Measuring and Managing Performance in Organizations

by Robert D. Austin

A US employment agency faced the problem of improving the number of successfully placed candidates in the 1950s. Management followed the maxim that "you can only manage what you measure". You interview people, find a matching job, place them and if they are still on the job half a year later ... success! Since it was difficult to get this information from the companies half a year after a placement (they had no interest in answering questions, as their needs were filled), management measured the number of placements and ratio of interviews to placements instead as a proxy. The result: not only did interviewers place unqualified candidates, they also destroyed records of interviews that did not lead to placements.

It's quite simple: If you reward what you measure, and you can economically only measure a proxy for your goal, what you get is improvement of the proxy, not the goal. In extreme cases, this may actually hurt your goal.

The goal of the principal or employer differs from the goal of the agent or employee. The principal has just one incentive: he wants to fulfill the needs of his customer and by this obtain profit. The agent, doing the work for him, has two conflicting goals: he wants to fulfill the needs of the customer and he also wants to fulfill the needs of his principal.

To deal with that situation, the principal has two options: he can try to measure and control what the agent is doing. Or he can trust his employee to do the best thing for the customer. In the first case, he will have costs for measuring and controlling, and as we have seen, for inducing dysfunctional behavior. In the second case he will have agency costs if the employee decides not to work as hard as he could.

Work that is repetitive, simple and does not need a lot of specialized knowledge, has a better chance to be usefully measured. Work based on long-term relationships, personal relationships as for example see in small companies, mutual commitment and frequent interactions tends to have lower delegation costs.

If you want to use measurement in complex environments like software development for learning how do, be wary on tying it directly to incentives. It may be more useful, to give it to employees so they can understand how things go, but not give it, or give it only in an anonymous, aggregate form for entire teams to the team's management.

Take home message: Measurement is about people. Thinking "I will just select what I measure very carefully" is setting yourself up for hurt. People adapt. They are smarter than your carefully selected measure.

I am amazed how smart free markets and competition are, in that they are able to weed out such dysfunctional behavior between buyer and seller. How can that be applied to the agent-principal relationship?