Organization Design

The Multitask Problem and Why Paying for One Thing Kills the Other

Learn the result that explains why so many well-designed bonus schemes make a business worse: when a job has parts you cannot measure, the correct strength of the incentive on the parts you can measure is lower than intuition says — and sometimes it is exactly zero.

  • Expert
  • 15 min total
  • 14 chapters

What decision this helps you make: Whether to attach an incentive to a measurable part of a job at all, how strong it can safely be given what else the role is for, and whether the real fix is a different plan or a different job.

What this topic is

The multitask problem is what happens when one person allocates limited effort across several activities and only some of those activities can be measured. Holmström and Milgrom formalised it in 1991. The result is counter-intuitive and load-bearing: the right strength of an incentive on a measurable task does not depend only on how well you can measure that task. It also depends on what else the person is supposed to be doing, on how valuable those unmeasured activities are, and on how easily effort moves between them. When the tasks compete for the same attention and one of them cannot be measured, the optimal incentive on the measurable one falls — and in the limiting case, a flat wage with no incentive at all is the correct contract.

Why it matters

Nearly every real job is multitask. A salesperson sells and also keeps a customer honest about what the product does. A support agent closes tickets and also protects a relationship. An engineer ships features and also leaves a codebase that someone else can work in. Every one of those jobs has a measurable half and an unmeasurable half, and standard incentive reasoning — better measurement, steeper slope — makes the second half worse. This is the single most expensive error in compensation design because the damage lands precisely where nothing is watching, and it therefore never appears in a review of the plan.

Who should learn it

Anyone designing an incentive for a role that is more than one thing, anyone who has seen a bonus scheme produce exactly the measured result and a visibly worse business, and any operator trying to understand why large organisations pay so many people flat salaries when the theory of incentives seems to say otherwise.

What you will understand

  • The Holmström-Milgrom result and why substitutability between tasks, not just measurement noise, sets the optimal slope
  • The equal compensation principle, and the constraint it places on any plan covering more than one activity
  • The three repairs that are not "a better bonus": job design, asset ownership, and deliberately weak incentives
  • How to compute whether an incentive on the measurable task is worth its damage to the rest of the role

Prerequisites

Common misconception

"If the bonus is producing bad behaviour, the measure must be wrong — find a better one." Sometimes true, and it misses the harder half of the result. Even a perfect, noiseless measure of one task can be destructive if the job contains a second task you cannot measure at all, because the problem is not the accuracy of the measure but the pull it exerts on finite attention. Holmström and Milgrom's sharpest conclusion is that a flat wage can be strictly optimal when tasks are substitutes and one is unmeasurable — not as a compromise, not as a failure of nerve, but as the right answer. The second misconception is that this is the same as Goodhart's law. It is not: Goodhart is about a measure drifting from its own goal, while multitasking is about a measure working perfectly and starving everything beside it.