Retention & Metrics

Cohort retention, key metrics, and knowing if you've made something people want.

Why Cohort Retention Is the Best Metric

David Lee, YC partner who built Google Photos to over a billion users, calls cohort retention the single best way to answer: "Did we make something people want?" Instead of looking at all users mixed together, track individual groups of new users (cohorts) over time.

The Core Insight

The only thing that matters is whether your cohort curves get flat. A product that retains 20% of users forever is fundamentally different from one that retains 80% initially but trends toward zero. Flat at 20% is better than declining from 80% to 0%.

2026 Benchmarks

Seed Stage

80–85% annual retention is good. NRR 95–105%. Don't panic at early churn.

Series A

88–92% annual retention. NRR 100–112%. Growth stage: 93–97%.

How to Calculate

Define cohort

Group new users by signup time (week or month).

Pick the active action

What counts as real value? Instagram: viewed 3+ posts. Uber: completed a ride.

Choose time period

Daily for social apps, weekly for utility, quarterly for travel.

Build the triangle chart

Each row is a cohort, each column is a subsequent period. Numbers can go UP — users who left may come back.

How to Improve Retention

Improve the product

New features, better speed, simpler flows. Look at chronological cohorts — if middle ones perform better, improvements are working.

Acquire better users

Often the product is fine but you're targeting wrong customers. Slice cohorts by channel, country, or customer type.

Improve onboarding

Cheapest fix. What was the customer doing yesterday? What do you want them to change today?

Build network effects

Products where every subsequent user makes it better for existing users.

The Retention Mistakes That Kill Startups

✗ Picking too large a time period — if you expect daily use, measure daily

✗ Picking too easy an action — 'opened the app' doesn't prove value

✗ Using payment as the action — users stop using first, then stop paying

✗ Looking at a single point instead of the full curve

✗ Trusting analytics tools blindly — build your own curves first

Next: Growth Strategies →