01 / UNDERSTAND
What is funnel analysis?
Funnel analysis measures how a defined population progresses through a sequence of events toward an outcome. It shows how many reach each step within an agreed window. A drop off locates a question; it does not explain the cause.
Unlike Customer Journey Map, a funnel does not describe the whole experience or what people thought. It is a deliberately narrow measurement model. A person may leave the measured path and still accomplish their purpose through another channel.
Why use Funnel Analysis?
A broad conversion number hides where progress is lost. Splitting the sequence lets a designer ask a more useful question: did people fail to find the next action, encounter an unavailable option, or decide the offer was unsuitable? Those explanations imply different interventions. Treat the chart as a way to choose where to investigate, rather than a ranking of screens to redesign.
When to use it
USE IT WHEN
A product has an identifiable sequence and a meaningful completion event.
A recent change may have altered progression through one part of a flow.
A team needs to locate a problem before planning qualitative research.
When to choose another method
Use User Interviews when the sequence is based on assumptions about behaviour.
Use Customer Journey Map when channel changes and interrupted progress matter.
Use A/B Testing to evaluate a causal claim about a proposed improvement.
What you’ll need
Agree the event definitions, identity rules, entry period and completion window with an analyst or engineer. Check tracking against actual sessions, including retries, failed requests and consent related gaps. Decide whether you count people, accounts or attempts. Include only the information needed for the question and document any important activity that happens outside the measured system.
What defines a useful conversion funnel
Keep the elements that help answer your question and support a decision.
Who qualifies to enter, and during which period.
Observable actions in a defined order.
How long each entrant has to finish the sequence.
People, accounts or attempts, with retries handled consistently.
The share progressing from one defined step to the next.
Missing events, alternate channels and explanations to investigate.
02 / PUT IT INTO PRACTICE
How to analyse a conversion funnel
Keep evidence and assumptions distinct. Revisit your work when new information changes your understanding.
01
Define completion
Start with an outcome that matters to the person, then work backward to a small number of necessary events.
A repair request accepted by the service is different from an opened form.
02
Set the population
Specify entry eligibility and the unit of counting before looking at results.
Count unique residents who start a request, not every repeated form view.
03
Check instrumentation
Reproduce ordinary and failed paths to see which events actually fire.
Confirm that server rejection cannot trigger the accepted event.
04
Choose order and window
Decide whether intermediate actions are allowed and how much time is available to finish.
A resident returning the next day may still complete one request.
05
Calculate progression
Show counts, entry to finish conversion and adjacent step rates with their denominators.
Sixty completions from one hundred starts means 60% total conversion.
06
Inspect the boundary
Compare a small number of relevant segments and inspect journeys around the largest loss.
Availability messages could explain a drop that looks like a form problem.
07
Choose an investigation
Form a testable explanation and select research or an experiment suited to it.
Observe people interpreting unavailable appointment slots before changing the button.
Example: a repair request funnel
Fictional scenario: 100 residents start a repair request, 80 describe the issue, 64 choose a slot and 60 receive an accepted request. Total conversion is 60%. Description to slot conversion is 64 divided by 80, or 80%; slot to acceptance is 60 divided by 64, or 93.75%. These counts are invented.
The largest absolute loss is before the description is complete. The team nevertheless checks both this boundary and the slot boundary: a smaller loss may have a more serious consequence. Session inspection would ask whether residents can describe unusual faults and whether available slots fit their circumstances.
The visual keeps counts and denominators beside four descending blocks. An annotation links each loss to an unanswered question. No abandoned request is automatically labelled a usability failure. Residents who telephone successfully belong in a separate channel account, not an invisible failure bucket.
03 / BELOW THE SURFACE
Below the surface: the best funnel can exclude the people who need help
A cleaner path can improve the measured rate while reducing access to the service.
ENTRY
Who never enters this funnel?
Check eligibility and alternative channels.
CAUSE
What else could produce this loss?
Separate technical, practical and interface explanations.
GUARDRAIL
Could the rate rise while fewer people are helped?
Monitor the consequence outside the sequence.
Imagine hiding a difficult repair category so fewer people start a request they cannot finish. Entry to completion conversion could rise even though the service now helps fewer residents. A measurement model that begins after eligibility screening would make that exclusion particularly hard to see.
Review who disappears before the first measured event and what happens to them. Pair the funnel with counts of denied entry, alternate channel completion and unresolved need where these can be collected responsibly. Improving a conditional rate is different from improving access.
This is an editorial interpretation check, not a new analytics framework. Before recommending a change, describe both the measured gain and a plausible way it could be produced without helping people. That counterexample gives the team a concrete guardrail to monitor.
Funnel Analysis companion worksheet
The original two page companion records event definitions, counts, time rules and tracking limitations. Its review page turns a loss of progress into an investigation with an owner. Use a query or analytics tool for computation; the worksheet preserves the assumptions behind the chart.
Practical tips
Keep counts visible alongside percentages.
Check the meaning of the final event with operations.
Record one plausible alternative explanation before recommending a fix.
Common mistakes
Counting page views as distinct people.
Changing the completion window between comparisons.
Treating the biggest drop as proof that its screen is badly designed.
What you’ll get
A reproducible funnel and a prioritised question about lost progress. Carry that question into observation or experimentation, then check both conversion and the service consequence of the change.