Jul 11, 2026
By Victor Teran
Session Replays: How to Find the Screen That Kills Your Retention
A session replay is a recording of a real user moving through your product: every scroll, tap, hesitation, and exit, replayed exactly as it happened.
Analytics
PostHog
UX
It is the closest thing software has to sitting behind the glass in user research, except the users do not know they are being studied and there are thousands of sessions instead of six.
Funnels tell you where users leave. Replays tell you why. The technique is unmoderated usability testing at a scale no research budget buys. At On the User, replays are how we found the onboarding break that Q.AI could not diagnose, the fix that took their activation from 38% to 67% and cut support tickets 73%. This post covers how to use replays without drowning in them: which sessions to watch, what to look for on screen, and how to turn what you see into a fix you can measure.
Key takeaways
Never browse replays randomly. Filter to new users who hit your worst funnel step and left, and watch five of those.
Four signals: hesitation, wandering, rage interactions, and the silent exit. Each points to a different class of fix.
Name the single stumble you saw repeated, change only that, and compare the same funnel step on the next cohort.
Which sessions are worth watching?
Never browse replays randomly. Filter to new users who hit your worst funnel step and left, and watch five of those.
Opened without a question, a replay library is a time sink; teams binge an afternoon of sessions and come away with anecdotes. The discipline is to arrive with a question written down: why do users leave at this specific step?
That means the funnel comes first. Build your signup-to-activation funnel, find the steepest cliff, then filter replays to exactly that population: new users, reached the cliff step, did not survive it, ideally from the last two weeks. In PostHog this filter takes a minute, and it turns the library into a stack of evidence about one question.
Then watch five. Five is enough because product friction is not random: if a screen is broken, it breaks the same way for almost everyone. By the third replay you will start predicting what the user does next. That prediction is the diagnosis forming.
FREE GUIDE
The 4 product leaks costing you growth
A short audit guide for founders. Find the four places your product leaks revenue, and what to fix first.
What should you actually look for on screen?
Four signals: hesitation, wandering, rage interactions, and the silent exit. Each points to a different class of fix.
Hesitation is the long pause on one screen with no input. The user is reading, re-reading, and deciding whether this is worth the effort. Hesitation points to a comprehension problem: unclear value, unclear next step, too much asked too early.
Wandering is movement without progress: opening menus, backing out, visiting settings, returning. The user is looking for something they expected and cannot find. Wandering points to a navigation or expectation problem, often created upstream by marketing promising a feature the first screen does not surface.
Rage interactions are the repeated taps on something unresponsive, or aggressive scrolling. These point to bugs and dead ends, and they are gold because they are unambiguous.
The silent exit is the most important and least dramatic: the user completes a step, lands on the next screen, and simply leaves. No rage, no wandering. That screen failed to give them a reason to continue. Paywalls placed before value produce this signature constantly.
Signal | What you see on screen | What it points to |
|---|---|---|
Hesitation | A long pause on one screen with no input | Comprehension: unclear value, unclear next step |
Wandering | Menus opened, backing out, settings, returning | Navigation or expectation mismatch |
Rage interactions | Repeated taps on something unresponsive | Bugs and dead ends |
Silent exit | Completes a step, lands on the next screen, leaves | That screen gave no reason to continue |
How do you turn observations into a measurable fix?
Name the single stumble you saw repeated, change only that, and compare the same funnel step on the next cohort.
After five replays, write one sentence: "New users stall on the workspace setup screen because it demands a team invite before showing any value." That sentence is the work order. Resist bundling it into a broader redesign; change the one thing you observed.
Then measure like you mean it. Same funnel, same step, next two weeks of new users, before versus after. If the cliff softened, you found the screen. If it did not, you fixed a symptom, and the replays of the new cohort will tell you what you missed.
This loop, funnel to replays to one fix to remeasure, is exactly what we run in the first weeks of every engagement. If you would rather have us run it with you live on your own data, book a free 30-minute Growth Teardown: Book your teardown.
WHAT NEXT
Want this fixed in your product, not just explained?
Jul 11, 2026
By Victor Teran
Session Replays: How to Find the Screen That Kills Your Retention
A session replay is a recording of a real user moving through your product: every scroll, tap, hesitation, and exit, replayed exactly as it happened.
Analytics
PostHog
UX
It is the closest thing software has to sitting behind the glass in user research, except the users do not know they are being studied and there are thousands of sessions instead of six.
Funnels tell you where users leave. Replays tell you why. The technique is unmoderated usability testing at a scale no research budget buys. At On the User, replays are how we found the onboarding break that Q.AI could not diagnose, the fix that took their activation from 38% to 67% and cut support tickets 73%. This post covers how to use replays without drowning in them: which sessions to watch, what to look for on screen, and how to turn what you see into a fix you can measure.
Key takeaways
Never browse replays randomly. Filter to new users who hit your worst funnel step and left, and watch five of those.
Four signals: hesitation, wandering, rage interactions, and the silent exit. Each points to a different class of fix.
Name the single stumble you saw repeated, change only that, and compare the same funnel step on the next cohort.
Which sessions are worth watching?
Never browse replays randomly. Filter to new users who hit your worst funnel step and left, and watch five of those.
Opened without a question, a replay library is a time sink; teams binge an afternoon of sessions and come away with anecdotes. The discipline is to arrive with a question written down: why do users leave at this specific step?
That means the funnel comes first. Build your signup-to-activation funnel, find the steepest cliff, then filter replays to exactly that population: new users, reached the cliff step, did not survive it, ideally from the last two weeks. In PostHog this filter takes a minute, and it turns the library into a stack of evidence about one question.
Then watch five. Five is enough because product friction is not random: if a screen is broken, it breaks the same way for almost everyone. By the third replay you will start predicting what the user does next. That prediction is the diagnosis forming.
FREE GUIDE
The 4 product leaks costing you growth
A short audit guide for founders. Find the four places your product leaks revenue, and what to fix first.
What should you actually look for on screen?
Four signals: hesitation, wandering, rage interactions, and the silent exit. Each points to a different class of fix.
Hesitation is the long pause on one screen with no input. The user is reading, re-reading, and deciding whether this is worth the effort. Hesitation points to a comprehension problem: unclear value, unclear next step, too much asked too early.
Wandering is movement without progress: opening menus, backing out, visiting settings, returning. The user is looking for something they expected and cannot find. Wandering points to a navigation or expectation problem, often created upstream by marketing promising a feature the first screen does not surface.
Rage interactions are the repeated taps on something unresponsive, or aggressive scrolling. These point to bugs and dead ends, and they are gold because they are unambiguous.
The silent exit is the most important and least dramatic: the user completes a step, lands on the next screen, and simply leaves. No rage, no wandering. That screen failed to give them a reason to continue. Paywalls placed before value produce this signature constantly.
Signal | What you see on screen | What it points to |
|---|---|---|
Hesitation | A long pause on one screen with no input | Comprehension: unclear value, unclear next step |
Wandering | Menus opened, backing out, settings, returning | Navigation or expectation mismatch |
Rage interactions | Repeated taps on something unresponsive | Bugs and dead ends |
Silent exit | Completes a step, lands on the next screen, leaves | That screen gave no reason to continue |
How do you turn observations into a measurable fix?
Name the single stumble you saw repeated, change only that, and compare the same funnel step on the next cohort.
After five replays, write one sentence: "New users stall on the workspace setup screen because it demands a team invite before showing any value." That sentence is the work order. Resist bundling it into a broader redesign; change the one thing you observed.
Then measure like you mean it. Same funnel, same step, next two weeks of new users, before versus after. If the cliff softened, you found the screen. If it did not, you fixed a symptom, and the replays of the new cohort will tell you what you missed.
This loop, funnel to replays to one fix to remeasure, is exactly what we run in the first weeks of every engagement. If you would rather have us run it with you live on your own data, book a free 30-minute Growth Teardown: Book your teardown.
WHAT NEXT
Want this fixed in your product, not just explained?
Jul 11, 2026
By Victor Teran
Session Replays: How to Find the Screen That Kills Your Retention
A session replay is a recording of a real user moving through your product: every scroll, tap, hesitation, and exit, replayed exactly as it happened.
Analytics
PostHog
UX
It is the closest thing software has to sitting behind the glass in user research, except the users do not know they are being studied and there are thousands of sessions instead of six.
Funnels tell you where users leave. Replays tell you why. The technique is unmoderated usability testing at a scale no research budget buys. At On the User, replays are how we found the onboarding break that Q.AI could not diagnose, the fix that took their activation from 38% to 67% and cut support tickets 73%. This post covers how to use replays without drowning in them: which sessions to watch, what to look for on screen, and how to turn what you see into a fix you can measure.
Key takeaways
Never browse replays randomly. Filter to new users who hit your worst funnel step and left, and watch five of those.
Four signals: hesitation, wandering, rage interactions, and the silent exit. Each points to a different class of fix.
Name the single stumble you saw repeated, change only that, and compare the same funnel step on the next cohort.
Which sessions are worth watching?
Never browse replays randomly. Filter to new users who hit your worst funnel step and left, and watch five of those.
Opened without a question, a replay library is a time sink; teams binge an afternoon of sessions and come away with anecdotes. The discipline is to arrive with a question written down: why do users leave at this specific step?
That means the funnel comes first. Build your signup-to-activation funnel, find the steepest cliff, then filter replays to exactly that population: new users, reached the cliff step, did not survive it, ideally from the last two weeks. In PostHog this filter takes a minute, and it turns the library into a stack of evidence about one question.
Then watch five. Five is enough because product friction is not random: if a screen is broken, it breaks the same way for almost everyone. By the third replay you will start predicting what the user does next. That prediction is the diagnosis forming.
FREE GUIDE
The 4 product leaks costing you growth
A short audit guide for founders. Find the four places your product leaks revenue, and what to fix first.
What should you actually look for on screen?
Four signals: hesitation, wandering, rage interactions, and the silent exit. Each points to a different class of fix.
Hesitation is the long pause on one screen with no input. The user is reading, re-reading, and deciding whether this is worth the effort. Hesitation points to a comprehension problem: unclear value, unclear next step, too much asked too early.
Wandering is movement without progress: opening menus, backing out, visiting settings, returning. The user is looking for something they expected and cannot find. Wandering points to a navigation or expectation problem, often created upstream by marketing promising a feature the first screen does not surface.
Rage interactions are the repeated taps on something unresponsive, or aggressive scrolling. These point to bugs and dead ends, and they are gold because they are unambiguous.
The silent exit is the most important and least dramatic: the user completes a step, lands on the next screen, and simply leaves. No rage, no wandering. That screen failed to give them a reason to continue. Paywalls placed before value produce this signature constantly.
Signal | What you see on screen | What it points to |
|---|---|---|
Hesitation | A long pause on one screen with no input | Comprehension: unclear value, unclear next step |
Wandering | Menus opened, backing out, settings, returning | Navigation or expectation mismatch |
Rage interactions | Repeated taps on something unresponsive | Bugs and dead ends |
Silent exit | Completes a step, lands on the next screen, leaves | That screen gave no reason to continue |
How do you turn observations into a measurable fix?
Name the single stumble you saw repeated, change only that, and compare the same funnel step on the next cohort.
After five replays, write one sentence: "New users stall on the workspace setup screen because it demands a team invite before showing any value." That sentence is the work order. Resist bundling it into a broader redesign; change the one thing you observed.
Then measure like you mean it. Same funnel, same step, next two weeks of new users, before versus after. If the cliff softened, you found the screen. If it did not, you fixed a symptom, and the replays of the new cohort will tell you what you missed.
This loop, funnel to replays to one fix to remeasure, is exactly what we run in the first weeks of every engagement. If you would rather have us run it with you live on your own data, book a free 30-minute Growth Teardown: Book your teardown.
WHAT NEXT