Why a user is here decides whether they stay. This skill covers the reasons that exist before any mechanic is applied to them, and that survive when the mechanic is switched off.
Two halves. The first is segmentation — finding out which cohort actually retains, which is almost never the cohort your dashboard is loudest about. The second is attachment — making an optional behaviour feel like belonging rather than like a cost.
It does not own the mechanics that manufacture obligation (streaks, wagers, leagues), which live in [[../duo-streak-mechanics/SKILL]], nor the cue and surface work of getting a session started, which is return-trigger territory. Come here for the reason; go there for the machinery.
Segment by why, not by who
- [[references/segment-by-stated-motive-not-demographics]] — rank cohorts by the reason they gave at signup; relational motives out-engage instrumental ones, and externally imposed use is the most fragile fuel of all.
- [[references/identity-motives-outlast-utility-motives]] — a goal tied to who someone is has no completion condition and no substitute, so your smallest segment is often your most committed one.
- [[references/segment-by-life-stability-not-enthusiasm]] — a daily habit needs an unbroken daily context, so the most interested cohort is frequently the least consistent; check the age curve of your return metric before you design for it.
Read your engagement numbers honestly
- [[references/intensity-and-frequency-are-two-axes]] — session depth and return frequency are independent; name the off-diagonal cells, because the shallow-but-frequent and deep-but-rare users need opposite fixes.
- [[references/the-growth-cohort-is-not-the-retention-cohort]] — novelty-seeking and persistence are different traits and often inversely distributed, so split acquisition and retention by segment or you will build for the loud new users.
Build the reason to come back
- [[references/attach-usage-to-a-dated-external-event]] — a real dated event the user already committed to supplies both a deadline and a picture of success that an open-ended aspiration cannot.
- [[references/attachment-not-exposure-sustains-optional-behaviour]] — for behaviour nobody requires, volume is not the lever; ship the recurring cast, the shared marker and the ritual that turn a cost into an identity.
How to use this skill
- Start with segmentation. Most retention arguments are two cohorts being averaged together and mistaken for one population.
- Split your engagement metric into its two real axes before you pick an intervention, or you will apply the frequent-but-shallow fix to a deep-but-rare user.
- Only then reach for attachment work. It is slow and hard to attribute, which is why teams skip it for impressions that will not move anything.
Sibling skills
- [[../duo-streak-mechanics/SKILL]] — the commitment counter itself, and how much slack to build into it.
- [[../duo-learner-motivation/SKILL]] — what to do when the block is the user's belief that they are bad at the thing.
- [[../duo-character-systems/SKILL]] — designing the recurring cast that attachment is built on.
- [[../duo-proprietary-data-reports/SKILL]] — publishing segment data externally without collapsing it into one score.
Sources
Every claim here is cited inline in the reference nodes against a dated blog.duolingo.com post; the bibliography of crawled posts lives in scripts/sources.json.