Before you announce a bonus, take a sheet of paper and test what you will count to pay it. At the top, write the job you want done well. Under it, write the measure you meant to pay on, the one that is quick to count, such as boxes picked. Then write what that measure leaves out, whether the worker can raise it by working better, and the easiest shortcut to raise it without doing the job well. That page is your test sheet. Once it is filled in, you keep the measure, change it, or decide to pay no bonus.
Say you paid your pickers by the box, and the picking got faster and worse. Fruit was knocked to the ground and plants were roughed up. Fruit that should have stayed on the plant to ripen for the next picking was bruised or pulled off green. The bonus did what it was built to do: you paid for boxes, and boxes are what came in. Picking with care and leaving the plant in good shape were never part of the count, so that is the work that got dropped.
When it looks like the crew is gaming the bonus, the cause is usually a measure that was easy to count. And the easiest measure to count is often the easiest to push up without doing the job well.
Why does a bonus on the easiest measure make the work worse?
Because pay steers where people put their effort. When one task is easy to count and another is not, paying for the first pulls attention away from the second. The second is usually the one that mattered: fruit picked without bruising, or the picking ladders and bins kept in good repair.
Two economists, Bengt Holmström and Paul Milgrom, set out this argument in a 1991 paper. They worked it out in mathematics rather than testing it on farms, but the mechanism is the one you see at harvest.
Pay, they write, does two jobs: it rewards hard work, and it steers which of their duties workers give their attention to. When output is easy to count but its quality is not, piece rates can lead workers to raise the volume at the expense of quality. Where the care of shared equipment is hard to check, piece rates can lead workers to abuse that equipment or neglect it.
At harvest, this costs money. Fritz Roka, in a University of Florida extension bulletin for growers, notes that “productivity incentives tend to drive workers to harvest at faster speeds, which in turn could damage or prematurely destroy a plant”. Fruit picked below the buyer’s grade, he adds, “must be culled or discarded, resulting in a direct financial loss to the grower”. You set the bonus to raise the harvest and ended up paying people to spoil part of it.
Is the worker gaming the bonus, or doing what it pays for?
The worker is doing what the bonus pays for, and that is the part that stings. Someone paid by the box will push the box count up. If the easiest way to push it up damages what you actually wanted, you paid for the damage. The fix is in how the bonus is set up, not in more discipline.
This pattern has a name, Goodhart’s law. David Manheim and a co-author sorted its different forms in a 2018 paper. As they describe it, once people are pushed hard to raise a measure that stands in for the real goal, the measure stops moving with the goal. Boxes go up while sound fruit goes down.
Sarah Fogleman, in a guide to incentive programs published by Kansas State University, gives the example of a hog operation that raises pigs from birth to market. To cut piglet deaths in the farrowing barn, where the sows give birth, the owner paid a bonus on live pigs leaving that barn. The result, in her words: “nursery mortality starts to rise”. Sick and weak pigs that would normally have stayed in the farrowing barn were sent on to the nursery, the next barn, so they would count as live, and died there.
The staff were not cheating. They “were following the carrots”, doing what the bonus paid for. So when a bonus gets gamed, take it as a sign that the measure is wrong. The test sheet is where you can catch that before you pay out.
Which measure belongs in the bonus?
The right measure passes two tests: the worker can raise it by working better, and it only goes up when the work you care about is done well. The first test is an old rule: pay people on what they control. Fogleman puts it in one line for farms: a manager should “search for things the employees influence” and build the bonus on those.
The second test matters most when the job has an early step and a finished product. On a fruit farm, the box filled in the orchard is the early step, and the fruit that passes grading at the packing shed is the finished product. If careless picking shows up later as fruit that fails the grade, then paying on the graded fruit already pays for careful picking, and a separate bonus on the box count adds little.
This only works if you can tell whose fruit is whose: the foreman checks each picker’s bin, or each bin carries the picker’s tag to the packing shed. Where you can, pay on the fruit that was graded and accepted, not on the raw count of boxes. Where the shed grades whole loads and no bin can be traced to a picker, a bonus on the grade cannot reward each person, and a fixed wage, covered below, may be the better choice.
Which mistakes make a bonus plan fail?
A bonus plan goes wrong in a few known ways, and you can check for each one on paper before the harvest starts. Hold the measure you meant to pay on against each row of this table; if it fails a row, change it before you announce the bonus.
| The mistake | What people end up doing | What to do instead |
|---|---|---|
| The measure depends on things the worker cannot change | Extra effort does not change the measure, because weather or price decides it | Pay on the part the worker can influence, and leave the rest out of the bonus |
| An important task is left out of the count | Effort drains from the uncounted task toward the counted one | Name the left-out task first. Then count it too, or stop paying extra per box, since the extra per box is what pulls effort away from the left-out task |
| The measure is easy to push up | People push it up the cheap way and damage the work it was supposed to reflect | Pay on the result you want, such as fruit that passes the grade, not on a count that is quick to run up |
| You pay on an early step when the finished product can be graded | You count boxes in the orchard when the packing shed already grades the fruit; the count adds little, and keeping a record of it takes someone’s time | If the picking shapes the grade and you can trace the grade to the picker, pay on the graded fruit and stop counting boxes |
| No one checks the quality the measure ignores | Speed goes up, the grade goes down, and the loss turns up when the buyer grades the load | Tie every volume bonus to a quality check, and pay only on what passes it |
| You pay a bonus where a steady wage would do the job | A strong bonus on work you cannot grade rewards the shortcuts | Where quality is hard to measure, pay a fixed wage and drop the bonus |
The left-out task is the one people most often forget to write down, so on the test sheet it gets its own line, right after the measure you meant to pay on. Name it in plain words: the plant left in shape to bear again, the ladders kept in repair, the damaged fruit left out of the bin. If you cannot name what your measure leaves out, paying on it is the likeliest way to end up with fast, careless work.
When is the answer no bonus at all?
Sometimes the right answer is a wage and no bonus, and that is a sound choice. Holmström and Milgrom show that when the quality of the work is hard to measure, “an optimal incentive contract can be to pay a fixed wage independent of measured performance”. In plain words, a fixed wage can be the best way to pay, because a strong bonus on work you cannot grade rewards the shortcuts.
Between a strong bonus and no bonus there are two middle options, and it is worth trying them on the test sheet first. The first is a quality gate: keep the volume bonus, but pay it only on output that passes grading, so speed without quality earns nothing. This works where someone can grade the output and tell whose it is, such as the foreman at the bin or the packing shed with tagged bins. Roka is blunt that “supervision of piece rate workers to ensure performance quality is necessary to avoid adverse consequences”. With the gate, that supervision becomes part of how you pay.
The second option is to split the job, so the task that is easy to count goes to some workers and the care goes to others. Holmström and Milgrom offer this as an idea, not a proven result: “the tasks in which performance is most easily measured are assigned to one worker” and the rest go to another.
The split works for a care task that can really be taken off the pickers, such as keeping the ladders and bins in repair. The pickers keep the bonus on the boxes, and whoever does the repairs is paid a fixed wage. It does not fix bruising, because care in picking is part of the picking itself: a worker paid by the piece, they note, may always be able to trade attention to detail for speed.
Choosing among the gate, the split and a fixed wage is the hard part, and skipping the choice usually means paying on the raw count by default. Once a measure gets through the test sheet, write it into the terms of the bonus, with the goal, how the money is divided among the crew and the payout date. Hand those terms to the crew before the harvest starts. If your bonus terms are not on paper yet, a companion article shows how to write the bonus terms before the results come in.
A bonus measure is one of the numbers you keep on your crew, with money attached, so the same question applies to the others, such as turnover or absences: does the number fairly reflect the work? Another article covers which numbers about your crew to track. Writing the rule before the work starts and checking the result against it afterwards is a habit that serves the rest of farm management as well. More on reviewing, rewarding and paying a crew is in the people section of the library.
Where to start
One sheet of paper, about an hour, and the bonus you were about to announce.
When a bonus gets gamed, look at the measure before you look at the crew: the measure is what you told people to chase, and that is where the plan went wrong. The fix is a line rewritten on the test sheet, not a lecture about effort. Pay on what the worker can influence and on what the farm can sell, and the work you forgot to count is less likely to be the work that gets dropped.