The story, from colonial Delhi, is that the British administration wanted fewer cobras and offered a bounty per dead snake. It worked briefly. Then people began breeding cobras to collect on them. When the scheme was scrapped, the breeders released their stock, and there were more cobras than at the start.

Whether the anecdote is history or folklore, the structure is entirely real: an incentive rewards a proxy, and the cheapest way to move the proxy is not the way that was intended.

Why it is not a people problem

Nobody in the story did anything irrational. The administration was buying dead cobras. The suppliers produced dead cobras. The gap was between what was bought and what was wanted, and that gap is a design error.

This is the same mechanism as Goodhart's law, stated from the incentive side rather than the measurement side. Goodhart says a measure degrades when it becomes a target; a perverse incentive is what that degradation looks like when money is attached and the degradation runs backward.

Common ones

Paying for activity. Calls made, tickets closed, lines written, meetings held. Each is easy to generate without the underlying value, and each will be.

Rewarding the fix rather than the prevention. The person who resolves a crisis is visible and the person who prevented one is not, so crises become career-useful and prevention does not get staffed.

Penalizing reported problems. If raising a defect is costly to the person raising it, defects stop being raised and start being discovered by customers.

Commission without retention. Paying at signature buys signatures, including from customers who will leave in three months.

Utilization targets. Billing every available hour rewards taking on work that should have been declined and discourages the unbilled thinking that makes the billed work good.

Designing against it

Ask what the cheapest way to earn this is. Not the intended way — the cheapest. If a cheaper route exists it will be found, usually within a quarter, usually by someone who is not trying to game anything.

Pay for outcomes that are hard to fake and that arrive late. Retained revenue rather than closed deals. Resolved-and-stayed-resolved rather than closed.

Pair the reward with its damage. Speed against defects, volume against retention. A pair is much harder to move in one direction only.

Watch the number that gets better first. When one metric improves sharply and nothing else changes, the incentive has been found rather than the outcome.