The first real problem I found was not a broken process. It was a broken instrument — a number everyone trusted, reported every month, that wasn't measuring what anyone thought it was.

I had just started as chief operating officer of a twenty-person professional services firm, fully remote. In week two I asked for the intake numbers, because intake is where owner-led firms usually leak first. The consultation show-up rate came back at 24%.

Twenty-four percent is a catastrophe. It says three out of four people who booked time never appeared. If that number were real, nothing else in the business would matter until it was fixed: not pricing, not staffing, not marketing spend. I was ready to spend my first quarter on it.

The following month the same report read 74%.

Nothing had changed. I hadn't touched intake. No process had been rewritten, no one had been retrained, no reminder sequence had been added. A number had tripled because I had started asking about it, which is the clearest possible sign that you are looking at a measurement problem rather than a performance problem.

The cause was mundane. The system was recording consultations when they were scheduled and, in most cases, never recording whether they were completed. Scheduling created a row. Completion required someone to go back and mark it, and mostly nobody did. So the denominator was accurate and the numerator captured only the fraction that had been closed out by hand that month. The 24% was a measure of administrative diligence. It had almost nothing to do with whether clients were showing up.

That's a more interesting finding than a bad intake rate, and a more common one.

Most owner-led businesses run on numbers nobody has ever tested

The reports exist. They arrive on schedule, they have decimal places, and they have usually been arriving for years. What has almost never happened is someone sitting down and asking whether the number describes the thing its label claims.

This isn't incompetence. It's a predictable consequence of how small firms grow. Someone configures a system during a busy quarter, sets up a report that looks right, and moves on. The business scales, staff turn over, the original assumptions leave with the person who made them, and the report keeps arriving. By the time it matters, the number has the authority of long service. Nobody questions the metric that has been on the dashboard since before they were hired.

The cost isn't that decisions get made on bad data, though they do. The cost is that bad data is confidently actionable. A 24% show-up rate points you at a specific fix, and you can spend a whole quarter executing that fix competently against a problem that doesn't exist. Missing data makes you cautious. Wrong data makes you fast in the wrong direction.

Three tests

Before I trust any number in a business I have just walked into, I run it through three things. None of them takes long, and all three can be done before you have any real understanding of the operation.

Reconcile it against something that doesn't share its plumbing. If the metric comes out of the CRM, check it against something outside the CRM: the calendar, the payment processor, the ledger, the phone records. A number that only agrees with itself hasn't been verified. In the case above, half the metric reconciled to something real and half reconciled to nothing: a booking creates a record whether or not anyone follows up, while a completion existed only if someone went back and said so.

Ask what would make it move, then check whether it moved. This one catches more than anything else, and it's the test the show-up rate failed in front of me: the figure tripled across two consecutive months in which nothing had been done to it. A metric that changes sharply when you've done nothing is measuring your data collection. A metric that stays flat through a genuine intervention is doing exactly the same thing in the other direction. Before you act on a series, look at what happened to it during a month when you know something real occurred, and see whether it noticed.

Find its logical ceiling and see whether it ever breaks it. A conversion rate can't exceed 100%. A utilization figure that sits above capacity for a whole quarter is telling you something about the formula, not the team. This is the fastest of the three and the one people skip, because a number above its own ceiling looks like good news.

That third test came back to me later. Once the tracking was rebuilt, individual months occasionally read slightly above 100%, and I want to be precise about what that meant, because it's a distinction worth carrying. The metric was captured as a snapshot rather than as a cohort. A consultation booked at the end of one month and completed in the next counted as scheduled in the first and completed in the second, so the boundary leaked in both directions and single months could exceed the ceiling. Across a quarter it washed out.

That's a measurement basis, not an error — but you only get to say so if you know which one you're looking at. The test isn't “never show a number above 100%.” The test is: can you explain it? If you can't, you've found something.

What we actually did

I did the first reconciliation myself, which isn't a good use of a new chief operating officer and was unavoidable. The scheduled side of the record was already sound, so I listed every scheduled appointment across a three-month window and took that as the denominator. Then I went to the three people responsible for those appointments and asked them to complete the record — what had actually happened to each one. That's reconstruction, and it deserves to be named as such. But it wasn't pure recall: the same people were already logging meeting notes, whether the person converted, and whether a fee was collected, all recorded at the time and for other reasons. Most appointments could be anchored to one of those. Where money had moved, the consultation had happened. That went into the dashboard I was building, and only after it started returning figures that behaved sensibly did anyone get retrained.

The order was the point. You can't teach a process against a number you haven't established, and every week spent retraining before the audit would have been a week spent correcting behavior we could not yet measure. Then I made the ongoing version as small as I could. One new field — a consult completed date — and two automated reminders prompting the team to fill it. That's the entire mechanism.

That split (a reliable scheduled count and a fictional completed one, inside the same system) is worth carrying, because it generalizes. The accurate half was the half the software captured on its own. The fictional half was the half that required a person to go back and enter something after the fact. If you want to know which of your numbers to distrust, start there. And when you fix them, don't ask people to try harder. Shrink what they have to do until it is a single field, move the prompt inside the workflow so it doesn't depend on recall, and let the software do the chasing.

The first clean quarter ran 74, 77 and 70 — call it 74% on a quarterly basis. Nine months later we were running 95% on the same basis. Both of those numbers are real in a way the 24% never was, and the improvement between them is the one I'd defend, because it was measured the same way at both ends.

But the thing I'd tell an owner isn't about intake. A business can spend years unable to see one of its own core conversion rates, and nobody notices, because the report never stops arriving. If you have a metric you have never reconciled, you don't have a metric. You have a habit.

That is why the first step of the method I use is the one most businesses skip: the TAG method — See the Truth. Implement Actions. Achieve your Goals. Setting a goal against a number you haven't tested isn't planning. It's guessing with a spreadsheet open.