Publishers ask about D1 first, but they decide on D7. A puzzle game with a strong first session and a weak first week is a game people liked and then forgot. Seven-day retention is the earliest number that says "this is a habit", and it is the number that determines whether user acquisition can ever pay back.
We track cohort retention for every game on our framework from the first test build, so we have spent a lot of time looking at what moves D7 and what merely looks like it should. This is a summary of that experience. It is written for producers and publishing managers rather than data scientists, and it deliberately avoids quoting our own titles' figures, which we share with partners under NDA.
First, what D7 is not
D7 is not a bigger D1. The levers are different. D1 is mostly about the first session: does the mechanic read, does the tutorial get out of the way, is there a reason to open the game tomorrow? D7 is about the second to sixth sessions: does the game give the player a structure for coming back, and does it keep its promise when they do?
That means the most common "retention fix" — polishing the first five minutes — mostly moves D1 and D2. Useful, but not the same job.
Lever 1: Session structure
Players return to games that fit into their day. For puzzle games that means the session has a natural length and a clean end. Our most consistent D7 gains have come from tightening session structure:
- Chapter shape. Sessions that end on an easy "release" level after a hard peak (see our difficulty curve article) end on a win. Players who end on a win come back.
- Lives and energy. Systems that stop a session at a predictable point are not evil if the stopping point is satisfying. A life lost on an unfair level is a churn event; a life lost after a real attempt is a reason to come back.
- Daily goals. A short, visible list of things to do today — three levels, one event round — gives the second and third session a purpose.
Lever 2: Difficulty pacing across days, not just levels
Difficulty design is usually discussed per level. For D7 what matters is difficulty per day. If day 2 contains the first real spike and the player has not yet built any attachment, they leave. If day 4 is flat, they get bored.
We plan the week explicitly: where the first peak falls, when the first new mechanic arrives, when the first event opens. Then we check the cohort data — retention by day and the level distribution of churned players — to see whether the plan matches reality.
Lever 3: Meta goals that mature over a week
A meta layer is not there to be deep; it is there to make tomorrow more interesting than today. The features that have paid off for us in D7 terms share a property: they build across several days.
- Win streaks that grow over days and reset on a miss create a reason to protect the streak tomorrow.
- Collections and building — a city, a garden, a collection book — where each day's play adds something visible.
- Weekly leagues where position updates daily and the reward lands at the end of the week.
Features that resolve inside a single session (a spin, a one-off chest) are fine for monetization and delight, but they do not move D7 much on their own.
Lever 4: Events with a calendar
Time-limited events are the most reliable D7 lever we have, provided they are on a predictable cadence. A player who learns that a new event starts every Monday has a reason to check the game on Monday. A player who sees random events has a reason to shrug.
The framework we use runs events through remote configuration, so the calendar is a design decision rather than an engineering release. Each event reports participation, completion and — importantly — the retention of participants versus non-participants in the same cohort. That last comparison keeps us honest: an event that participants love but that does not lift their next-week return is entertainment, not retention.
Lever 5: Fixing the levels that eat cohorts
The least glamorous D7 lever is also the one with the fastest payback: find the three levels between day 2 and day 6 where the most players quit, and fix them. Per-level funnels make this a weekly routine rather than a quarterly investigation. Most fixes are small, and each one returns a slice of the cohort.
What has not moved D7 for us
Some things we tried that improved other metrics but did not move D7 in a measurable way:
- More boosters in the shop. Revenue moved; retention did not.
- Push notifications alone. Without a reason to return inside the game, reminders mostly move D2 and then fade.
- Bigger tutorials. Beyond making the mechanic readable, extra teaching moved D1 down, not D7 up.
- Cosmetic polish after a certain point. Polish protects retention; it does not create it.
How to run the analysis
If you want to reproduce this kind of work on your own titles, the minimum dashboard is:
- Cohort retention by install day, filterable by acquisition source and country.
- Last-level-played distribution for churned players in each cohort.
- Per-level funnels with fail rate, retry rate and quit-after-fail.
- Event participation joined to retention of the same cohort.
- Session count and length by day since install.
With those five views you can attribute most D7 movements to a specific cause within a week of a change shipping. Without them, retention discussions become opinion — and opinion is expensive at UA prices.
We build this dashboard set into every game on our framework. If you would like to see how it works on a real title, get in touch.
