How to build a metrics system
A metrics system is a tree: one number at the top for direction, broken down to the layer a single change can move.
A metrics system is a tree: one number at the top for direction, broken down to the layer a single change can move. When the tree doesn't grow, it's usually not the analytics tool's fault — it's that nobody thought through the number at the top.
What you'll run into:
- Twenty numbers on the dashboard and you don't know which one to watch
- You shipped a feature but can't say which number it should improve
- Two people's "daily active users" don't match up
Start with the top metric
The top metric must match the product goal, and it shifts with the lifecycle. The same product has a different number at the top in different stages.
| Stage | Business focus | Top metric |
|---|---|---|
| Introduction | Polish the core feature, get the first users | Total users, week-1 retention |
| Growth | Activate and keep existing users | Activity, retention rate |
| Maturity | Monetize | Paid conversion, average order value, GMV |
| Decline | Cut costs, find a new growth point | Acquisition cost, profit, validation data for the new direction |
Common top metrics fall into six families: revenue, acquisition, activity, retention, referral, and composite. When choosing, ask: if this number improves, has the product genuinely gotten better?
One more filter: a metric must be actionable. A number you cannot move with any strategy is not worth watching. The top metric gives everyone one shared reading of whether the product is getting better — but it doesn't tell you what to do this week. That's what breaking it down is for. Pick the top one wrong and no amount of fine splitting helps.
Break it down
There are a few common ways to break a metric down. Pick one that fits your situation; don't mix them.
| Approach | How to use it |
|---|---|
| AARRR | Acquisition, activation, retention, revenue, referral — set second- and third-level metrics per stage. Good for viewing the whole funnel |
| User flow | Follow the steps users actually take and set a metric per step. Good for locating where people get stuck |
| OGSM | Objective, goals, strategies, measures. Emphasizes separating process metrics from outcome metrics |
| OKR | Objectives plus key results, then break each key result into concrete metrics |
Keep process metrics and outcome metrics separate. Outcome metrics (orders, revenue) are lagging and hard to touch directly; process metrics (search conversion, coupon redemption) are where you can actually optimize. When do you stop breaking down? Stop when the layer can be moved by one concrete change.
Define one metric clearly
Ambiguous metrics are the most common source of reconciliation fights on a team. A complete metric has three parts:
Time window + qualifier + atomic measure. In "articles browsed in the last 7 days," the window is 7 days, the qualifier is "browsed," and the atomic measure is article count. Drop the window and you can't compare anything; drop the qualifier and nobody knows which behavior counts — does opening count as active, or only reading? "Daily active users" so often fails to reconcile because it was never defined.
Once metrics are set, write them up as a metrics dictionary: name, formula (how it is computed), and constraining dimensions. Everyone on the team checks the same document — that is how you stop everyone quoting different numbers.
A lighter version for solo builders
- Just one at the top level. During exploration, that's week-1 retention. Ignore everything else for now.
- Three to five at the second level. These map to the key steps of the main flow, used to locate where a problem sits.
- Add the third level on demand. When you spot an issue in one segment, break that segment down further. No need to lay it all out from day one.
- Write the definition for every metric. Even if it's just you — in three months you'll forget.