Metrics and KPIs
Not every number is a KPI
A metric is a measurement of an activity; a KPI is a metric tied to a strategic objective with a target. Good metrics are comparable, actionable and resistant to gaming.
What you will learn
- Distinguish metrics, KPIs and operational measures
- Write a metric definition with numerator, denominator and filters
- Evaluate a metric's quality and gaming resistance
What is this?
A metric is a defined, repeatable measurement (e.g., sessions per day). A KPI is a metric tied to a strategic objective with a target and an owner (e.g., 'increase free-to-paid conversion from 3.0% to 3.5% this quarter'). Operational measures run dashboards; KPIs run the business. The difference is purpose and accountability, not math.
Why do you need it?
- Strategic alignment: everyone steers toward the same few numbers.
- Accountability: a KPI has an owner and a target, so performance is discussable.
- Faster learning: a defined metric can be compared over time and between teams.
- Terminology hygiene removes the 'my revenue vs your revenue' fights.
Mental model
A car's instruments vs. its safety rating
Metrics vs KPIs
Speed and fuel (metrics) tell you the current state; the safety rating (KPI) tracks whether you're progressing on the strategy of 'safe vehicles'. Both matter; only the KPI is strategy-bound.
How it works
- Define the measurement precisely: what is counted, over what window, at what grain, with which filters.
- Write the formula — numerator and denominator — so two analysts compute the same number.
- Attach a target and a cadence, and assign an accountable owner.
- Instrument the metric in the warehouse, then validate against a hand-computed sample.
- Review the KPI set periodically; metrics that no longer drive decisions should retire.
From activity to KPI.
In production
- A subscription company tracks NRR and churn as KPIs, and trial starts/activation as supporting metrics.
- An e-commerce team sets checkout conversion as its weekly operational metric and CSAT as a quarterly KPI.
- A SaaS board reviews 5 KPIs monthly; everything else lives in supporting dashboards.
Best practices
- Define metrics with explicit formulas: numerator, denominator, filters, timezone, grain.
- Keep the KPI set small: a team can honestly focus on 3–5.
- Pair each KPI with guardrails to catch gaming (customer churn balanced by support load).
- Document version: when a definition changes, label old vs new series.
Common mistakes
- Declaring every dashboard number a KPI — nothing becomes accountable.
- Gaming-able targets, e.g., 'reduce tickets' encourages hiding tickets rather than fixing causes.
- Changing the formula after the result disappoints, invalidating comparison.
- Celebrating a KPI that moved for seasonal reasons, not because the work worked.
Debugging tips
- When two dashboards disagree, diff the formulas: one likely includes a filter the other lacks.
- Check KPI spikes against events outside the metric (promos, launches, code changes).
- Recompute the KPI by hand on a 50-row sample to validate instrumentation.
- Version-lock metric definitions when they change so history stays comparable.
Interview checkpoint
Hands-on lab
- Write a precise metric definition from a vague business goal
- Compute the metric from raw CSV with pandas
- Validate the figure against a hand-computed sample
Define and Validate a KPI from Raw Data
25 min · intermediate
Quick quiz
Which of these is a KPI rather than a metric?
A metric formula must include its numerator, denominator, filters and grain.