PostHog experiments follow a 3-stage pipeline: flags define variants, metrics define success, and analysis validates the outcome.
Step 1
Feature Flags
Create a multivariate flag with control and variant. Assign users deterministically by distinct_id.
→
Step 2
Metrics
Define a primary metric (e.g. activation rate) and guardrail metrics like api_error or churn.
→
Step 3
HogQL Analysis
Run SQL queries to segment by tenant size, join with revenue data, and check guardrail metrics the panel misses.