Level Funnels: How to Find the Level Where Players Quit

Every game has a level that eats players. Somewhere between the tutorial and the credits there's a jump that's too long, a boss with too much health, or a timer that's three seconds too tight. Players hit it, sigh, close the app, and never come back. They won't email you about it either.

A level funnel helps you find where progression drops. Pair it with attempt completion rates and fail reasons to choose a level to investigate. This guide explains the difference between reach counts and an ordered funnel, how to compare the right players, and what to test once a level stands out.

The screenshots are from AnyAnalytics with sample data from our demo game, Space Hopper. The method works in any analytics tool.

What is a level funnel?#

An ordered funnel follows the same cohort through a sequence of steps within a chosen conversion window: started level 1, then level 2, then level 3. Its bars get shorter as fewer of those players progress.

A per-level reach report is different: it counts players who reached each level in the selected date range. Returning players can start at higher levels without replaying level 1, so those counts need not decline. Use Insights → Funnels when you need a defined sequence and conversion window; use the Game report to inspect level reach and attempt results.

The question is never "do players drop off?" They always do. The question is where the drop-off is bigger than it should be.

Step 1: Send three events#

These three events power the Game report. Send a start for each attempt and exactly one completion or failure for its result, with the same level value:

analytics.track("level_start", { level: 12 });
analytics.track("level_complete", { level: 12, duration: 92, score: 3400 });
analytics.track("level_fail", { level: 12, duration: 38, reason: "timeout" });

Use the same level value on the start and end of an attempt, include duration in seconds, and give fails a reason from a short, fixed list. If you're in Unity, here's the five-minute setup. For the full list of events worth tracking, see 8 events every mobile game should track.

Step 2: Read "players reached"#

In AnyAnalytics, the Game → Players reached view shows per-level reach from your events. Start by checking the date range and which players it includes. The screenshot below is sample data with 1,000 players at level 1 and 749 at level 12:

Enlarge screenshot

Explore the Game demo to try these views with sample data.

This sample has no sudden cliff. That shape is a useful starting signal, but it does not establish a healthy retention rate on its own. Compare new players with enough time to progress, and separate platform and game version. A recent player who has not reached level 12 may still be playing.

Here's what to look for:

Step 3: Check fail rates per level#

Reach counts suggest where to look. The Start → finish view adds evidence by showing how each level's attempts end; it cannot establish the cause by itself:

Enlarge screenshot

Three columns do most of the work:

• Fail rate. Fails as a share of starts. A smooth climb is a difficulty curve. A sudden jump is a wall.
• Median time. If one level takes three times as long as its neighbors, players might be lost rather than challenged.
• Hardest level. AnyAnalytics flags the level with the highest fail rate among levels started at least 10 times, so one unlucky attempt can't crown a level. In this sample it's Level 56, which players fail 52.9% of the time.

High fail rate + small drop-off = hard but fair. Players retry and get through. That's often fine, even fun.

High fail rate + big drop-off = a wall. Players try, fail, and leave. Fix this one first.

Step 4: Find out why they fail#

Break level_fail down by the reason property in a trend chart to test a possible explanation:

Enlarge screenshot

Add a filter for level = 12 and you're looking at how that level beats people:

• Mostly timeout: investigate the timer, route length and instructions. Test a longer timer and compare fail rate and progression.
• Mostly died: inspect enemy damage, trap spacing and checkpoints. Separate failures by attempt number if you track it.
• Mostly quit: investigate length and clarity. Only explicit quits you instrument are counted; closing or killing the app may leave an attempt without a result.

The Where players are view shows the level each player started most recently in the report's scope. A crowd on one level is a reason to investigate, not proof that those players are currently online or permanently stuck.

Bonus: the funnel before level 1#

Levels aren't the only funnel in your game. You can chain any events in Insights → Funnels. This one follows players from opening the game to their first purchase:

Enlarge screenshot

In this sample, 40% of the starting players did not reach the level-start step within the selected funnel window. For them the main menu is level zero, and it's the hardest one in the game. Real games see this too: long loading screens, a sign-in wall or a menu with nowhere obvious to tap. Let new players start playing in one tap.

The median time between steps is worth checking as well. If "opened the game → started a level" takes minutes instead of seconds, something is in the way.

A worked investigation#

Suppose a new-player cohort has a larger drop between levels 11 and 12 than neighboring steps. First confirm that all cohorts had the same time to progress. Then:

1. Filter level_fail to level = 11, since a drop before reaching level 12 may originate in level 11.
2. Break failures down by reason and compare platforms and app_version.
3. If timeout failures rose after one update, inspect the timer and any level changes in that build. Treat it as a hypothesis.
4. Ship one change, such as more time, and compare equally mature new-player cohorts.
5. Look for both fewer timeouts and more players reaching level 12. A lower fail rate alone is not enough.

This is an example investigation, not a measured result from Space Hopper.

How to fix a problem level#

Once you've found the level, change one thing at a time and watch the numbers:

1. Ship one fix per version. Every event carries your app_version, so you can filter the report to the new build and compare it with the old one.
2. Make later attempts easier after repeated failures. Test more time, a checkpoint or a hint, then compare progression as well as fail rates.
3. Teach before you test. If a new mechanic appears in the level players fail, add an easier level that introduces it first.
4. Move the gate. If a paywall or energy limit lines up with your cliff, players may be leaving because of the gate, not the level.
5. Re-check in a week. You want the fail rate down and more players reaching the next level, not just one of the two.

FAQ#

What is a good level completion rate?#

There's no universal number, since it depends on genre and on where the level sits. Compare each level with its neighbors instead. A level that's much harder than the ones around it is the problem, whatever the absolute rate.

What's the difference between drop-off and fail rate?#

Fail rate is about attempts: how often a level beats the player. Drop-off is about people: how many of the starting cohort did not reach the next level within the conversion window. A level can have a high fail rate and almost no drop-off if players happily retry.

How many players do I need before the numbers mean anything?#

The more the better. A rough guide: wait for at least a few dozen attempts per level before reacting. AnyAnalytics only names a "hardest level" once it's been started at least 10 times.

Can I build a funnel for my tutorial?#

Yes. Send tutorial_step_completed with a numeric step property, then chain that event in Insights → Funnels with a different step filter on each stage. Choose a conversion window that gives new players enough time to finish.

Find your level zero#

Start by verifying three level events, then let enough players attempt your levels. The Game report gives you reach and attempt results; an ordered funnel lets you follow a cohort through the next steps.

Start free with AnyAnalytics (no card required), send the level events, and open the Game report to choose your next investigation.