Combine compatible rates by summing their underlying numerators and denominators, then dividing. Do not average percentages equally unless equal group influence is the intended measure. Verify eligibility and counting rules, disclose zero denominators, and show group populations when composition affects the pooled result.
Check that each rate means the same thing
A ratio is a numerator divided by a denominator, and its label should describe both. Before pooling branch conversion rates, confirm that converted and eligible enquiries follow the same definitions, outcome window and identity unit. One branch counting people and another counting opportunities cannot produce a coherent shared rate simply because both outputs use a percentage sign. Align the concepts or keep the rates separate.
Ensure numerator events belong to their denominator population. Purchases this month divided by unrelated enquiries this month may be an activity ratio, not cohort conversion. A numerator should not exceed its eligible population under a one-conversion-per-member definition. If repeated events are intended, call the measure an event frequency or another appropriate name rather than assuming it must behave like a probability.
Pool counts before calculating the percentage
For compatible groups, add converted counts and add eligible counts, then divide the combined totals. This weights each group rate by its denominator. Keep the underlying integers or appropriately precise quantities until the final display step. Averaging rounded percentages can create errors even when the intended weights are later supplied. Where only percentages are available, request denominators before claiming a pooled result.
A group with zero eligible cases has an undefined rate and contributes no eligible cases to a pooled calculation. It should not enter as a zero-percent group in a simple average. Missing eligibility is different: it means the denominator is unavailable, so excluding that group may leave an incomplete population. Report the distinction rather than allowing software defaults to turn unknowns into measured zero performance.
Interpret the combined rate alongside its groups
A pooled rate is useful for the observed overall population, but changing mix can move it even when group-specific rates do not improve. Show group counts and rates where managers need to understand performance. For a like-for-like comparison, a separately labelled standardised mix may be useful, but it represents an analytical scenario. Preserve the actual pooled result so the report remains tied to observed activity.
NIST's proportion material supplies statistical background; this article's arithmetic is an original descriptive example, not a hypothesis test. A larger rate is not automatically a significant improvement or evidence of causation. Keep sample size, eligibility coverage and observation maturity visible, and escalate inconsistent group definitions before spending effort on a highly precise but conceptually invalid aggregate.
Check that groups do not overlap before summing their denominators. A customer assigned to two regions can enter both branch reports legitimately, but a combined unique-customer conversion rate requires deduplication at the shared identity level rather than simple addition.
Fictional branches: unequal enquiry populations
A fictional scenario to explain the method, not a customer result or a market benchmark.
A fictional branch A converts 18 of 20 eligible enquiries, giving 90%. Branch B converts 20 of 100, giving 20%. The unweighted average of branch percentages is 55%, but the combined population has 38 conversions among 120 enquiries. Its pooled conversion rate is approximately 31.7%.
The difference arises because branch B contributes five times as many eligible enquiries. A report showing only 55% would give the small branch accidental equal influence. The analyst publishes the combined counts and both branch rates, with the same thirty-day outcome window applied to each. A third branch with no eligible enquiries is labelled not applicable rather than zero percent.
If branch A later receives a larger share of enquiries, the pooled rate can rise without either branch improving. These invented counts illustrate denominator weighting and composition. They do not establish statistical significance, comparable customer difficulty or the reason the branches have different outcomes.
TAKE THIS INTO YOUR NEXT CONVERSATION
A practical next-step checklist.
- Align numerator, denominator and outcome windows.
- Sum compatible underlying counts first.
- Distinguish zero eligibility from missing coverage.
- Show group rates and composition where useful.
Common questions.
Can I average two percentages when denominators are unknown?
That yields an equal-group average, not a pooled rate. Obtain the denominators before claiming overall conversion.
Why can the overall rate rise with unchanged branch rates?
The population can shift toward a branch with a higher rate. That is a composition effect.
Further reading & source context.
These primary references provide background. They are not evidence for the fictional example above and do not endorse Accuralytix.
Found an error or a missing qualification? Send an editorial correction.
Check my data readiness.
Use a practical tool to turn this idea into a clearer starting point.
Check my data readiness