Where Do Players Quit Your Game? How to Find Out
List the stops a player is supposed to pass, in order, and count how many got to each one. The biggest fall between two neighbouring stops is where to look first, as long as enough people got that far.
What does a list of stops look like with real numbers?
Salt Road Couriers is an invented delivery game, and everything below is made up for the example. The maker wrote down the path he planned for the first day, then counted 200 sessions. A session is one sitting.
| Stop | Players | Went on from the stop before | Very likely between |
|---|---|---|---|
| Started the game | 200 | ||
| Learned to steer the cart | 171 | 86% | 80% and 90% |
| Made the first pickup | 158 | 92% | 87% and 96% |
| Made the first delivery | 92 | 58% | 50% and 66% |
| Made the second delivery | 84 | 91% | 84% and 96% |
| Got paid for day one | 79 | 94% | 87% and 97% |
People call this a funnel, because the list gets narrower as players leave. The last column is a sure range: with this many players, the true share very likely lies in there.
Do not start with the bottom line. Read it stop by stop.
Between the first pickup and the first delivery, 66 players were lost. The other four steps together lost 55. And the range for that step, 50% to 66%, does not touch any other, the next lowest of which starts at 80%. So this is not luck. Something happens on that road.
The next question is what those 66 did last. In the log, 48 of them ended on the event cart-tipped, 11 on map-opened, and 7 on other things. So nearly three in four left straight after the cart fell over.
The maker looked at the game and saw it. A tipped cart drops the parcel, and the player has to go all the way back to the depot. On the first day, with no skill yet, that is a long walk for a small mistake.
What should you change at that stop?
Change the one stop. The maker did not touch the tutorial, the prices or the second delivery. He made the first road wider and let a tipped cart keep its parcel. His target: at least 80% get from pickup to delivery. Telling 58% from 80% takes about 70 players per version, so he can check that with a small group.
Some cautions before you do the same:
- A session ending is not a verdict. People leave because dinner is ready. That is why you look for many people leaving at the same place, doing the same thing.
- You cannot see who came back. With anonymous data, someone who returns tomorrow looks like a new session. That is the price of not knowing who anyone is.
- Check the plan first. If a later stop ever shows more players than an earlier one, an event is missing or there is a way round. Fix that before you trust anything else.
With only 20 sessions, the same list would be a hint at best, so watch a few people play instead. The drop-off recipe does this count for your AI and refuses to call a stop the worst one unless it clearly is.
Paste this to your AI
Find out where people stop playing my game. I have an event log in [file or folder].
1. From my game, draw the path a player is meant to take as a Mermaid state diagram. Include every way round a step, and show me which event proves each stop. Wait for my yes.
2. Use get_recipe with drop-off-funnel if Spielerando is connected. If not, count it yourself: a player counts for a stop if they reached that stop or any later one.
3. Show a table: stop, players, share who went on, the 95% range. Then name the worst stop, and whether its range is clear of the next worst.
4. List every place the log does not fit your diagram, such as a later stop with an event when an earlier one has none.
5. For the worst stop, show the last event of those who left there, with counts.
Then give me one change at that stop and the number that would show it worked. Tell me what you actually know and what is your guess. Where fewer than 30 players reached a stop, say it is too thin to judge.
Or skip this and connect the MCP: your AI then finds all of this by itself.Connect the MCP
How to check it worked
- Add it up. Players who left at each stop, plus those at the end, must equal the players at the start.
- Go there. Play that exact stop as a new player. Does the problem show up?
- Measure again after the fix. Did only that step move? If everything moved, something else changed too.