Metric drop diagnosis
Locate where a number moved before you decide what to do about it.
A metric that dropped tells you something changed — not what, where, or why.
Weekly active users on a B2B tool fell 9% last week. Leadership wants a fix by Friday.
What do you do first?
One number, many stories
Weekly active users is a sum across enterprise and self-serve, new and tenured, mobile and web. A 9% dip could be one large account offboarding, a tracking change that stopped logging events, a seasonal lull, or a genuine product regression.
Reacting to the aggregate treats all of those as the same problem — and each needs a completely different response. The skill is refusing to act on the top-line number until you've found the slice it moved in.
How to localize a drop
- 01
Split by segment
New vs returning, plan tier, account size, geography. Which cohort moved?
- 02
Split by surface
Which feature, page, or step lost usage — not the product as a whole?
- 03
Split by time
A cliff on one day points to a release or outage; a slow slope points to something structural.
- 04
Check instrumentation
Rule out a tracking or pipeline change before you trust the number at all.
- 05
Form one hypothesis
Name the single most likely cause the slices point to — then test that, not everything.
- 06
Then act
Now the fix is aimed at a cause, not a symptom.
The last time a metric moved on a product you know, what was the first cut you made — and what did it let you rule out?
Strong PMs don't fix the metric that moved. They find the slice it moved in, then fix the cause there.