Retention and engagement

Acquisition without retention is a leaky bucket. You can spend forever pouring users in the top; if they leave, revenue, referrals and compounding growth loops never form. Retention is the percentage of a cohort that still uses the product after a given period — and it is usually the clearest signal of whether you have product-market fit.

For engineers, this is why reliability on the critical path, notifications that do not spam and features that deepen a habit all count as product work. Match the window to natural frequency: daily tools use D1/D7/D30; weekly tools use W1/W4/W12; monthly tools use M1/M3/M6. Using D30 for a monthly product makes healthy customers look churned.

When a growth proposal is "send more push," ask for the opt-out and long-term retention impact. When a product proposal is "add another tab," ask whether it deepens the core habit or dilutes it. Retention work is often subtraction and reliability, not net-new surface area.

Concepts in this section

Read them in order. Each one builds on the one before.

Step	Concept	What it answers
1	The leaky bucket	Why does fixing churn beat buying more traffic?
2	Retention curves	What shape is PMF, and how do I compare cohorts?
3	Engagement metrics	How deep is use, and does DAU/MAU even apply here?
4	Retention levers	Habit, notifications, resurrection, quality, or dunning — which first?

When a cohort looks bad, diagnose before adding features: wrong users (organic flattens, paid decays), never activated, activated but no habit, quality on the critical path or mix/seasonality. For paid products, split voluntary cancel, involuntary (failed cards — dunning is often the cheapest save) and contraction, which shows up in NRR.

SDE2-shaped habits: know your team's retention definition and where the curve flattens; instrument events needed for cohort queries before launch; when scoping, name which part of the curve the work moves. In interviews, draw a retention curve, mark the flat line and propose levers tied to user frequency.

Common mistakes

Scaling acquisition on a curve that decays to zero. Paid channels then accelerate cash burn.
Blended actives as retention. Mix of new, old, organic and paid hides the story.
Copied DAU/MAU targets. A weekly B2B tool at 20% can be healthy; a messenger at 20% is on fire.
Engagement features before diagnosis. Wrong users, never activated and broken sync each need a different lever.

Key takeaways

Retention compounds; acquisition without it only scales the leak.
Flatten above zero for PMF; decay to zero for a miss; smile for resurrection or network effects.
Match metrics to natural frequency; breadth of return before depth of power users.
Diagnose the curve, then pick habit design, careful notifications, resurrection, quality or dunning.