Can You Average Conversion Rates? A US and UK Reporting Example
Calculate the overall rate from counts, and explain when an equal-market average answers a different question.
How do you combine conversion rates correctly?
For disjoint groups using the same conversion definition and reporting period, divide total conversions by the total eligible denominator. This is equivalent to weighting each group’s rate by its denominator. A simple average of group percentages generally does not give the overall conversion rate when group sizes differ.
A regional report with two different answers
In this synthetic example, a US site records 1,000 eligible visits and 40 converting visits: 4%. A UK site records 100 eligible visits and 10 converting visits: 10%. Each visit belongs to exactly one region and can count as converting at most once. Both reports use the same complete reporting interval and event definition.
Averaging 4% and 10% produces 7%. But there are 50 converting visits among 1,100 visits, so the combined rate is approximately 4.55%. The US contributes ten times as many visits, so its rate receives ten times the weight. Labeling 7% as the overall rate would imply 77 converting visits, whereas the data contains only 50.
Verify the calculation in Python
This standard-library script needs no packages. The final assertion checks that splitting the US row into two identical-sized subgroups leaves the combined rate unchanged.
from math import isclose
rows = [("US", 1000, 40), ("UK", 100, 10)]
visits = sum(r[1] for r in rows)
conversions = sum(r[2] for r in rows)
pooled = conversions / visits
unweighted = sum(c/n for _, n, c in rows) / len(rows)
assert isclose(pooled, 50/1100)
assert isclose(unweighted, 0.07)
split = [("US-a", 500, 20), ("US-b", 500, 20), ("UK", 100, 10)]
assert isclose(sum(c for _,n,c in split)/sum(n for _,n,c in split), pooled)
print(f"Overall: {pooled:.2%}; equal-market average: {unweighted:.2%}")
Expected output: Overall: 4.55%; equal-market average: 7.00%. Keep the full precision for subsequent calculations and round only for display.
When is the simple average useful?
An equal-market average can be useful if the explicit question is “What is the average regional rate when every region receives equal importance?” That is a different quantity from the share of all visits that convert. State the weighting rule and give it a clear label rather than presenting one number as interchangeable with the other.
In our example, splitting the US into two reporting rows would change the simple three-row average to 6%, even though no customer behavior changed. A metric should not change merely because a report designer subdivided a region unless the metric intentionally treats each reporting row as a separate unit.
What should you check before adding denominators?
- Overlapping users: monthly distinct users from two sites cannot automatically be added if a person can use both. Rebuild the combined distinct population before calculating a user-based rate.
- Different units: a purchase-per-session rate and a purchaser-per-user rate are not the same metric. Do not pool them.
- Incomplete periods: do not combine a complete US week with a partial UK week and describe the result as a comparable weekly rate.
- Missing data: an unavailable regional denominator is not zero. Report incomplete coverage instead of silently omitting it.
- Zero denominators: a group with no eligible visits has an undefined individual rate, not a measured 0% conversion rate. If the total denominator is zero, the overall rate is undefined too.
Which weights should I use?
Use the denominator of the metric being combined. For converting visits divided by visits, weight by eligible visits. For total order value divided by order count, weight regional average order values by order counts, provided values use a consistent currency and revenue definition. Do not use revenue weights simply because the dashboard is a sales dashboard.
If you already have exact counts, summing numerators and denominators avoids reconstructing totals from rounded percentages. A report that shows only 4% and 10%, without counts, does not contain enough information to recover the overall rate.
The NumPy weighted-average reference documents the weighted-sum divided by sum-of-weights calculation. The example above implements the same arithmetic directly; it does not depend on NumPy.
How should you explain the result in an interview?
“I would ask whether these are disjoint visits with a common conversion definition. If so, I would calculate 50 divided by 1,100, about 4.55%. The 7% figure is an equal-market average. I would show the regional counts beside the rates so the weighting is visible, and avoid rounding before combining them.”
This exercise concerns combining rates in one period. For a separate question about changes over time, work through conversion falling while every channel improves. Continue with business-case interview practice and choosing between mean and median. These are fictional learning examples, not measured regional benchmarks or employer questions.