The Case for Lean: The Principal Agent Problem

Jun 27, 2016

I’ve found myself advocating for Kanban methodologies as I’ve advised agile teams, and I’m working through my reasoning in the hopes of understanding it better myself. In the first article I talked about program oversight, and in the second I talked about team dynamics. This time I’m admitting to the fact that I went to business school as I talk about the principal agent problem.

At the University of Minnesota Carlson School of Management (Ski U Mah), I had a professor who wrote a book on franchising. In it, he quotes a cleaning company CEO with a great assessment of the joys of business. “You see, a manager will do what you want, but he won’t work very hard. A franchisee will work hard, but he won’t do what you want.”

In business school terms, this idea in general is called the “principal agent problem”. Let’s say you start a business. As the owner and sole employee, you want the business to get bigger. Now you hire your first employee. Obviously their interests are somewhat aligned with yours; if the business fails, they are out of a job. But their interests are not fully aligned with yours. As long as the business can afford to pay them, they don’t really need the business to grow. As Peter Gibbons says in Office Space, “that will only make someone work just hard enough not to get fired.”

So lots of professors have thought hard about this, and lots of managers live with it every day. And, of course, managers try to deal with it by measuring their employees’ output, and incentivizing them to increase their output (either with bonuses, raises, or with threats of getting fired). But as a business gets bigger, the direct impact of each employee on the growth of the business gets less and less direct. So it becomes harder and harder to use things like profit sharing or a direct equity stake to motivate someone’s behavior. Instead, managers measure things that the employee can control. But those things aren’t perfectly aligned with the growth of the business, so just because the employee can maximize that measurement (and their bonus) doesn’t mean the business benefits. And when employees know they’re being measured on something (or even when they think they’re being measured on something) they will change their behavior accordingly, even if that’s not the best thing for the business.

This is especially a problem where the work involves creativity and has qualitative attributes, which certainly is true in engineering. We software engineers can’t generally agree on a standard as to what makes some code better than other code, and even where we do agree on a standard it wouldn’t allow a useful comparison of two completely separate pieces of code written by different people to do different things. There’s also the problem introduced when measuring on individual output while people are working in teams, which can harm team dynamics.

You can’t make this problem go away. The best you can do is two things. First, make sure that, as much as possible, what you measure is well aligned with what is good for the business. Second, break the direct link between measurement and incentives at the level of the individual, both by measuring just at the team or organization level, and by clearly demonstrating that a wide variety of behaviors are considered valuable (not just individual output, but also teaching, mentoring, and making others more efficient).

One more digression, while on this topic, since this is a major hobby horse of mine. For those managers that think, “if you can’t measure it, you can’t manage it,” I would say first, Deming didn’t say that, and second, you’d better figure out how to manage it.

OK, so what does this have to do with agile and Kanban? Well, there has been a lot of work around measuring the performance of agile teams, and our perspective on the principal agent problem can help us understand and assess it. First, think about typical agile measurement used in a sprint methodology. The team ends up with a velocity per sprint, usually measured in story points. Note that this metric is unitless as long as story points are unitless. This is on purpose. The people who invented story points were quite aware of the principal agent problem and the fact that measuring something alters it, and set out to create something that is hard to use for comparing teams or measuring changes in performance over time.

Unfortunately, we’re not really left with much that can be used to measure value to the business. And that is a problem, because while we must manage all sorts of things we can’t measure, large organizations are going to have to manage a lot of things by measurement, because there’s just no other way to have a large organization. (Maybe the right answer would be to not have large organizations, but while we have them, that answer is of limited utility.)

So what ends up happening, in my experience, is that teams get encouraged to make their story points “unitful” so they can be compared across teams and over time. Or, some other method is used to achieve the same thing, like measuring completed stories or requirements. The disadvantage of this, in a sprint-based methodology, is that there is an incentive to declare a story “done” to improve the metrics, especially at the end of a sprint when the sprint goals are in danger. This kind of temptation is hard to resist, since that’s the only apparent thing that can make the metric look better.

On the other hand, if we’re using a Kanban methodology, what we are typically measuring is flow: how much time it takes to get some business requirement from the point at which we start working it until it is delivered to the customer. In this context, while there can still be an incentive to declare victory on something to improve the metrics, there are other, less-risky ways of improving the numbers. First, we can break work into smaller pieces so it can flow through the process more quickly. Second, we can identify and remove bottlenecks that are slowing things down.

The key insight about these two approaches is that they actually align with what is valuable for the business. Finding ways to break the work up so we can deliver quicker makes us more agile, so we can respond better to change. Removing bottlenecks removes waste from the process. Of course, it is possible to do both those things in a sprint-based methodology, and they would have similar positive results. But in a sprint-based methodology, the need to do them is less obvious because it’s not being directly measured.

I’m not suggesting that this will immediately fix what’s wrong with an organization. But what the organization measures and how the measurements are used have a significant impact on culture. Sometimes just starting to measure something is enough to identify that it is important.