DAU/MAU stickiness

Benchmarks

Every benchmark carries its year and a confidence grade for its source. Use them to calibrate, not as targets.

• Metric

  Cadence-matched cohort retention

  cohort retention at the natural cadence

  Share of a sign-up or activation cohort that does the value event in a later bracket matched to the product's natural cadence (for example week 4 or week 8). Read by segment for curve shape and across cohorts for stacking.

• Metric

  Current-user retention rate (CURR)

  share of current users who stay active

  Share of yesterday's current users (active that day and in the prior week) who are active again today. In Duolingo's growth model, lifting this rate moved daily actives more than any other transition.

• MetricARobustEvidence grade A. Robust evidence: replicated, peer-reviewed research.

  Email unsubscribes per campaign

  email opt-outs per send

  Unsubscribes divided by emails delivered, per campaign. A guardrail for email triggers and win-back flows; lower is better.

• Metric

  Habitual-user rate

  share at the devoted cadence

  Share of activated users who meet the route's devoted cadence in the read week (week 8 by default). Compared with a 5% bar that is Hooked's rule of thumb, not a scientific cut-off, and read on the habit horizon.

• Metric

  Hook cycle time

  time from investment to the next visit

  Median time from an investment that loads a trigger, to that trigger's delivery, to the session it starts. Shorter cycles mean more loops and faster habit formation.

• Metric

  Investment rate

  sessions where users put something in

  Share of sessions in which the user invests something that improves the product with use (content, data, follows, reputation, skill), plus the share of new users who invest in their first session. A leading Return proxy.

Sources

• DAU/MAU is an important metric, but here's where it fails · Andrew Chen, 2018

  APrimaryBenchmark confidence A. A primary source for the benchmark.

Book content is paraphrased, never quoted at length. Last verified October 24, 2026.