The numbers looked… too good. CTR up. CPA down. Acronyms. Acronyms all round. Graphs heading up on the right.
Looker Studio glowed green: -48.7% cost per conversion. Kerching!
I stared at it, a mug of Tesco Gold Instant cooling beside my keyboard, the kind that humbles you with overtones of radiator water. The flat hummed with the sound of the dehumidifier. Everything about the room felt normal, except the data.
A good result should make you feel proud. This one made me suspicious. I took a sip of cold instant and thought: it’s learned how to cheat, hasn’t it?
The snake in the system
A client wanted more leads. Fast. We set a cost-per-lead target, flicked on automated bidding, and watched the algorithm do its thing.
Within a fortnight: victory. Leads up 300%. CPAs sliced in half. Except the sales team wasn’t seeing a single new customer.
After a bit of forensic Looker Studio archaeology, we found it. The optimiser had decided that existing customers filling in forms again counted as “new leads.”
Technically correct. Commercially useless.
It hadn’t broken the system. It had simply taken us at our word.
A short history of snake farming
The somewhat spurious origin story of the Cobra Effect goes something like this: faced with too many cobras in colonial India, officials introduced a bounty for every dead snake.
At first, it worked. Then people started breeding cobras, killing them, and collecting the reward. When the bounty ended, breeders released their snakes. The cobra population exploded.
The problem wasn’t bad intent. It was a bad incentive.
What it means for marketers
The Cobra Effect happens when a metric or reward backfires, because people or algorithms optimise for the number instead of the meaning. Rewarding the metric instead of the mission.
Where you’ll see it
- Paid search optimised for clicks, not qualified leads
- Engagement campaigns rewarded for reactions, not relevance
- Sales teams bonused on sign-ups, not retained clients
- Smart bidding chasing cheap traffic, not profitable conversions
- Loyalty schemes gamed by serial voucher-hunters
- Content reports celebrating impressions, not impact
From the dashboard view, everything looks like success. From the business view, it’s a snake pit.
How to catch yourself
Before you celebrate a metric, ask:
- Are we rewarding the result, or the appearance of it?
- If someone wanted to game this, how would they?
- Would this metric still feel believable if it doubled overnight?
- Who benefits if the system cheats?
If the honest answers make you uncomfortable, there may already be a snake around your neck.
Guardrails
- Define success in words before numbers. “More good leads” is meaningless until you say what a good lead is.
- Run a pre-mortem. Ask, “If this goes wrong, how will it go wrong?” before you launch.
- Keep a human in the loop. Dashboards count. Humans notice when the counting stops making sense.
The napkin test
You pay affiliates £10 per “lead.” Someone builds a bot that fills in your contact form 200 times a day. Revenue doesn’t move. The dashboard glows green.
That’s not performance. That’s snake farming. And we’ve got plenty of snake farmers, charmers, and dashboard evangelists in marketing.
Tactful pushbacks
Slack / email: “Before we optimise for X, can we check that X actually measures success? Otherwise we might just be teaching the algorithm to look busy.”
Stakeholder version: “It’s like paying salespeople for the number of meetings booked, not whether the deals close. You end up with a calendar full of non-starters.”
Close but not quite
- McNamara Fallacy: measuring only what’s easy.
- Overfitting: mistaking the past for the future.
When to ignore this
If no reward, bonus, or pressure is tied to the metric, you’re probably fine. Cobras only appear when there’s a prize. (Although you might be surprised what some people consider a prize, especially when it features in a performance report.)
Subscribe on LinkedIn