AI Customer Growth + Retention — Onboarding, Lifecycle, Churn Watch

What's in this pack

This is the rig for the growth or PMM lead at a SaaS who has been told to lift activation, cut churn, and prove it with numbers — not vibes from a Slack thread. Every pick here is either open source, self-hostable, or has a real free tier — and where possible, has an MCP server or webhook so an AI agent can read the data instead of you screenshotting dashboards into the standup.

The stack covers the five loops growth teams actually run: measure → segment → nudge → retain → close the loop with feedback. The order matters: you cannot A/B test what you have not instrumented, and you cannot nudge a segment that does not exist yet.

Install in this order

1. PostHog (#1615) — analytics foundation. Event tracking, funnels, retention cohorts, session replay, feature flags, and a SQL layer in one self-hostable app. This is the single dependency every other pick assumes you have running. Start here even if you keep GA on the marketing site.
2. PostHog LLM Observability (#2826) — if your product has AI features, wire LLM call traces into the same warehouse. Now "users who hit the AI chat" is a segment, not a guess.
3. PostHog Session Replay (#2828) — watch ten replays of users who churned in week one. Saves a quarter of survey work. PII auto-masked so legal stops asking.
4. GrowthBook (#2516) — open-source A/B testing and feature flag platform that reads your warehouse. The activation funnel is where most retention is won; this is the cheapest place to experiment.
5. Intro.js (#2421) — in-product onboarding tours and tooltips. Once a cohort shows it stalls at step 3 of activation, you have one week to ship a nudge before the data goes stale. Intro.js is the boring, reliable choice.
6. Claude Code Agent: Customer Success Manager (#4298) — an agent skill that ingests usage signals + support tickets and drafts the at-risk-account brief your CSM team actually reads. Closes the gap between PostHog and a human follow-up.
7. Formbricks (#457) — open-source NPS, CES, and in-product micro-surveys. Self-hostable; feeds responses back into PostHog as events so churn predictors get a real label.
8. Listmonk (#918) — self-hosted newsletter and lifecycle email engine. Cheap, fast, no per-contact pricing. Pair with PostHog cohort exports for trigger-based campaigns.
9. Resend Audiences (#2982) — modern email API with managed audience lists, useful if you do not want to run Listmonk and need deliverability that Just Works. Has an MCP server, so an agent can move a user between lists.
10. Mautic (#1773) — full marketing automation suite when you outgrow scripted Listmonk and need visual journeys, lead scoring, and a CRM lookup. Heavyweight; install last and only if needed.

How they fit together

[Your app + AI features]
   │  events, traits, LLM traces
   ▼
PostHog (analytics + replay + flags)
   │            │                │
   │   cohort exports             feature flag gates
   │            │                │
   │            ▼                ▼
   │   Listmonk / Resend     GrowthBook (A/B tests)
   │   (lifecycle email)         │
   │                             │ winner ships everywhere
   ▼                             ▼
Formbricks (NPS) ──▶  back into PostHog as event
   │
   ▼
Customer Success Manager agent ──▶ at-risk brief for CSM

The loop everyone forgets: NPS responses go back into PostHog as a property on the person, so your churn cohort and your detractor cohort can be the same SQL query. Without that wiring, NPS is just a vanity dashboard.

Tradeoffs you'll hit

• PostHog vs Countly vs Matomo — PostHog wins for product analytics + replay + flags in one box. Countly is leaner if all you want is mobile event tracking. Matomo is web-marketing analytics, not product analytics — different job.
• GrowthBook vs PostHog Experiments — PostHog has experiments built in; GrowthBook is a more mature, dedicated tool that can read from your warehouse without ingesting events twice. Pick GrowthBook only if you already have a warehouse you trust.
• Listmonk vs Resend Audiences — Listmonk is free and self-hosted, but you own the deliverability problem. Resend has paid pricing but inboxing Just Works and the MCP server is a force multiplier. Most teams: Listmonk for in-product transactional, Resend for marketing.
• Mautic vs everything above stitched together — Mautic is one product instead of five, but the install is heavy (PHP + DB + queue) and the UI is dated. Worth it only when you have ≥3 marketers running campaigns and need a shared journey builder.
• Intro.js vs custom React onboarding — Intro.js is 10 KB and works in 20 minutes. A custom React tour is prettier but eats 2 weeks. Ship Intro.js first; replace later if the rough edges actually hurt.

Common pitfalls

• Instrumenting after launching the experiment — by far the #1 mistake. Wire PostHog first, ship at least one funnel report, then start the activation tour. Without a baseline, you cannot say the nudge worked.
• Sending lifecycle email from your transactional domain — destroys your sender reputation in a quarter. Use a subdomain (mail.yourapp.com) for marketing, keep transactional on the root domain or a separate one.
• Treating NPS as a number — the number is useless on its own. The free-text comments are the gold. Pipe every Formbricks response with text into an LLM tag-extractor and group by week.
• Cohort drift from rename churn — every time marketing renames a plan, your retention cohort math silently shifts. Bind cohorts to event properties (plan_tier_at_signup), not display names.
• Letting the CSM agent autoreply to detractors — never. The agent drafts; a human sends. Auto-replied apologies make detractors angrier than no reply.

What to measure week one

1. Activation rate by signup week (PostHog funnel: signup → key action within 24h).
2. Week-2 retention by segment (free vs trial vs paid, and by acquisition channel).
3. NPS by cohort (Formbricks → PostHog, segmented by plan + tenure).
4. Lifecycle email opens by stage (Listmonk / Resend dashboards).
5. At-risk account count (CSM agent flags; review weekly with the human CSMs).

If any of those five numbers does not exist in your dashboard by end of week one, fix that before installing tool #6.