company.

“The best way to reward people is to backload their rewards as much as possible,” Powell says—and the same goes for their punishments.

Understanding Nonmonetary Incentives

Powell and Matouschek’s findings offer a plausible answer to another question that puzzles strategy experts: Why do companies keep making bad decisions even after they realize that they are bad?

Eventually, managers no longer need to depend on subordinates to tell them what is best—they can just look at their competitors making smarter choices and reaping the benefits. Powell returns to the history of the Sears catalog: “At some point it did become public knowledge that this was not good for the firm overall, and even the firm knows that.”

Yet, a promise is still a promise. Without a new incentive scheme, the all-powerful employee has no reason to change course.

The use of promises as rewards fascinates Powell, who sees this research as part of a broader mission to understand nonmonetary incentives and their frequently strange consequences. In a related paper, he used game theory to explore how promise-keeping can lead companies to make imperfect hires.

Such work stands to shed light on a little-understood mechanism that is quietly, but powerfully, shaping how employees behave at all levels and across all industries.

“When I was working at In-N-Out Burger back in the day, if I was nice to customers, at some point they would maybe let me be a supervisor, too,” Powell says. “This notion of future power as a reward does resonate.”