The Cobra Effect: Why Your Best Metric Might Be Breeding Snakes

From Product Theory: The Hidden Forces That Shape User Behavior — 40+ short chapters on why users behave the way they do.

The British government in Delhi had a cobra problem. Too many venomous snakes, too many people getting bitten. So they did the reasonable thing: they offered a bounty for every dead cobra. Cash for corpses. Simple incentive, clear outcome.

At first, it worked. Dead cobras came in. The population seemed to drop.

Then someone got clever.

The bounty that backfired

Why hunt cobras in the wild when you could breed them at home? Raise cobras, kill cobras, collect bounty. Repeat. A cottage industry sprang up around the exact animal the government was trying to eliminate.

When officials figured this out, they scrapped the program. No more bounties. Which left the breeders holding a lot of worthless snakes—so they released them.

The cobra population ended up higher than before the program started.

That's the Cobra Effect: a well-meaning incentive that produces the exact opposite of what you wanted. Economist Horst Siebert coined the term in 2001, but the pattern is old and it's specific. It's not just that the metric became useless. It's that the metric made things actively worse.

Every incentive creates a game. Make sure it's a game you want people to play.

The same trap, wearing a lanyard

You don't need a colonial bounty program to breed cobras. You just need a dashboard.

Say you want more referrals, so you pay a bonus for every new signup. It works—signups spike. Then you look closer. A handful of power users have figured out they can spin up fake accounts and collect the bounty. You're not paying for growth. You're paying people to breed cobras.

The conversation always goes the same way:

PM: "Our referral program is crushing it. 500 new signups a week."

Analyst: "How many convert to paid?"

PM: "Uh. Not many. Most don't even log in after day one."

Analyst: "Are they real people?"

PM: "Define real."

That's the moment the snakes hit the floor.

The pattern shows up anywhere you reward a proxy instead of the outcome:

• Track tickets closed per hour → support reps close tickets without solving anything. "Have you tried restarting?" Close ticket. The metric looks great; satisfaction drops.
• Make the free tier generous to drive conversion → you grow a massive population of free riders who will never pay and cost you money every month.
• Pay engineers to find bugs → if the bounty is good enough, what stops someone from leaving bugs in place to "discover" later? You've made sabotage the optimal strategy.

In every case the incentive created a game. Just not the game you wanted.

Why incentives lie to you

Here's the mechanism. Incentives don't change what people want. They change how people behave to get what they want.

Pay for dead cobras and people optimize for dead cobras—not fewer wild cobras. Pay for closed tickets and people optimize for closed tickets—not happy customers. Pay for signups and you get signups—not engaged users.

Every incentive opens a gap between two things:

• The proxy — what you measure.
• The goal — what you actually want.

When the proxy is easier to game than the goal is to achieve, people will game the proxy. Not because they're villains—because you built a game and told them the score.

The sabotage question

Before you launch any incentive, ask one thing: "If someone wanted to hit this target without achieving the goal, how would they do it?"

If the answer comes easily, your incentive is already broken. Fix it before it ships.

This is also why some metrics are far safer than others—it comes down to how tightly the proxy is coupled to the outcome.

Revenue is harder to fake than signups. Retention requires real value in a way acquisition never does. The closer your metric sits to the real outcome, the less room there is for breeding snakes.

And that's the whole point of the sabotage question. Every vanity metric—downloads, signups, page views—is a potential cobra, because they're all easy to juice without creating anything real. Once you reward people for moving the number, the game becomes make the number go up. Not make the product better.

Rule of thumb: If hitting the target is easier than solving the problem, you're measuring the wrong thing.

If you're already eyeing one number on your own dashboard, quietly working out how you'd hit it without solving anything, that's the sabotage question doing its job. I send one idea like this at a time over at Context Limit — no cobras.