Sub Club by RevenueCat · 2026-01-21
Why Your Free Users Are Your Real Growth Engine – Cem Kansu, Duolingo CPO
Cem Kansu, Duolingo CPO (joined 2016 to figure out monetization), walks through Duolingo's monetization history: failed crowdsourced translation model, streak repair IAP, end-of-lesson ads with strict category blocking and complaint-volume analysis, an ad-free month IAP test that led to Duolingo Plus (no ads, offline, support the mission), free trials, unlimited Hearts in the subscription, CTA copy tests ('Try for $0.00'), ML-chosen internal subscription ads, and moving paid features back to free to protect word of mouth. Also: separate subject apps were 'hard mode'; human tutors flopped.
Play the episode17 tactics
Numbers mentioned
| Subject | Metric | Value | Quote |
|---|---|---|---|
| Duolingo | share Subscription share of total revenue; ads are a much smaller share | 80% approximate | “So, I think, it's above 80% and ads is a much smaller percentage.” |
| Duolingo | conversion rate Share of MAU who pay (subscribe) | 10% approximate | “So 10% of our users subscribe.” |
| Duolingo | price Streak repair in-app purchase, Duolingo's first monetization experiment | USD 1–3/one-time approximate | “But I think the first thing that users saw at Duolingo as a thing they could buy was repairing your streak for I think a dollar or $3.” |
| Duolingo | price One-time 30-day ad-free purchase tested at $3, $5 and $10 | USD 3–10/one-time | “And we tested different price points, we tested $3, $5 and $10 and everyone thought this wouldn't work.” |
| Duolingo | team size Headcount when Cem joined, before monetization | 60 employees approximate | “So until the time I joined, there was 60-ish employees.” |
| Duolingo | count Concurrent A/B tests; a good chunk on the purchase flow | 400 A/B tests approximate | “So Duolingo concurrently runs somewhere around 400 AB tests.” |
| Duolingo | count Complaints about a missing language course vs complaints about ads (normalized per DAU), in the pre-ads-launch analysis | 3 x approximate | “we looked at, for example, the number of complaints about why we didn't have the finished language on Duolingo was three extra number of people that are complaining about Duolingo having ads.” |
| Duolingo | time to milestone First idea to full launch for each new subject (math, music, chess) | 6–12 months approximate | “these projects all took probably somewhere from six to 12 months to go from the first idea to full launch.” |
| aggregateProduct pipeline speed-up from AI in prototyping and coding, at Duolingo and other companies the guest sees | growth rate Productivity improvement from AI | +10–20% estimate | “Really the speed up of the product pipeline is in the 10 to 20% range.” |
| arpu US ad revenue per user, host recollection | USD 250per user /year approximate | “I think I saw 250 per user per year in the US” |