Funnel analysis
A funnel is a sequence of steps toward a goal (visit, signup, activate,
purchase), and users leak out at every step. Funnel analysis finds where
the leaks are, how big they are, and which one is worth fixing, turning a
vague "conversion is low" into a specific, sized problem.
Method
- Define the funnel steps precisely. Name each step as an observable
event with a clear definition and order, and decide the window (does a
user have to complete the funnel in one session, a day, ever?). Fuzzy
step definitions produce fuzzy, unactionable numbers (see product-metrics'
define-once rule).
- Measure conversion at each step, not just end to end. The overall
rate hides where the loss happens; the per-step drop-off is the map. A
step where 60% leak is the finding, and it is invisible in the aggregate.
Show both the step-to-step rate and the cumulative.
- Size the opportunity of each leak. The step with the worst drop-off
is not always the best to fix; weight by volume and value. Fixing a 50%
leak that only 100 users hit matters less than a 10% leak that 100,000
hit. Prioritize by users-lost times value-per-user, not by the scariest
percentage (see prioritize-tasks).
- Segment the funnel to find who leaks. The same step may convert fine
for one segment and terribly for another (mobile vs desktop, new vs
returning, by source, by geography). Segmenting reveals whether a leak is
universal (fix the step) or specific (fix it for that segment), and a
bad-fit traffic source shows up as step-one drop-off (see
user-activation, correlation-causation on selection).
- Diagnose the why, do not just measure the where. A drop-off number
says where, not why. Combine it with qualitative evidence (session
recordings, user testing, support tickets) to understand the cause
before designing a fix (see usability-testing, empty-and-error-states).
Guessing at the cause of a leak wastes the fix.
- Mind the funnel's honesty. Watch for survivorship and selection (the
users who reach step 4 are not like those who dropped at step 1, so
later-step rates are conditional), and time effects (a "funnel" measured
in one session undercounts users who complete it later). State the
window and the conditioning.
Boundaries
- Funnel analysis shows where users leave, not why; pair the where (data)
with the why (qualitative and experiments) before investing in a fix
(see usability-testing, experiment-analysis).
- Not every flow is a linear funnel; users loop, skip, and return, and
forcing a messy journey into rigid steps can mislead. Model the real
paths where they matter (see user-flows).
- Optimizing a funnel step can shift the problem downstream (more signups
of worse-fit users who churn); check the downstream metrics, not just the
step you fixed (see churn-analysis, product-metrics' guardrails).
1---2name: funnel-analysis3description: Analyze conversion funnels to find where users drop off and why, and size the opportunity of fixing each step. Use when diagnosing where a multi-step flow loses people, or prioritizing what to fix.4---56# Funnel analysis78A funnel is a sequence of steps toward a goal (visit, signup, activate,9purchase), and users leak out at every step. Funnel analysis finds where10the leaks are, how big they are, and which one is worth fixing, turning a11vague "conversion is low" into a specific, sized problem.1213## Method14151. **Define the funnel steps precisely.** Name each step as an observable16 event with a clear definition and order, and decide the window (does a17 user have to complete the funnel in one session, a day, ever?). Fuzzy18 step definitions produce fuzzy, unactionable numbers (see product-metrics'19 define-once rule).202. **Measure conversion at each step, not just end to end.** The overall21 rate hides where the loss happens; the per-step drop-off is the map. A22 step where 60% leak is the finding, and it is invisible in the aggregate.23 Show both the step-to-step rate and the cumulative.243. **Size the opportunity of each leak.** The step with the worst drop-off25 is not always the best to fix; weight by volume and value. Fixing a 50%26 leak that only 100 users hit matters less than a 10% leak that 100,00027 hit. Prioritize by users-lost times value-per-user, not by the scariest28 percentage (see prioritize-tasks).294. **Segment the funnel to find who leaks.** The same step may convert fine30 for one segment and terribly for another (mobile vs desktop, new vs31 returning, by source, by geography). Segmenting reveals whether a leak is32 universal (fix the step) or specific (fix it for that segment), and a33 bad-fit traffic source shows up as step-one drop-off (see34 user-activation, correlation-causation on selection).355. **Diagnose the why, do not just measure the where.** A drop-off number36 says where, not why. Combine it with qualitative evidence (session37 recordings, user testing, support tickets) to understand the cause38 before designing a fix (see usability-testing, empty-and-error-states).39 Guessing at the cause of a leak wastes the fix.406. **Mind the funnel's honesty.** Watch for survivorship and selection (the41 users who reach step 4 are not like those who dropped at step 1, so42 later-step rates are conditional), and time effects (a "funnel" measured43 in one session undercounts users who complete it later). State the44 window and the conditioning.4546## Boundaries4748- Funnel analysis shows where users leave, not why; pair the where (data)49 with the why (qualitative and experiments) before investing in a fix50 (see usability-testing, experiment-analysis).51- Not every flow is a linear funnel; users loop, skip, and return, and52 forcing a messy journey into rigid steps can mislead. Model the real53 paths where they matter (see user-flows).54- Optimizing a funnel step can shift the problem downstream (more signups55 of worse-fit users who churn); check the downstream metrics, not just the56 step you fixed (see churn-analysis, product-metrics' guardrails).
Amey-Thakur