People Respond to Incentives, Not Instructions

The most reliable force in economics isn't what people are told to do. It's what they're rewarded for doing — and the two are misaligned far more often than anyone expects when they design the reward.

The bounty that bred more cobras

In colonial-era Delhi, the government grew concerned about the number of venomous cobras in the city and offered a bounty for every dead cobra turned in. At first, it worked — people hunted cobras, the population fell, everyone declared it a success. Then enterprising residents realized breeding cobras was easier and more reliable than hunting wild ones, and started farming them purely to collect the bounty. When officials caught on and cancelled the program, the breeders — now holding cobras worth nothing — released them. The city ended up with more cobras than before the bounty ever existed.

The instruction and the incentive were two different things

"Reduce the cobra population" was the instruction. "Get paid per dead cobra" was the actual incentive. Nobody involved was acting irrationally or maliciously — they were responding, with complete rationality, to exactly what was being rewarded. The instruction and the incentive looked aligned right up until someone found a cheaper way to satisfy the incentive without producing the outcome the instruction was written to achieve. Once that gap opened, the incentive won, because incentives are what people actually optimize against — instructions are just what the incentive was supposed to represent.

"Gaming the system" is usually the system working as designed

When an incentive produces a result nobody wanted, the common response is to blame the people for "gaming the system." That framing lets the actual design off the hook. The cobra breeders didn't break the bounty program — they satisfied its exact terms more efficiently than the officials who wrote it anticipated. An incentive that pays out for the wrong thing isn't a people problem waiting to be lectured away. It's a specification problem, and it needs to be fixed the same way any other broken specification gets fixed: by changing what's actually measured, not by asking people to want the "right" outcome more sincerely.

Where the same pattern shows up everywhere else

The cobra effect isn't a quirk of 19th-century colonial administration — it's a template. Measure a call center on "calls resolved per hour" and reps learn to close tickets fast without actually solving anything. Measure them on "average handle time" instead and reps learn to transfer difficult callers to someone else's queue before the clock runs against them. Neither group is lazy or dishonest. Both groups are doing exactly what the metric rewards, which is the only thing a metric can ever actually ask for — it can't ask for the intent behind it.

The one-question check

Before trusting that an incentive will produce the outcome you actually want, ask: if someone wanted to satisfy this incentive in the cheapest, laziest way possible while ignoring the instruction entirely, what would they do — and does that cheap path pay out the same reward as doing it properly? If it does, that cheap path is the outcome to expect. Not the instruction you wrote down, but the incentive you actually built.