The screen marks your order ready long before it is.
You are at a fast food joint. You order at the counter. A screen lights up with your number, and a few minutes later it shows as ready. Then it vanishes from the screen, implying that the order has been delivered.
Except you’re still standing there waiting. The food arrives ten, sometimes fifteen minutes after the screen said you got it.
I’ve seen this happen many times across fast-food outlets. Same brand names you’d recognise, same gap between what the screen says and the reality.
The number rolls up to someone who never stood at the counter.
That screen isn’t decoration. It originated to improve the user experience and now it’s a KPI. It measures how fast the outlet serves you, and that number rolls up to someone who runs many outlets and maybe has never stood at this one.
So the staff do the smart thing when they’re running behind. They mark the order ready and delivered before it is. The dashboard records a quick, on-time delivery. The head office sees a fast kitchen. But unfortunately, the customer sees a wait with no signal.
Somewhere far from that counter, decisions get made on the basis of that number. Staffing levels. Bonus payouts. Which outlet is a top performer and which needs a talking-to. All of it based on a handover that didn’t happen when the system said it did.
A KPI you enter about yourself is not a measurement.
There is a name for numbers that look good and mean little: vanity metrics, like follower counts and page views that present a rosy picture without telling you anything concrete. The usual advice is to swap them for actionable ones, measure outcomes not outputs, pick a better KPI.
And that advice isn’t wrong. It just isn’t the problem here.
What’s happening at the counter is a KPI being gamed. A gamed KPI is a real, useful metric distorted by the person who enters it. The order-ready time is a good metric. The problem is who enters it: the person reporting the number is the same person being graded by it, and no amount of metric-picking can fix that.
We built a tracker that measured nothing.
It’s easy to sneer at an overworked teenager at a fast food counter and miss yourself. So here’s our own version.
We once built a tracker for our field team. The idea was simple, maybe too simple. An executive would reach the visit location, open the app, start the visit, fill the details in real time, then stop it when they left. Start, stop, and we’d know how long each visit took and what happened in the visit.
That was the SOP: the standard operating procedure, how it was meant to be done.
And here’s what actually happened. People did their three or four visits for the day, then sat down in the evening and marked all of them in one go. Start, stop, start, stop, sitting at home, a cup of chai in hand, hours after the fact.
The data came in clean. Every visit had a list of tasks done and a duration, but none of it was true.
We have a shorthand for this gap now: SOP versus GOP. SOP is how the work is supposed to be done. GOP, or ground operating procedure, is how it’s done on the ground. The KPI is always written against the SOP. It gets gamed against the GOP. The distance between the two is where the gaming happens.
The phrase came out half-jokingly, in a review meeting where we were discussing how people behaved while freezing the requirements for a new feature. What we saw on the ground didn’t match what the process assumed. It didn’t match what we, sitting in the room, assumed either.
The staff aren’t cheating — the system asked the wrong question.
You could call this dishonesty. You could also argue it isn’t.
The counter staff aren’t villains. The field team wasn’t lazy. Both were handed a system that asked “are you fast?” — or “are you efficient?” — and then made the person being judged the one who answers. Given that, marking the order done early, or batching the visits at night, isn’t cheating. It’s the sane response to a badly built question.
Eliyahu Goldratt put it flatly: tell me how you measure me, and I will tell you how I will behave. Measure me in an illogical way, then don’t complain about illogical behaviour.
Donald Campbell named this mechanism decades ago — the more a number is used to make decisions, the more it corrupts the thing it was meant to measure.
And then distance finishes the job. Nobody upstream stands at the counter or rides along on the visits. They see the dashboard, and it looks fine. Everyone relaxes.
One self-reported number is never enough.
So how do you stop a KPI being gamed? Not by finding the one true metric — that’s the same trap in a nicer suit.
The failure is trusting a single, self-reported number watched from far away. The fix is plural, and deliberate.
Use more than one, and make them different in kind. Pair the number people enter with one they can’t touch. If a visit is self-started, check it against a system event: a location ping, a call log, a signature. A second number that can’t be gamed the same way is what turns it into a measurement.
Measure what the button is supposed to prove. The button press is easy to capture. What it stands for — a fed customer, a visit that happened — is the point. Before you trust a number, ask what it would look like if someone were gaming it, then go and measure that instead. Treat a spotless number as a prompt to look closer, not to celebrate.
Close the distance. No dashboard beats standing where the work happens. A manager who sees the queue catches what the service-time report never will. Where you can, separate the person who enters the number from the person it grades: a mystery audit, a second pair of eyes, anyone who isn’t scored by the same figure they’re reporting.
This isn’t only instinct. In Harvard Business Review, Michael Harris and Bill Tayler call it surrogation: when a metric stands in for the goal, people chase the number and forget the goal. Wells Fargo set a cross-selling target; staff opened around 3.5 million accounts no one asked for. Studying it, the authors conclude what we did: no single yardstick should carry the whole judgement.
None of this is expensive. It’s a willingness to distrust your own dashboard on the first reading. The same restraint that keeps a product simple keeps a metric honest.
The delivery companies already fixed this.
You’ve been the second check yourself, without noticing.
A courier marks your parcel “attempted delivery, customer unavailable.” Once, that was the end of it: the agent entered the number, and the number was believed.
Now your phone vibrates a moment later: did the agent actually attempt delivery, or reach you? You confirm it, or you say it never happened. The platform has stopped treating the agent’s word as the only record and started checking it against the one person the agent can’t overrule: you. Too many mismatches against the same person and lo and behold, you caught a bad apple.
That’s the whole idea, running at a national scale.
And here’s what we did on our own tracker: geolocation tagging, selfie capture at the location, OTP inputs and other data points on the visit. Together they made it difficult to game the system.
The self-report still exists. It just doesn’t get to be the truth on its own.
Find the number nobody checks.
A green dashboard buys you a feeling of control. It tells you the kitchen is fast, the visits are logged, the work is getting done. All is well.
Sometimes that’s true. And sometimes it isn’t, because you asked people to report on their own work from far away — and far away is exactly where you can’t catch it.
The counter was never the point. Your product is. Look at your own dashboards and workflows, find the number someone enters about their own work and nobody checks, and make sure you’re not making the same mistake.
And if you do find one being gamed, the question to ask is this: why does the employee feel the need to game it? That’s the harder question to answer, but the right one to ask.
Minimum Viable Thinking is for product people searching for clarity in a culture of clutter. If this one landed, subscribe, and forward it to anyone who owns a dashboard.
I write about building products without the clutter, and about using AI in the actual work, not just having opinions about it. Most of it comes from more than a decade in product, much of it in health-tech in India.
Subscribe on Substack