Auto-survey churned users and read the answers in chat

Why this matters

Email-based exit interviews get 15% response rates. People hate writing emails. They especially hate writing emails about why they left a product they don’t care about anymore.

In-app surveys get 60%+ response rates. Right place, right time, low friction, single click.

The problem is most teams have churn data in one system (Stripe), users in another (PostHog), surveys in a third (Typeform). Wiring them together is a project. Agentry has the cohort + survey + events all in one place — the agent stitches them in one prompt.

Then the reading of the responses is the second hurdle. 200 free-form text answers is a lot to wade through. Asking the agent to “summarize what people said this month” turns it into 5 bullet points you can actually act on.

What you get

• A 3-question survey that triggers when a user cancels
• A cohort churned_users updated daily
• A weekly Routine that summarizes new responses and DMs you the highlights

Walk through it

agentry_create_cohort {
  project_id: "default",
  name: "churned_users",
  filter: { has_event: "subscription_canceled", in_last_days: 30 }
}

{ "cohort_id": "co_churn30" }

agentry_create_survey {
  project_id: "default",
  name: "Cancellation feedback",
  type: "popup",
  targeting: { cohort: "co_churn30" },
  display_conditions: {
    show_when: "always",  // show on next page load after cancellation
    dismiss_after_responses: 1
  },
  questions: [
    {
      type: "single_choice",
      question: "What's the main reason you canceled?",
      choices: [
        "Too expensive",
        "Missing a feature I need",
        "Found a better alternative",
        "Not using it enough",
        "Bugs / reliability issues",
        "Other (please specify)"
      ]
    },
    {
      type: "open_ended",
      question: "What would have made you stay?",
      placeholder: "One sentence is fine"
    },
    {
      type: "rating",
      question: "If we fixed that, would you come back?",
      scale: 5,
      labels: { 1: "Not a chance", 5: "Definitely" }
    }
  ]
}

{ "survey_id": "sv_can1", "active": true }

/schedule "0 9 * * MON" "
  Fetch responses to survey sv_can1 from the last 7 days.
  Group by question 1 (cancellation reason).
  For question 2, extract themes from the open-ended responses (cluster similar feedback).
  For question 3, compute average willingness-to-return.
  Format as a brief markdown report and DM it to me via Slack webhook
  https://hooks.slack.com/services/.../...
"

{ "routine_id": "rt_chsm", "next_run_at": "2026-05-19T09:00:00Z" }

The output

A weekly Slack DM in your #me-only channel:

📋 Cancellation Survey — Week of 2026-05-19

Responses: 23 (out of 41 cancellations — 56% response rate)

WHY THEY CANCELED:
  Missing feature              9   ████████ 39%
  Not using enough             6   █████    26%
  Too expensive                4   ███      17%
  Found alternative            3   ██       13%
  Bugs / reliability           1   ▌         4%

THEMES from "what would have made you stay" (open-ended):
  • Want better Slack integration (5 mentions)
  • Want native mobile app (4 mentions)
  • Pricing should have an annual discount (3 mentions)
  • UI is confusing during onboarding (3 mentions)

AVG willingness to return if we fixed it: 3.4 / 5
  3 respondents said "Definitely (5)" — they're the highest-priority winbacks

ACTIONS THIS WEEK:
  → 3 hot winback candidates: [email protected], [email protected], [email protected]
    (their reasons are addressable — DM in #cs to assign)
  → Top theme: Slack integration. Already on Q2 roadmap.
  → Top "easy" win: annual discount — talk to billing.

Setting it up

Three pieces:

1. Fire subscription_canceled on cancellation. Agentry has no SDK — this is a raw POST to /v1/analytics/:

// In your Stripe webhook handler or cancellation endpoint
await fetch(`https://api.agentry.sh/v1/analytics/${process.env.AGENTRY_PROJECT_ID}/`, {
  method: "POST",
  headers: {
    "Authorization": `Bearer ${process.env.AGENTRY_DSN}`,
    "Content-Type": "application/json",
    "User-Agent": "myapp/1.0",  // REQUIRED — Cloudflare 403s default UAs
  },
  body: JSON.stringify({
    event: "subscription_canceled",
    distinct_id: user.email,
    properties: {
      reason_self_reported: cancelReason,  // from the cancel form, if any
      plan: user.plan,
      tenure_days: daysSince(user.subscription_started_at),
    },
  }),
});

2. Make sure the PostHog snippet is loaded on your app (this is what actually renders the survey in the user’s browser — surveys are PostHog-served). Agentry’s MCP wires this automatically via agentry_install_guide, but if you set up analytics manually, ensure the PostHog posthog-js snippet is loaded. The PostHog snippet handles the survey UI; Agentry is the brain that creates / targets / reads it.

3. (Optional) Add a manual cancellation page that links to the survey directly. For users who fully bounce from the app, an email with the survey URL gives a second shot:

Hi {{first_name}},

You canceled last week. We'd be grateful if you spent 60 seconds telling us why:
https://agentry.sh/s/sv_can1

It's three questions. Doesn't go to a marketing team — it goes to me directly.

— Henrik

In-app surveys outperform email, but the union of both maximizes response rate.

Variations

• “Create a survey for users who downgraded (not canceled) — different cohort, slightly different questions.”
• “Show a different survey to users who hit a specific error 3+ times — ‘is this stopping you from using us?’”
• “After we ship the Slack integration, survey the cohort of users who said that was their reason — ‘we built it, want to come back?’”
• “Group survey responses by company size — do enterprise churners say different things than indie hackers?”