Two signals of product-market fit. The PMF survey asks users a single question. Retention curves watch what they actually do. Each catches what the other misses.
Last updated: 2026-04-01
A single survey question popularized by Sean Ellis: "How would you feel if you could no longer use this product?" The PMF threshold is 40% or moreof users answering "very disappointed".
Best for early-stage products with active users but limited cohort data. Gives a signal in days, not months.
Cohort retention plotted over time. For each cohort, you track what percentage are still active at D7, D14, D30, D60, D90. The PMF signal is the curve flattening to a plateau.
Best for products with at least three to six months of cohort data. Retention curves are slower but harder to fool.
% who answered "very disappointed" / total respondents. Threshold = 40%.Ellis published 40% after analyzing more than 100 startups. He noted the number isn't magical, it's a benchmark. Don't agonize between 38% and 42%.
For each cohort: active_users(t) / cohort_size, plotted at intervals.A healthy curve drops fast as casual users churn, then flattens into a stable plateau. The plateau represents your core users.
| Criteria | PMF Survey | Retention Curves |
|---|---|---|
| Type of signal | Stated preference | Behavioral data |
| Time to result | Days | Months |
| Threshold | 40%+ "very disappointed" | Curve flattens to a stable plateau |
| Origin | Sean Ellis, after analyzing nearly 100 startups | Cohort analysis tradition; Andrew Chen popularized for PMF |
| Sample size | 100+ responses per persona | Multiple cohorts of meaningful size |
| Risk of false positive | Higher. Self-selecting respondents | Lower. Hard to fake using the product |
| Pairs with | NPS, jobs-to-be-done research | LTV, churn, cohort analysis |
| Best stage | Early product launch | Growth and scaling stage |
Pros
Cons
Pros
Cons
Score your own data with both frameworks. Compare results and pick the one that fits your team.
Sean Ellis published the 40% threshold after analyzing over 100 startups. Above 40% "very disappointed", products consistently grew. Below it, growth was rare. Ellis himself emphasized the threshold isn't magical. It's a benchmark. Don't sweat the difference between 38% and 42%. Sweat the difference between 25% and 50%.
You can have a high survey score and a curve that's still stable but at a low level. That usually means a small but very engaged user base. Andrew Chen ranks flattening retention curves as the strongest behavioral signal of PMF. A flattening curve plus a strong survey score is the cleanest combination.
Steep drop in the first days as casual users churn out. Then flat. The plateau height matters by product type. Consumer apps often target D30 above 40% to call it strong PMF. SaaS targets lower D30 but with higher revenue per retained user.
At least 100 responses for each persona. Below that, the 40% threshold is statistical noise. If you have multiple personas, sample 100 from each. Filter to users active in the last 30 days, otherwise inactive churned users dilute the signal.
Yes. Use the PMF survey to get a fast signal at three to six months in. Use retention curves once you have enough cohort data to see plateaus. The two metrics rarely disagree, and when they do, retention curves are the more conservative answer.