A reward aimed at solving a problem can backfire by rewarding the wrong behavior. Pay people to kill cobras and someone starts breeding cobras. Whenever you incentivize a proxy, expect people to game the proxy — sometimes worse than the original problem.
The cobra effect: a well-intentioned incentive backfires because people optimize the literal reward rather than the outcome you wanted — sometimes making the problem worse than before you intervened.
The story (told by economist Horst Siebert): colonial Delhi had too many venomous cobras, so the government offered a bounty for dead cobras. It worked briefly — then enterprising locals began breeding cobras to cash in. When officials discovered the scheme and scrapped the bounty, the breeders released their now-worthless snakes, leaving more cobras than at the start. The fix didn't just fail; it reversed. A real version happened in Hanoi with rats, where a bounty paid for severed rat tails produced tailless rats roaming the city — breeding more rats to harvest.
It's a vivid case of perverse incentives, close kin to Goodhart's law: the moment you reward a measurable proxy ('dead cobras') instead of the true goal ('fewer cobras'), people optimize the proxy by whatever means, including means that defeat your purpose. Pay developers per bug fixed and you may get bugs introduced to be fixed. Reward call centers for closed tickets and problems get closed, not solved. Fund agencies by the size of the problem they manage and the problem mysteriously never shrinks. People aren't being evil — they're responding rationally to exactly what you rewarded.
The defense is to design incentives by asking 'how would a clever, self-interested person game this?' before launching it — run the cobra test. Reward outcomes you actually want, not easy proxies; watch for the behavior the incentive will breed, not just the behavior you hope for; and stay ready to kill an incentive that's being gamed (carefully — even removing it can backfire, as the released cobras showed). Whenever you attach a reward to a number, assume the number, not the goal, is what you'll get.
Every incentive you set — at work, in policy, in your own life — risks rewarding the wrong thing; the cobra effect is the warning to design for how people will actually game the reward, not for how you hope they'll behave.
It's not 'all incentives backfire, so never use them.' Well-designed incentives work powerfully. The lesson is to design for the behavior the reward will actually breed — anticipate gaming, reward true outcomes over proxies — not to abandon incentives, which would just leave the original problem unaddressed.
Lock this idea into memory with a 5-minute active-recall session — the science of spaced repetition, no signup.
Try this idea free →