Metrics, metric trees and guardrails
Which behaviours and outcomes should the team try to change?
How I use it
Name the outcome
Start with what the business needs over a defined period: profitable growth, durable adoption, expansion or something else. Then ask what useful customer behaviour would produce that result. A metric earns attention when it informs a decision.
Break it into drivers
Separate arithmetic identities from causal hypotheses. Completed workflows can be decomposed into active accounts, attempts per account and completion rate. Whether faster setup causes more completed workflows is a hypothesis to investigate.
Keep asking what drives the next level until you reach behaviour the team can influence.
Define the unit and the window
Choose users, accounts, transactions or another meaningful unit. Define the event, eligibility, exclusions, time zone, observation window and source. Reconcile important totals and inspect a few real journeys before trusting the dashboard.
Add guardrails and ownership
Decide what to maximise, minimise or keep within a constraint. More usage might be unhelpful if errors, support demand or AI inference costs rise faster.
Give each important measure an owner and connect it to a review or decision.
Illustrative metric tree
- Active accountsHow many accounts do useful work?
- Starts per accountHow often do they try?
- Completion rateHow often do they finish?
An arithmetic identity: active accounts × starts per account × completion rate. A feature causing one of those drivers to improve is a separate hypothesis.
Weekly completed workflows = weekly active accounts × workflows started per active account × completion rate.
With 1,000 active accounts, four starts per account and 80% completion, the total is 3,200 completed workflows. Improving completion to 85%, with other inputs fixed, adds 200. That is an arithmetic scenario, not evidence that a particular feature will cause the change.
| Measure | Role |
|---|---|
| Completed useful workflows | Product outcome |
| Active accounts / starts / completion | Drivers |
| Failure rate / support burden / cost | Guardrails |
What to watch for
- Choosing a number because it is easy to measure rather than connected to value.
- Multiplying averages drawn from different populations or time windows.
- Allowing one team’s target to damage the customer or another part of the business.
A brief for the next discussion.
A small metric tree, agreed definitions and explicit guardrails.
Know your metrics
Where is your growth work getting stuck?
Use the growth diagnostic to find where to focus, or talk it through with me.