Mid-core RPG

How a mid-core RPG studio used retention cohorts to cut early D7 churn

Picture a small mid-core RPG studio — roughly a dozen people, one live title with a few hundred thousand monthly players, and a tutorial-to-mid-game loop they had been polishing for a year. Installs were steady, but the team had a nagging feeling: a large share of new players seemed to vanish in the first week, and nobody could say exactly where or why.

The challenge

Their old dashboard reported a single D7 retention number that drifted between releases without explanation. When a build shipped and D7 dropped two points, the team could argue about it for a week and never agree on a cause. They had three open questions and no clean way to answer any of them: Which cohort was leaving? Where in the early game did players stall? And did the new-player experience actually get better, or worse, after each balance patch?

What they did with Keentics

The team started with the retention matrix: rows by install day, columns by day offset. Immediately the shape of the problem was visible — a sharp drop between D1 and D3, well before the content most players had complained about. The cliff wasn’t at the hard boss everyone assumed; it was earlier, around the first dungeon.

To confirm it, they built behavioral segments: players who cleared dungeon one versus those who attempted and abandoned it. Splitting the retention curves by those two groups made the gap undeniable — clearers retained far better, and a meaningful slice of installs never cleared at all.

Then they reconstructed the early-game funnel: tutorial complete → first hero unlock → dungeon one entered → dungeon one cleared. The biggest leak sat at “entered → cleared.” A path / Sankey view showed why: a chunk of players bounced off a gear-check wall, wandered back to the shop, and quit without the currency to progress. The mechanic was working as designed; the pacing wasn’t.

The fix was a content change — softening the gear gate and adding a guided reward — shipped behind a version tag. Because Keentics computes retention on full event data, they could read the new cohort’s matrix the same way and compare it column-by-column against the previous build, instead of waiting for a blended average to settle.

The result

In scenarios like this, once the early stall is removed, teams typically see the D1→D3 cliff soften and D7 retention recover by roughly one to two percentage points over the following cohorts — directional figures, not an audited claim, and the exact lift depends entirely on the game. The more durable win was process: a repeatable loop of matrix → segment → funnel → path → ship → re-read by version that turned “D7 dropped, argue about it” into a question the data could actually answer.

For the team, the unlock wasn’t a single magic number. It was finally being able to see which players left, where, and whether a change helped — within days of shipping it, and without exporting anything to a spreadsheet.

This is a representative mid-core RPG scenario rather than a named customer story, but the workflow — and the questions it answers — is exactly how retention analysis tends to play out in practice.