The retention curve that signals product-market fit falls steeply, then flattens into a horizontal plateau. A curve that keeps decaying toward zero shows no lasting retained base, as described by Lenny's Newsletter quoting Brian Balfour in its product-market fit guide. Here product-market fit means a core group keeps using the product without push.
A cohort retention curve tracks one signup group over time to reveal that core. AI startups face noisy early curves because curious tourists try the product and leave. Andreessen Horowitz advises rebasing retention from Month 0 to Month 3 for that reason in its AI retention benchmarks.
Table of Contents
- What does a fit curve look like?
- Why judge AI fit after Month 3?
- What retention level counts as good?
- How do founders misread retention?
What does a fit curve look like?
Look for a fast early drop followed by flat retention. The drop reflects casual users leaving after first tries. The flat part reflects repeat users who found lasting value. A curve without a plateau keeps sloping down month after month.
That shape means new signups only replace churn. It does not show a self-sustaining user base. Founders can sharpen this read by splitting users by use case and onboarding path. Superhuman offers one cross-check model: it defined fit as 40% or more users saying they would be very disappointed without the product.
Why judge AI fit after Month 3?
AI tourist churn distorts signup-month behavior. Users arrive fast after a launch, test one prompt, then disappear within a couple months. Judging fit on Month 0 mixes those tourists with real adopters.
Top AI revenue cohorts break into acquisition from M0 to M3, retention from M3 to M6-M9, and expansion from M9 onward. Andreessen Horowitz reports that phased pattern in its analysis of hundreds of AI companies. Founders should therefore judge product-market fit after M3. Earlier reads reward novelty more than retention.
What retention level counts as good?
For monthly-billing products, use the retention curve as a direct fit signal. Other billing models need adjusted interpretation because usage gaps do not mean the same loss. Consumer founders need an external target once the curve flattens.
At about six months, roughly 25% retained is good and 45% is great for consumer social, according to Lenny's Newsletter in its consumer startup benchmarks. An AI consumer startup can use that range to pick a comparable goal. A flat curve below that range still shows some fit, but a weaker base.
How do founders misread retention?
Exact-day measurement understates returning users by 3.5 times. Only 8% returned exactly on Day 30 and 6% on Day 60, versus 23% returning sometime in Days 15-30 and 20% in Days 31-60, according to an Amplitude press study of 500M devices in its retention measurement study.
Use return windows, not exact-day checks, before calling a curve flat or decaying. Track return windows, not exact days, before you declare fit.
- Compare same-length monthly cohorts, not mixed launch spikes.
- Rebase AI curves at M3 before judging the plateau.
- Treat monthly billing as direct signal and adjust annual or usage billing.
- Pair the curve with a disappointment survey to confirm value.