The Numbers

I once spent an afternoon with a dashboard that had about forty charts on it.

It was good work. Somebody had built it carefully and kept it up to date. When I asked which of the forty had changed a decision in the last quarter, the answer, after a pause, was that they mostly confirmed what people already thought.

A dashboard watches the business. It does not change it. Those are different jobs and only one of them is worth doing.

What Works

The Numbers is the small set of measures that would change a decision, arranged so you can see which ones move early and which ones the board reads.

Two levels, and the split is the whole design.

Lagging measures are the outcomes. They confirm whether the strategy worked. You cannot manage them directly; you can only influence them. These belong to the company.

Leading measures are what you manage. They move first, and they are the ones a team can actually affect. These belong to teams.

Most companies have a pile with both kinds in it and no idea which is which, which is why reviews turn into interpretation.

Why it works, and the two tests

A leading measure earns its place by passing two tests. Both, not either.

One. We have genuine influence over it. Not correlated with our outcomes. Something we can move this week.

Two. We can show that the relationship to the outcome is real, with evidence rather than hope. And we write the evidence down.

That second test is the one that does the work, because it is very hard to fake and almost nobody applies it.

Here is what passing looks like.

First value inside 14 days. Accounts that reach first value inside 14 days renew at 96%. Accounts that do not renew at 71%. Three years of our own data, 340 accounts. Observed in our history, not assumed. Confidence: High.

And here is what honesty looks like when a measure does not have that behind it.

Onboardings completed without a named specialist. The only measure of whether this scales at all. We have no evidence for this one, and we should say so, because we have never done it. Confidence in the relationship: Low. This is what the experiment tests.

That second block is worth more than the first. Everybody's metric pack looks like the first one. Writing down which of your measures you are least sure about is what makes the whole model believable, and it is what tells you where to spend your testing.

The early warning signal

One more thing on the page, and it is cheap.

Live accounts past day 14 without first value, checked weekly. Above twelve and the programme is stalling.

The lag between this and the retention number is about eleven weeks. We need to know in week two, not in the quarterly review.

Every company has one of these available, and most have never written it down. It is usually a count; it is usually already in a system somewhere, and it usually leads the number the board cares about by two or three months.

How it goes wrong

Correlation dressed as a driver. A number that moves alongside your outcome and cannot be moved by you. It looks like a leading measure, and it is a weather report.

Too many. If reviewing them takes more than twenty minutes, it is a dashboard again.

And nobody edits it. The model is a set of claims about how your business works. Some of them will be wrong. Reviewing it monthly and changing it on evidence is the point; a metrics model nobody has edited in a year is a metrics model nobody has tested.