Pay, and how it is set · 3.6

Bonuses, and whether they work

Bonuses, and whether they work. What is actually the case, and how it compares with what is repeated.

For a practical software perspective on this topic, Monitask's practical guide provides a concise companion guide.

Variable pay is worth what it changes. The evidence on what it changes is much better than the enthusiasm with which it is adopted.

Where it works

Simple, measurable, individually attributable tasks where quality is observable and where the worker controls the output. The piece rate entry covers the strongest form and the evidence there is unambiguous: output rises.

Group and profit-sharing schemes show smaller and more mixed effects, because the link between an individual's effort and the payout is weak, which is a straightforward consequence of arithmetic rather than a mystery.

Where it does not

Complex work, interdependent work, work whose quality is only visible later, and work where the measurable part is a poor proxy for the valuable part. Which is most professional work.

The failure mode is not that the bonus does nothing. It is that it works: it moves effort towards whatever is measured, and if the measure is a proxy, effort moves towards the proxy and away from the thing it was standing in for.

Crowding out

Introducing a payment for something previously done for other reasons can reduce how much of it is done. The best-known demonstration involves a fine for collecting children late from a nursery, after which lateness increased: the fine converted a social obligation into a priced service.

The effect is real, it is not universal, and it is most likely where the payment is small relative to the intrinsic motivation it displaces. Small bonuses for things people already care about are the case to be most careful with.

Very large bonuses

Experimental work has found that very high stakes can reduce performance on tasks requiring cognitive flexibility, apparently through pressure rather than through effort. The finding is contested and the laboratory setting limits what can be drawn from it.

It is worth knowing chiefly as a caution against the assumption that more incentive is monotonically more performance.

Gaming, which is the main practical risk

Every scheme with a threshold creates an incentive to be just above it, and every scheme with a period creates an incentive to move activity between periods.

Sales pulled forward into December and pushed out of January. Cases closed on paper. Targets met by opening accounts nobody asked for, which is how one large bank produced a scandal that cost it several billion and its chief executive.

None of that required dishonest people. It required a target, a consequence, and insufficient attention to what else the target touched.

What variable pay does to risk

Moves it to the worker. A pay package that is seventy per cent fixed and thirty per cent variable transfers a share of the employer's uncertainty onto the household budget, and the household is less able to absorb it.

In some industries this is regulated for exactly that reason, including rules on deferral and clawback in finance introduced after 2008.

Discretionary bonuses

Where the amount is at management discretion, the scheme is not an incentive in the technical sense, because the relationship between action and payment is not known in advance.

It functions instead as retrospective recognition and as a retention tool, both of which are reasonable purposes. Describing it as performance-related pay when nobody can state the function is where the resentment comes from.

The design questions

What exactly is being measured. What else the measure touches. Whether the worker controls it. What happens at the threshold. Over what period. And what would happen if somebody tried to maximise the measure without regard for anything else, because somebody will.

The alternative that gets skipped

Paying more, in fixed pay, without a scheme. It has no gaming risk, no measurement cost, no threshold effects and no crowding out, and it is frequently the option nobody models because it does not sound like management.

The period problem

An annual scheme rewards a year and forgets it. Anything whose payoff arrives in three years is unfunded by it, which is one reason maintenance, training and documentation are chronically underdone in organisations that pay annual bonuses on annual results.

Telling people what it is for

A scheme nobody can explain has the cost of a scheme and the effect of a lottery. If the people covered by it cannot state what changes the payout, it is not influencing behaviour and it is still being paid for.

That is a cheap diagnostic and organisations rarely run it, because asking the question invites the answer.

For broader context, consult US bonuses guidance.