PostHog + Koji: Turn Funnels and Cohorts into AI User Interviews

PostHog tells you what happened. Koji tells you why. Connect the two and every funnel drop-off, churned cohort, or power-user segment becomes an AI interview — voice or text — triggered automatically the moment PostHog identifies the behavior. Then the themes and structured scores flow back so your analytics and your qualitative "why" live in one loop instead of two disconnected tools.

As the classic framing goes, "analytics tells you what happened; it does not tell you why." PostHog is the best-in-class what: events, funnels, cohorts, session replays, and feature flags. Koji is the AI-native why: moderated interviews at scale, no recruiter and no moderator required. Wiring them together closes the oldest gap in product research.

What data flows between PostHog and Koji

The integration is bidirectional:

• PostHog → Koji (trigger): When a user enters a cohort, completes (or abandons) a funnel step, or fires a tracked action, PostHog sends an event that launches a personalized Koji interview and delivers the link in-app or by email.
• Koji → PostHog (write-back): When the interview finishes, Koji sends a koji_interview_completed event back into PostHog carrying the top themes, an AI summary link, and typed scores from your structured questions — so you can build cohorts and funnels on the qualitative result.

Because Koji's outputs come from structured questions with typed values — a scale satisfaction score, a single_choice reason for churning, a yes/no "would you recommend" — they become event properties in PostHog you can filter, break down, and correlate against the behavioral data you already have.

Three ways to connect

1. Real-time with PostHog's CDP (recommended). PostHog ships a built-in customer data platform with 50+ realtime destinations, including a generic HTTP webhook destination that fires as events arrive, filtered by event, property, or SQL. Point a destination at Koji's API to launch interviews. One practical note: destinations that filter on a cohort require the cohort to use person-property filters only — behavioral dynamic cohorts are evaluated on a schedule, not in realtime, so trigger those via a scheduled export instead.

2. No-code with Zapier or Make. PostHog's Zapier trigger is Action Performed — it fires when a saved PostHog Action runs, so define the Action first (for example "abandoned checkout"), then wire it to Koji's start-interview step. Make is a first-class PostHog destination as well. On the return trip, send Koji's completed event back into PostHog via its Capture endpoint.

3. AI-driven with Koji's MCP and API. Koji ships a Model Context Protocol integration (15 tools) plus a developer API. Query a PostHog cohort via its API, feed the persons into a Koji study, launch interviews, and pull structured results back — all driven from Claude or your own code.

The best PostHog signals to interview on

Funnel drop-off cohorts. A funnel shows you where users abandon; it never shows you why. Build a dynamic cohort of users who reached step three of onboarding but never finished, trigger a Koji interview, and let the AI probe the exact friction. You turn a red bar on a chart into a fixable, prioritized reason.

Churned or dormant users. Define a cohort of users who were active and then went quiet. A Koji exit interview captures the real reason — the workflow that broke, the competitor they switched to, the value they never reached — instead of a support ticket you never got.

Power users and new-feature adopters. Cohort your most engaged users, or everyone who touched a feature behind a flag, and interview them about what to build next. Pair PostHog's behavioral truth (they actually use it) with Koji's stated priorities (what they want) for roadmap decisions you can defend.

Where PostHog Surveys end and Koji begins

PostHog has its own in-app Surveys — lightweight popover and widget prompts with display targeting. They are perfect for a quick in-context micro-question: a one-tap CSAT, a "did this help?" thumbs. Koji is the complementary depth layer. When the micro-survey surfaces a signal worth understanding — a dip in CSAT, a spike in a churn reason — trigger a Koji interview to get the two-minute explanation the popover was never meant to capture. Use both: PostHog Surveys for breadth and in-the-moment reactions, Koji for the adaptive, moderated conversation that explains them.

What syncs back to PostHog

After each completed interview, Koji sends an event carrying:

• Interview status and a completion timestamp
• Top themes surfaced in the conversation
• Structured question scores — a scale satisfaction value, a single_choice churn reason, a yes/no recommendation signal — each an event property
• A link to the full transcript and Koji report

Now you can build a PostHog cohort of "users who churned AND cited missing integrations," or break a retention chart down by interview-verified satisfaction — analysis that fuses the behavioral and the qualitative in one place.

Why a conversation beats another dashboard

Every product team eventually hits the same wall: the dashboards are green or red, but nobody in the room can say why. Adding more events does not answer a why question; only a person can. The trouble is that traditional user interviews do not scale — recruiting, scheduling, and moderating a dozen sessions can eat a week.

Koji removes that cost. Its AI interviewer runs the conversation in voice or text, asks adaptive follow-ups so a vague answer becomes a concrete reason, and supports all six structured question types in a single interview — open_ended, scale, single_choice, multiple_choice, ranking, and yes_no — so you capture a clean quantitative signal to correlate in PostHog and the narrative behind it at once. See the structured questions guide for how to combine them.

Koji's quality gate keeps the data clean: each session is scored for effort and coherence, so low-effort responses are flagged, excluded, and never consume credits. Koji's free tier includes 10 one-time credits to test a drop-off loop; paid plans start at €29/mo (Insights) and €79/mo (Interviews), with text interviews costing 1 credit and voice 3.

A worked example: closing a drop-off loop

Your PostHog onboarding funnel shows a 38% drop between "connected data source" and "created first report." The chart tells you where users stall; it cannot tell you why. Build a dynamic cohort of users who connected a source in the last 14 days but never created a report, and add a PostHog CDP webhook destination (or a scheduled export, since this is a behavioral cohort) that calls Koji.

Koji invites each stalled user to a short interview built from three structured questions — a single_choice "what stopped you from creating a report," a scale "how clear was the next step," and an open_ended "walk me through what you expected to happen" — with adaptive AI follow-ups. As interviews complete, Koji fires koji_interview_completed events back into PostHog carrying the reason and the satisfaction score as properties. Now you break the funnel down by interview-verified reason and find that most stalled users could not find the report button, not that they lost interest. One UI fix, discovered from the why behind the what — and measurable in the same PostHog funnel afterward.

Related Resources

• Structured Questions Guide — the 6 question types that make interview data correlatable
• Product Analytics vs User Research — the what-vs-why distinction in depth
• Mixpanel + Koji Integration — the same pattern for Mixpanel
• Amplitude + Koji Integration — for Amplitude cohorts
• AI Interviews vs Surveys — why adaptive follow-ups win
• Webhook Setup — the real-time path
• How the Quality Gate Works — why low-effort responses never pollute your data