Skip to content
Back to Analytics & Experimentation

Conversion Rate

The fundamental funnel metric —the fraction of users who complete a step —and the four common definitions that produce different numbers for the same funnel.

Why

Conversion rate is the most-reported and most-misdefined metric in analytics. The same funnel can have four different conversion rates depending on the definition: overall (step N / step 1), step-to-step (step N / step N-1), cohort (completed within window / entered cohort), and unique (unique users / unique exposures). These four numbers can differ by 2x for the same funnel, and when the team reports 'conversion is 26%' without specifying which definition, every dashboard shows a different number and every meeting has a 'which 26%?' argument. The discipline is to pick one definition per funnel (usually overall, for executive reporting), document it, and use the others (step-to-step for diagnosis, cohort for time-series) as secondary views. The definition is not a detail —it is the metric, and getting it wrong makes every downstream analysis (A/B testing, forecasting, benchmarking) wrong.

When

When defining any funnel. The definition is chosen once, documented, and used consistently for all reporting on that funnel.

How

1. Choose the primary definition for the funnel:
   - Overall: step N completions / step 1 entries (e.g., 3224/12400 = 26.0%) —best for executive reporting, one number
   - Step-to-step: step N / step N-1 (e.g., 3224/6448 = 50.0%) —best for diagnosing which step drops
   - Cohort: completed within window / entered cohort —best for time-series (compare Sep vs Aug)
   - Unique: unique users completing / unique users entering —best when users can enter multiple times
2. Document the chosen definition in the metrics dictionary with the SQL
3. Compute and report the primary definition in the dashboard; show the others as secondary views
4. For A/B tests on the funnel, use the overall conversion as the primary metric (it captures the end-to-end effect) and step-to-step as a diagnostic
5. For time-series (is conversion improving?), use the cohort definition with a fixed window (e.g., 7-day conversion) so the number is comparable across time
6. For funnels where users can enter multiple times (e.g., checkout), use unique conversion to avoid counting the same user's 3 attempts as 3 entries
7. Add a unit test that asserts the conversion rate on a fixture, with the definition documented in the test
8. When reporting conversion to stakeholders, always state the definition: "overall funnel conversion (deploy / signup) = 26%" —the definition is part of the number

Example

# Funnel: signup -> first deploy (4 steps, n=12,400 entered)
Step 1: 12,400  Step 2: 9,920  Step 3: 6,448  Step 4: 3,224

Four conversion rates for the SAME funnel:
  Overall (step4/step1):     3,224 / 12,400 = 26.0%
  Step-to-step (step4/step3): 3,224 / 6,448 = 50.0%
  Cohort (7d window):        3,180 / 12,400 = 25.6%  (44 deployed after 7d)
  Unique (unique users):     3,224 / 12,400 = 26.0%  (no duplicate entries here)

Primary definition (documented): Overall = 26.0%
  -> used in executive dashboard, A/B test primary metric, quarterly target

Secondary views:
  Step-to-step: [80%, 65%, 50%] -> diagnostic: step 3->4 is the bottleneck
  Cohort (7d): 25.6% -> time-series: comparable across months
  Cohort (14d): 26.0% -> the 44 late deployers are within 14d

A/B test on this funnel:
  Primary metric: overall conversion (captures end-to-end effect)
  Diagnostic: step-to-step (if overall moves, which step caused it?)

Reported as: "Overall funnel conversion (deploy/signup) = 26.0%
  (step-to-step: 80% -> 65% -> 50%; 7d cohort: 25.6%)"

Common Problems

  • ⚠Reporting 'conversion is 26%' without specifying the definition —the team uses overall (26%), the dashboard shows step-to-step (50%), and the A/B test uses cohort (25.6%); every number is 'conversion' and none of them agree.
  • ⚠Switching definitions mid-analysis —the team reports overall conversion in Jan (26%), switches to step-to-step in Feb (50%), and declares 'conversion doubled' when only the definition changed.
  • ⚠Using overall conversion for step diagnosis —'conversion is 26%' does not tell you which step to fix; step-to-step (80%, 65%, 50%) does, and using the wrong definition for the wrong purpose loses the diagnostic value.
  • ⚠Not deduplicating in funnels where users enter multiple times —a user who attempts checkout 3 times counts as 3 entries and 1 completion, deflating the conversion rate and making the funnel look worse than it is.

Best Practices

  • ✓Pick one primary definition per funnel (usually overall for reporting, cohort for time-series) and document it —the documentation is what prevents the 'which 26%?' argument and keeps the dashboard consistent.
  • ✓Always state the definition when reporting the number —'overall funnel conversion (deploy/signup) = 26%' is unambiguous; 'conversion = 26%' is not, and the ambiguity will cause a mismatch eventually.
  • ✓Use step-to-step conversion as a diagnostic view, not the primary metric —it tells you which step to fix, but it is not the end-to-end number the business cares about, and reporting it as 'conversion' misleads stakeholders.
  • ✓Deduplicate when users can enter multiple times —count unique users, not sessions, so a user's 3 checkout attempts are 1 entry with 1 completion (or 0 if they never complete), not 3 entries with 1 completion.