Retention levers

When a cohort decays toward zero, resist "add engagement features" first. Walk a short diagnostic: wrong users, never activated, activated but no habit, habit broken by quality, or competition/seasonality. Retention work is often subtraction and reliability, not net-new surface area.

Know your team's retention definition and where the curve flattens. When scoping features, ask which part of the curve they move (activation, engagement intensity, resurrection, churn save). Treat notification systems, sync reliability and cold-start empty states as retention infrastructure.

The big idea

Habit loops pair a trigger (time, notification, peer action), an action (open, scroll, create), a variable reward and an investment that makes the next visit better (content created, network grown, setup completed). Engineers implement the boring parts: reliable reminders, fast loads on the reward path and data models that remember investment.

Notifications are high-leverage and easy to destroy trust with. Good ones are timely, personal and tied to a reason to return ("Priya commented on your doc"). Bad ones optimize open rate while training users to disable the channel. Measure notification-driven retention against opt-out rate and long-term engagement; never celebrate CTR alone.

Resurrection (win-back email/push, "what's new" digests, seasonal hooks) can lift a smile curve. It is cheaper than net-new acquisition when the dormant base is large — but if the product still does not fit, you are spamming. Segment by reason for churn when you can (price vs value vs forgotten).

Product quality on the critical path. Latency, bugs and confusing IA on the daily path quietly destroy retention. A growth team cannot out-email a broken sync. Treat reliability on core workflows as a retention feature.

For paid products, split:

• Voluntary churn: user cancels (price, missing feature, switched, low usage).
• Involuntary churn: failed payment, expired card. Often fixable with dunning, retries and reminders — some of the highest-ROI retention engineering in SaaS.
• Contraction: seats or usage drop without full cancel — shows up in NRR.

Build a simple cancel survey and look at usage before cancel. Users who never activated will say "too expensive" when the truth is "never got value." Link reasons to cohorts in the warehouse; see product analytics.

Walking through it

2,000 monthly subscribers. 80 cancel: 50 voluntary, 30 involuntary. Of voluntary, 28 had fewer than 2 sessions in the prior 30 days. Dunning might recover half of involuntary (~15). Activation and engagement work targets the 28 low-usage cancels. A blanket 20% discount to all cancels mostly pays people who already decided the product was not a habit.

Another case: week-4 retention fell from 24% to 16% over a quarter. Channel mix shifted from 40% paid to 70% paid. Organic week-4 stayed at 28%; paid stayed at 11%. The product did not suddenly get worse — the blend did. Tighten paid creative to match the job organic users already hire you for, or accept lower blended retention and price CAC accordingly.

Diagnostic order: (1) compare retention by channel and signup reason; (2) split activated vs not; (3) if activated, is there a reason to return at natural frequency?; (4) error rates and "works on web but not mobile" on the retained path; (5) cohort by calendar month before panicking at seasonality.

Common mistakes

• More push as the first lever. Ask for opt-out and long-term retention impact.
• Discounting every cancel. You pay people who never got value.
• Adding another tab that dilutes the core habit instead of deepening it.

Check yourself

1. 80 cancels, 30 involuntary, 28 voluntary with <2 sessions. Where is the highest-ROI work?

Dunning for ~15 recoverable involuntary; activation/engagement for the 28 low-usage voluntary. Not a blanket discount.

2. Blended week-4 fell 24% → 16% while organic stayed 28% and paid stayed 11%. Product regression?

No. Mix shifted toward paid. Fix targeting or price CAC for the true paid curve.

Key takeaways

• Diagnose the curve (users, activation, habit, quality, mix) before adding features.
• Levers: habit design, careful notifications, resurrection, critical-path quality.
• Split voluntary vs involuntary churn; dunning is often the cheapest save.

Further learning

Lenny Rachitsky's talk on growing startups covers retention benchmarks, activation and common pitfalls.

Back to the map: retention.