Cobra Effect
A fix that creates a faster, worse way to produce the very problem it was meant to solve.
What Is It?
The Cobra Effect describes an intervention that backfires by handing people a more efficient path to the exact outcome it was designed to prevent. The name comes from an often-repeated story: colonial administrators in Delhi, alarmed by a cobra infestation, allegedly offered a bounty for every dead cobra turned in. People began breeding cobras specifically to kill them and collect the reward. When administrators caught on and cancelled the bounty, the newly worthless cobras were released, leaving the city with more cobras than when the program started.
That story is worth treating with real caution rather than repeating as settled history. The term itself was coined by German economist Horst Siebert in a 2001 book, and the underlying colonial anecdote has little solid historical documentation behind it, it functions more as an illustrative parable than a verified case. What’s genuinely well documented, across many independent, better-sourced examples, is the mechanism the parable illustrates: a reward tied to a proxy for the problem, rather than the problem itself, creates an incentive to produce more of whatever earns the reward, even when producing it makes the underlying problem worse. Goodhart’s Law asks how you can improve the number without improving reality. The Cobra Effect asks something sharper: how you can produce more of what earns the reward, even if doing so means producing more of the very problem the reward was meant to solve.
Why Does It Matter?
Organizations design incentives around observable proxies constantly, because the actual outcome they care about is often hard to measure directly. Paying a support team per ticket closed rewards fast closes, not resolved problems; a security bounty program paying per bug found can reward superficial, easily rediscoverable bugs over the deep audit work that actually improves safety; a hiring target measured in headcount added can reward filling seats over building the team that was actually needed. In each case, the reward attaches to a stand-in for the goal, and the stand-in is exactly what people optimize once real money or reputation rides on it.
The Cobra Effect is the sharpest version of this problem, the case where the proxy isn’t just imperfect, it’s actively gameable in a way that manufactures the underlying issue rather than merely failing to reduce it. A pest bounty that pays per animal doesn’t just fail to reduce the pest population, it creates a direct financial incentive to increase it. That’s a stronger and more dangerous failure mode than a metric simply drifting from reality, because the incentive is actively pulling in the wrong direction from the start.
The distortion doesn’t stop at manufacturing the proxy. Left in place long enough, an organization tends to reorganize around the manufactured proxy rather than just tolerating it. If support tickets closed becomes a revenue driver, the organization doesn’t just end up with more tickets, it gradually builds workflows, staffing, and reporting that depend on a steady supply of them. At that point the proxy isn’t merely being gamed. It’s been institutionalized.
What Changes Once You See It?
You stop asking only “does this reward move the number we care about” and start asking “does this reward create a faster way to produce more of the underlying problem than to actually solve it,” since those are different questions with very different failure profiles.
You start looking specifically for interventions where the reward is paid on a countable proxy, dead animals, tickets closed, bugs filed, referrals made, and checking whether the proxy can be manufactured more cheaply than the real outcome can be achieved.
You also get more careful about how an incentive program ends. The Delhi parable’s second half, the release of the now-worthless cobras once the bounty was cancelled, is a reminder that badly designed incentives can do damage on the way out as well as the way in, and an exit needs as much thought as the launch.
Common Misunderstandings
- It is not the same claim as Goodhart’s Law or Campbell’s Law, though all three are closely related. Those describe a measure losing its meaning or the underlying process distorting once a metric becomes a target. The Cobra Effect is the specific, sharper case where the incentive actively creates more of the problem it was meant to eliminate, not merely a distorted or gamed measurement of it.
- It doesn’t require dishonesty or rule-breaking to occur. In the classic telling, breeding cobras for the bounty wasn’t cheating, it was the individually rational response to the actual incentive on offer. The design created the outcome; no one needed to break any rules to produce it.
- It is not an argument against incentives or bounties generally. Well-designed bounty and reward programs work every day; the caution is narrower, that a reward paid on a proxy needs to be checked for whether the proxy can be manufactured, not just measured, before it’s trusted.
- It isn’t limited to one dramatic reversal at the end. The pattern shows up in smaller, chronic versions too: a steady trickle of manufactured proxies that never fully reverses the original goal but does quietly waste resources rewarding the wrong behavior indefinitely.
- It doesn’t require the original problem to grow in an absolute sense to count. Milder versions simply redirect effort toward manufacturing the rewarded proxy rather than solving the underlying problem, without ever fully reversing the intended goal.
Diagnostic Question
Could someone produce more of this reward by manufacturing the proxy we’re paying for, rather than by solving the problem the proxy was meant to represent?
Explore Further
Field Notes
None yet.
Related Field Guide
Origin
The term was coined by German economist Horst Siebert in his 2001 book Der Kobra-Effekt, drawing on an oft-repeated but poorly documented anecdote about a cobra bounty program in colonial-era Delhi. The specific colonial story has little verified historical support, but the term has become the standard shorthand for the broader, well-evidenced pattern of incentives that manufacture the problem they were meant to solve.