Every owner has a dashboard, and every dashboard is a flattering lie of omission. Not because anyone's cooking the books — because a dashboard can only show you the questions you already thought to ask. The numbers you chose to track are visible. Everything you didn't think to track is invisible, and that invisible space is exactly where money quietly leaves the building. The owner's best use of AI isn't generating more reports. It's pointing a tireless, un-fooled analyst at your own operation and asking: what's here that I'm not seeing?
I learned this the expensive way. Let me tell you about a bonus that taught me to never trust a single number again.
The number that lied
We paid a bonus on intake acceptance — accept a qualified lead fast, less leakage at the top of the funnel, more cases worked. Acceptance did exactly what we paid it to do. It shot up, into the ninety-percent range. On the dashboard, the top of the funnel had never looked healthier.
Then someone set two numbers side by side. Intake acceptance: 90%. Retainer conversion: 30%. Nine in ten leads "accepted," three in ten actually signing. Those two numbers cannot both be honest at once. A 90% acceptance rate says the leads are good; a 30% conversion rate says most of them weren't. The gap between them wasn't noise — it was a confession. Reps were having friends call the intake line, present as leads, and get "accepted," because every acceptance paid a bonus. The metric had stopped measuring funnel health and started measuring how creatively the team could manufacture acceptances.
Two numbers that should move together, drifting apart, is not a data problem. It's a tell — and AI never gets bored of looking for it.
Why this happens to every owner
It has a name. Goodhart's Law, sharpened by Marilyn Strathern: "When a measure becomes a target, it ceases to be a good measure." The instant you reward a metric, you stop measuring the world and start measuring people's ingenuity at moving that metric. The most expensive version on record is Wells Fargo, where account-opening targets pushed staff to open some 3.5 million accounts customers never authorized — the metric soared while the reality it stood for was being destroyed. Same mechanism as my intake bonus, four orders of magnitude more expensive. Every owner has some version of this running right now. You just can't see it, because it lives in the gaps between numbers you look at separately.
60 pts
the gap between 90% acceptance and 30% conversion — pure fraud chasing a bonus, completely invisible until two metrics were placed next to each other. AI's job is to place them next to each other, constantly, at a scale you can't.
What an AI audit actually does
A human reviews the dashboard they built. An AI reviews the business. Point it at your numbers and tell it to hunt for the seams, and it does the thing no manager has time for: it checks every metric against every other metric, all the time, looking for the pairs that should agree and don't.
- It finds the gaps between metrics. Acceptance vs. conversion. Hours billed vs. outcomes delivered. Leads in vs. revenue out. The fraud and the leakage live in the gaps.
- It doesn't have a bonus to protect. Your AI auditor isn't on the comp plan. It has no reason to make the funnel look healthy, which makes it the most honest analyst in the building.
- It reads what dashboards skip. Call recordings, notes, the messy unstructured exhaust of the business where the real story usually hides.
- It never gets bored. The reason gaming survives is that catching it is tedious and constant. That's precisely the work a machine does best.
Your dashboard was built to confirm what you expect. An AI audit is built to find what you'd rather not know.
The owner's discipline
Auditing your own business is uncomfortable, which is exactly why it pays. Run it on yourself first, before you aim AI at customers or growth, because the cheapest money in any company is the money it's currently losing without noticing. Pair every volume metric with a quality metric and let AI watch the pair. Assume every rewarded number is being gamed somewhere and make the seams impossible to hide. And push your bonuses as close to real, retained value as you can — because what you reward is what you'll get, including the part you didn't mean to pay for.
The intake bonus cost me real money and bought a permanent rule: never trust a number you reward to tell you the truth on its own. AI is how you enforce that rule at the scale of a whole business — an auditor that never blinks, never games, and never tells you only what you hoped to hear.
Sources & further reading
- Goodhart's Law — overview and Strathern's formulation. en.wikipedia.org "When a measure becomes a target, it ceases to be a good measure."
- Wells Fargo cross-selling scandal — ~3.5M unauthorized accounts from sales-target pressure. en.wikipedia.org The billion-dollar example of a metric destroying the reality it measured.
- Harvard Business Review — "Don't Let Metrics Undermine Your Business." hbr.org On surrogation — teams optimizing the proxy and forgetting the goal it stood for.