In 1975, an economist at the Bank of England named Charles Goodhart was working on monetary policy. He was watching the UK government try to control inflation by targeting specific measures of money supply. Goodhart noticed something strange: every time the government picked a particular measure as their target, that measure stopped being a useful indicator of the underlying economy. Banks and consumers would rearrange their behavior to optimize against the target, and the target would lose its predictive value.
He wrote down a sentence that has since become one of the most quoted laws in social science:
"When a measure becomes a target, it ceases to be a good measure."
This is Goodhart's Law, and it is, I'd argue, more useful than any individual cognitive bias in behavioral economics. It's the meta-principle behind why incentive systems almost always backfire. The Cobra Effect is just the most colorful illustration of Goodhart's Law in the wild.
The Cobra, the Rats, and the Apocrypha
The popular Cobra Effect story goes like this: in colonial India, the British government, worried about Delhi's cobra population, offered a bounty for dead snakes. Enterprising locals started farming cobras for the bounty. When the British figured this out and canceled the program, the farmers released their now-worthless snakes into the city. Cobra population: worse than before.
It's a great story. It is also, almost certainly, apocryphal. Historians have looked for primary documentation of this program and not found it. The story appears in a 2001 book by German economist Horst Siebert (Der Kobra-Effekt), but Siebert himself doesn't cite a primary source. The historical record is, at best, hazy.
There is, however, a much better-documented twin story from French colonial Hanoi in 1902. The French administration, trying to combat a plague outbreak, offered a bounty for rat tails. Workers were paid per tail. Hanoi residents began catching rats, cutting off their tails, and releasing the now-tailless rats to continue breeding — guaranteeing a perpetual rat-tail supply. French officials, on inspecting the city, started finding tailless rats running around the streets. The historian Michael Vann has documented this story extensively in his book The Great Hanoi Rat Hunt.
The lesson is the same in both cases. Set up a reward, and people will optimize against it instead of for the goal you actually wanted.
This is what economists call a perverse incentive. It is what social scientists call Campbell's Law (Donald T. Campbell, 1976: "The more any quantitative social indicator is used for social decision-making, the more subject it will be to corruption pressures"). It is what consultants now call unintended consequences. The vocabulary is overcomplicated. The phenomenon is one thing.
Three Modern Cobras
I want to walk through three recent cases because the historical examples can feel quaint. The mechanism is alive and well in modern business.
Bogotá's Driving Restriction
In 1998, Bogotá, Colombia introduced a policy called Pico y Placa — "rush hour and license plate" — designed to reduce traffic and pollution. The rule was simple: depending on the last digit of your license plate, you couldn't drive on certain days.
The policy worked for about eighteen months. Then it didn't.
What economists tracking the program found was that wealthier families simply bought a second car with a different license plate to fill the gap. Some bought a third and fourth. The total number of cars in Bogotá rose significantly. The cars bought specifically to circumvent Pico y Placa were often older, cheaper, and more polluting than the cars they supplemented. The pollution and traffic both got worse, not better. The Inter-American Development Bank published a detailed analysis of this in 2009.
The behavioral move here is deceptively obvious in hindsight. The government targeted which cars were on the road on a given day instead of how many cars existed. Citizens optimized against the measure, not the goal. Goodhart's Law in two-vehicle-garage form.
The NFL's Tanking Problem
The NFL's draft system is supposed to help bad teams improve. Worst record gets first pick. First pick gets the best college players. The bad team gets better. Equilibrium is restored. Parity is preserved.
The system has a Cobra Effect built into it. If having the worst record is rewarded with the first overall pick, then having the worst record becomes a strategic asset for teams that are out of playoff contention. The optimal play, late in a losing season, is to lose more — known in the league as "tanking."
In 2022, former Miami Dolphins coach Brian Flores filed a lawsuit alleging that Dolphins owner Stephen Ross had offered him $100,000 per loss in the 2019 season to deliberately throw games and improve the team's draft position. The NFL investigated and ultimately fined Ross $1.5 million and suspended him through October 2022. Ross denied the offer; the league's investigation could not definitively prove specific games were thrown but found enough to merit discipline.
Whether or not the specific allegations against Ross are fully accurate, the structural incentive is real. The NBA has had the same problem for decades and has tweaked its draft lottery system multiple times to try to defang it. Goodhart's Law again: the moment "worst record" becomes a target rather than a measure of need, it stops being a useful measure of need.
Wells Fargo and the Cobra in a Bonus Plan
The third modern cobra is, in some ways, the most expensive in history.
Starting in the early 2000s, Wells Fargo introduced an aggressive cross-selling target for branch employees: a minimum number of new accounts opened per customer per quarter. Bonuses depended on it. Front-line employees who missed targets were threatened with termination. The metric was number of new accounts. The presumed underlying goal was deeper, more valuable customer relationships.
You can guess what happened.
By 2016, internal investigations revealed that Wells Fargo employees had opened roughly 3.5 million unauthorized accounts in customers' names. Employees forged signatures, created fake email addresses, transferred small amounts of money between fake accounts to keep them "active," and bound real customers to credit lines they had never agreed to. The bank fired over 5,000 employees. It paid $185 million in the initial 2016 settlement. By 2022, the total cost — counting the CFPB's $3.7 billion fine, settlements with the SEC, the DOJ, and various state AGs — was approaching $10 billion in penalties, not including the reputational damage that drove customers to competitors for years afterward.
A new-accounts target became the measure. Once it became a measure, it ceased to be a measure of the goal. The goal was customer depth. The measure was account count. Employees gamed the measure because their livelihoods depended on it. The bank discovered, the hard way, that it had built a multi-billion-dollar cobra farm.
Why This Keeps Happening
The reason every reward system eventually gets gamed has a deep behavioral explanation. Steven Levitt and Stephen Dubner spend a substantial chunk of Freakonomics on this exact mechanism — what they call "the hidden side of everything" is mostly the hidden side of incentive design.
Three reinforcing factors:
- People are smarter than the systems designed to constrain them. Bureaucracies design simple targets. Humans find clever workarounds. The asymmetry compounds.
- The target is always less rich than the actual goal. "Reduce traffic" is rich. "Reduce license plates on the road on Mondays" is impoverished. Citizens optimize against the impoverished version because that's what the rule actually says.
- Most people are not unethical — they're locally rational. The Wells Fargo branch employee opening fake accounts wasn't a villain. They were a parent trying to keep a job. The system put them in an impossible position and they did what humans do under economic pressure.
This is the territory Jerry Muller covers in The Tyranny of Metrics and James Scott covers in Seeing Like a State. Both books are about the systematic ways simple measures fail to capture complex realities, and both are required reading for anyone designing an incentive system.
How to Build Incentives That Don't Backfire
The honest answer is: you can't, fully. But you can reduce the failure surface dramatically.
Make incentives simple. Complex incentives have more surfaces to exploit. The simpler the rule, the harder it is to find a clever workaround.
Adversarially red-team your own incentive plan. Before you launch any incentive, spend serious time trying to game it yourself. Imagine you're an unscrupulous version of the people you're trying to reward. What's the easiest path to the bonus that doesn't involve doing the work you actually wanted done? If you can find a path in twenty minutes, your employees will find it in twenty seconds.
Tie incentives to outcomes, not activities. The Wells Fargo failure was rewarding new accounts, an activity. The right reward would have been long-term customer lifetime value, an outcome. The activity is easier to fake. The outcome is much harder.
Build in lag. If the reward is paid out the instant the activity happens, gaming is encouraged. If the reward is paid out three years later, after the work has had time to either prove valuable or expose itself as fake, gaming gets a lot harder.
Pair quantitative metrics with qualitative oversight. Goodhart's Law fires hardest when a number is the sole criterion. Adding human judgment — sample auditing, peer review, supervisor verification — reintroduces the kind of texture that simple metrics can't capture.
What I Take From All This
Goodhart's Law isn't a curiosity. It's a structural law of how incentives behave. Every reward system you'll ever design — for employees, customers, vendors, AI agents — is subject to it. The only question is how much damage your particular Cobra Effect will cause before you notice.
The most useful operational instinct I've taken from years of staring at this is to assume my incentive plans will fail and design them with the failure mode in mind. If I cannot easily explain how my reward could be gamed, I haven't thought about it hard enough yet.
The Cobra Effect isn't a bug in human nature. It's a feature. People will always optimize against the system you build, not the goal you imagined. The job of incentive design isn't to eliminate that response. It's to make the optimization actually serve the goal.
Otherwise: cobras everywhere. And nobody to buy the skins.