App Onboarding
Part of Mobile app onboarding
Measuring where new users leave onboarding
Define new-user cohorts, check onboarding events and interpret funnel gaps without mistaking every missing step for abandonment.
To find where new users stop progressing through onboarding, define the eligible group, record the task steps they can take and compare each step with the next under stated time rules. A funnel can locate a measured gap. It cannot tell you why people left or whether every person needed that step.
Define the journey
Start with a completed first-use outcome and work backwards through the necessary actions. In a hypothetical booking app, one route might be first open, availability viewed, time selected and booking confirmed. A call to book or a saved time for later is a different route; forcing it into that sequence could mislabel valid behaviour as abandonment.
Specify what triggers each event and which completed state it represents. State who is eligible, its timestamp and how repeats are handled.
A screen view shows that a screen appeared. A button tap shows an attempt. Neither proves a booking was confirmed. Firebase Analytics logs some events automatically, but a task-specific completion event may need implementation.
Choose a starting event that matches your cohort definition. Decide how reinstalls and people using more than one device fit the cohort before using that event as a starting count.
Check event quality
On development devices, compare the expected route with the recorded event sequence. Firebase DebugView shows development-device events near real time. Seeing an event there helps inspect instrumentation; it does not establish that production data is complete. Include interrupted tasks, errors, restarts and supported app versions in the check plan.
| Transition | Eligible starting group | Next completed state | Check before calling it a drop-off |
|---|---|---|---|
| First open to task start | New users able to use the task | Task genuinely begun | Could another valid entry route bypass this event? |
| Task start to required setup | People who need that setup | Setup successfully saved | Was it optional or completed earlier? |
| Required setup to first value | People able to finish the task | Confirmed useful result | Could the result arrive after the report window? |
These are example event definitions. They are not automatic analytics events.
Choose a denominator and window
For each transition, count eligible people who reached the preceding state. Divide the number of those same people who reached the next completed state within the chosen window by that preceding count. Show both counts with the percentage. Keep cohort, identity, event and window rules consistent when comparing versions or changes.
Allow enough time for the task to finish. Someone who begins today and completes tomorrow should not be counted as lost at the end of today if the intended window is longer. Exclude incomplete observation windows from a finished-window comparison or report them separately. Investigate ineligible people, valid alternative routes and missing events instead of assuming every absence means abandonment.
Funnel reports can apply configured entry and step-order rules that differ from a custom transition rate. Record the chosen settings and do not treat a displayed rate as the custom transition rate above unless its population and step rules match.
Key Metrics in Onboarding Funnel Analysis
- Denominator
- Number of eligible users who reached the preceding step
- Numerator
- Number of users who completed the next step within the window
- Conversion rate
- Percentage of users progressing from one step to the next
Investigate a gap
Check event quality and the effect on the useful outcome before prioritising a gap. Review the screen, error states and relevant feedback. A confusing field, late eligibility check and permission interruption call for different changes.
Record one proposed change and compare later groups under compatible definitions. If progress through a step rises but confirmed outcomes do not, check whether that step became easier to trigger without helping people complete the task. A measured gap identifies where to investigate; it is not a diagnosis.
