Funnels
Locate conversion loss without assuming the largest drop is the best opportunity. · Data & Experimentation · Lesson 33 · 3 min
Funnels · 3 min
Situation
Signups are healthy; repeat use is weak.
A product gets 10,000 visits, 1,000 signups, 600 completed onboardings, 300 first actions, and 120 users returning the following week.
These illustrative counts describe a path. They do not yet explain why people leave or whether every visitor was a relevant potential user.
Mental model
A funnel needs a defined population and window.
Specify the ordered steps, which people are eligible, whether steps must occur in sequence, and how long users have to progress.
Do not divide this week's returns by this week's signups if returns belong to older cohorts. Each step should follow the same defined group or use an explicitly different analysis.
Example
Read the rates and the volumes.
In the example, onboarding-to-first-action conversion is 300 divided by 600, or 50%. First-action-to-return is 120 divided by 300, or 40%.
Improving either step could matter, but the causes and costs differ. Low-intent traffic may explain a large visit-to-signup drop that a product redesign cannot fix economically.
Investigate
Segment before choosing a fix.
Break the affected step by platform, acquisition source, customer type, or other relevant characteristics. Inspect sessions or interview people who stopped there.
Check whether the event fires reliably. A missing event can manufacture a funnel drop, while duplicate events can inflate progress.
Failure case
Counting easy actions as meaningful progress.
Removing required setup might raise onboarding completion while leaving users unable to perform the first useful action. The funnel looks healthier at one step while downstream value deteriorates.
Evaluate the complete path. Sometimes adding helpful friction, such as selecting the right workspace, improves later success.
Remember this
A funnel tells you where to investigate.
Combine step-level conversion, population size, user intent, and evidence about the obstacle. Choose the intervention that improves a valuable outcome, not merely one screen's rate.