Zentrik + PostHog

PostHog is the evidence source. Zentrik turns each saved product question into a fixed aggregate measurement, checks it on a schedule, and keeps one durable Insight current as the result changes.

Use the Insight in an idea or brief when the evidence should change product work. Zentrik does not infer a cause or change an active measurement between runs.

A workspace owner or admin can connect PostHog and select the project.

1. Open Settings → Integrations → Connections and select PostHog.
2. Approve the read-only scopes in PostHog: project:read and query:read for measurements, plus event_definition:read, property_definition:read, action:read, cohort:read, and insight:read so Zentrik can read the names and descriptions in your project.
3. Select Open Imports when you return to Zentrik.
4. Choose the PostHog project that contains the events you want to measure.

The connection supports PostHog US and EU Cloud. Reconnecting updates the existing workspace connection. Zentrik does not need a PostHog API key.

After you choose a project, Zentrik reads its catalog: event and property definitions, actions, cohort names, and saved Trends and Funnels insights. Imports shows the catalog counts and when they were last read. A connection made before these scopes existed keeps working; Imports asks you to reconnect when a catalog section needs a scope you have not approved yet.

1. Select New question.
2. Ask about one tracked behavior, such as Are more users completing onboarding each week?, or about an ordered process, such as How well does our onboarding flow convert? Use the name of a behavior or process that your project already records.
3. Choose a daily or weekly check.
4. Select Define measurement. For a single behavior, review the exact PostHog event, measure, audience, and comparison. For a process, review the ordered steps Zentrik matched from your project's real events.
5. If Zentrik cannot match the question safely, choose the correct event from the live project list.
6. Select Preview result. Save the question only when the result has enough evidence for your decision.

You can save more than one question. Each question has its own schedule, Insight, and evidence history. The saved definition stays fixed until you edit it.

A question about a process, such as a signup flow or a participant flow, becomes a funnel: an ordered sequence of two to six real events. Zentrik reads the people who reached the first step in a 30-day window and reports how many reached each step, the conversion from the step before, the median time between steps, and the largest drop in the sequence. The window ends 14 days before the check, so everyone in it has had the full conversion window, and the 30 days before it are read as the comparison. A change of 5 points or more that is unlikely to be sampling noise is reported as improving or worsening; anything smaller or noisier is reported as stable. A funnel needs at least 20 people at the first step and 5 at the last before Zentrik reports a result; until then it says so plainly rather than showing a number built on too little evidence.

• Current Insight shows the newest supported result and evidence window.
• Evidence history keeps the result from each run.
• Use in an idea carries the Insight into product work. It does not require prior acceptance.
• Pause stops scheduled checks without deleting the question or its history.

If a user rejects the Insight, Zentrik removes it from active product context and later runs do not recreate it.

• Zentrik reads a bounded catalog of definitions from the selected project: event and property names and descriptions, actions, cohort names and sizes, and saved insight names. It does not read cohort membership.
• Zentrik reads aggregate query results from the selected project.
• Zentrik stores aggregate measurements, evidence windows, versions, and safe run receipts.
• Zentrik does not copy raw events, people, or session recordings.
• A single-behavior question counts event occurrences or unique users over complete periods in the project's timezone. A process question reads two funnel windows, both closed a full conversion window ago, in the project's timezone.
• A model can map the initial question to one event or to an ordered sequence of events from your project's catalog. It cannot invent an event, calculate the result, or rewrite an active query.