Choose a PostHog Experiment Primary Metric
Choose the metric that most directly represents the hypothesis, define it before inspecting results and add related guardrails. A large collection of unrelated metrics creates more opportunities to mistake noise for success.
Before you start: Have a concrete hypothesis, validated events and a meaningful exposure definition. Agree what outcome would justify a rollout and what harm would block it.
Worked example: A clearer invite flow aims to increase accepted invitations among eligible workspace owners. Track successful invitations as the primary outcome and failed submissions as a guardrail.
Step-by-step guide
Write the causal expectation
Describe how the change should affect user behavior and why. Keep the metric close enough to that behavior that a result can inform the decision.
Choose the counting unit
Decide whether the experiment concerns people or groups. Match assignment and analysis to the unit that can independently experience the change.
Define the metric
Select an appropriate supported metric type and specify event filters, denominator and window. Test a known successful and unsuccessful journey.
Set related guardrails
Choose a few plausible harms such as errors or abandonment. Keep unrelated marketing or site metrics outside the decision rule.
Record the decision criteria
Agree the practical effect worth acting on and how uncertainty will be handled. Avoid rewriting the success definition after seeing a favorable secondary result.
Review before launch
Ask another analyst or engineer to verify exposure and metric events. Save the definition with the experiment so later readers can distinguish planned from exploratory analysis.
Common mistakes
Easiest event wins
A convenient click may not represent user value. Choose the intended outcome.
Metric chosen after results
Selecting the favorable metric introduces bias. Define it first.
Unrelated guardrails
A long dashboard does not strengthen the hypothesis. Monitor plausible side effects.
Tips
Keep the primary metric definition in one sentence.
Test the full event sequence before launch.
Label unplanned analyses as exploratory.
Sources and verification
Based on product documentation checked on 2026-10-24. This is a documentation-based guide, not a hands-on benchmark. Features and availability can change.