Guide
How to set one metric that proves a fix is working
A fix without a signal is a belief. But most teams pick the wrong signal: a headline number that only moves a quarter later, by which point three other things have changed and nobody can say what worked.
What a good signal looks like
- It moves within weeks. If your first reading is three months away, you cannot correct course.
- It is caused by the thing you changed. Revenue is affected by everything; repeat contacts on one fault type are affected by your fix.
- It already exists, or costs nothing to collect. A signal that needs a new reporting project will not be collected.
- It is hard to improve dishonestly. If the easiest way to move it is to game it, it will be gamed — without anyone intending harm.
- One number, one owner, one place to look.
Leading and lagging, paired
Keep the result you actually want (the lagging number) written down, and pick one leading signal that should move first. Resolution time is the result; repeat contacts on the top fault type is the signal. If the signal moves and the result doesn't, your causal story is wrong — which is useful, and much cheaper to learn in three weeks than in three quarters.
Set the review date before you start
Choose a date at which the signal should have visibly moved, and write down what “moved” means numerically. Deciding this in advance is what stops a stalled fix from being defended after the fact. If nothing has moved by then, do not push harder — return to the evidence and diagnose again.
Where the diagnostic helps
Every diagnosis on this site ends with the evidence of movement to watch for your specific constraint, plus a review date, so the signal is chosen before the work starts rather than justified afterwards.