Skip to content

Case · M4 × Return (consumer subscription)

Duolingo: smarter reminders, not more reminders

Duolingo used a bandit algorithm to choose the wording of each daily practice reminder while holding notification volume flat. Over a strong baseline, total DAU rose 0.5% and new-user retention 2% (peer-reviewed, KDD 2020).

M4 ×ReturnConsumer subscription

The outcome · 2020

The bandit algorithm lifted total DAU 0.5% and new-user retention 2% over a strong baseline, without sending more notifications (Yancey & Settles, KDD 2020; company authors, peer-reviewed).

Case grade for the number: A. Evidence grade for the mechanism link: B.
Read the source

Stage cell: M4 × Return (consumer subscription).

What they did. Instead of sending more notifications, Duolingo optimized which reminder text each learner received, using a multi-armed bandit that treats recently seen messages as temporarily less effective ("sleeping, recovering" arms). Volume stayed flat on purpose.

The mechanism. A daily reminder tied to the learner's own goal is an owned trigger HK-TRG31 (Hooked Ch.7) whose cadence bridges the gap until an internal trigger forms HK-TRG12 (Hooked Ch.2). Novelty decays, so rotating fresh, relevant copy keeps the trigger alive without the cost of extra volume. Test before you fear notifications, and test against a strong baseline HK-TRG37 (Hooked Ch.7) F26.

The outcome. +0.5% total DAU (2020) and +2% new-user retention over a strong baseline Case grade A. From a regulator, a public filing or a peer-reviewed study..

What it means at your stage. At M3-M5, tune reminder content and timing before volume. Track opt-outs and uninstalls as guardrails, and evaluate long enough to see novelty fade.

Caveats. Peer-reviewed, but written by company authors about their own product. Mechanism link Evidence grade B. Moderate evidence: solid studies with some limits..

Would you have called it?

This case is also a round of Predict the Outcome: read the situation, pick what you think happened, then see the evidence.

Play this round

Try this

The techniques behind it

Each one is a starting point, not a recipe. Gamify It turns a technique into an experiment card for your product: the change, the metric, the guardrail and how long the test needs at your traffic.

  • TechniqueEvidence grade B. Moderate evidence: solid studies with some limits.

    A reminder tied to the plan the user chose

    if-then reminder at a chosen time

    Let the user pick a goal and a time during setup, then send one reminder at that time that names their own plan and opens today's unit directly.

  • TechniqueEvidence grade D. Practitioner framework or anecdote: plausible, untested.Practitioner heuristic

    Size reminders to the behavior's natural cadence, base them on events the user cares about, and send fewer as self-started use grows. Reminders bridge the weeks or months before an internal trigger forms; they shouldn't become the reason people return.

  • TechniqueEvidence grade B. Moderate evidence: solid studies with some limits.

    Before adding or cutting a message, send it to most users and hold out a random group. Compare sessions, opt-outs, uninstalls and organic shares, so one good or bad anecdote doesn't set the cadence.

Limits

What we don’t claim

Sources

Source IDs: CASE-M01HK-TRG31HK-TRG12HK-TRG37F26

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

Your product

Find the stage your product is stuck in

Your product has its own version of this story. Find the stage it’s in and the move that fits it. Connecting is free; a Pro subscription starts it.