An overall average combines subgroup results with the weight of each subgroup. Calculate both components before explaining a movement. A changing mix can move the total even when every subgroup is stable; different subgroup movements can also cancel each other out.
Identify the weights hidden in the overall result
Start with the unit that enters the calculation. A conversion rate uses eligible visits or people as its denominator, while average order value weights orders. The same channel can therefore have different importance in different measures. Reconstruct the total from subgroup numerators and denominators, rather than averaging displayed percentages. Check that each observation belongs to one subgroup or document how overlapping membership is handled.
Choose segments tied to the business question rather than every available label. Acquisition channel may explain a campaign result; service type may explain handling time. Preserve an unknown category when classification is missing. Removing it changes the population and can make the remaining segments appear complete. Ensure a category means the same thing in both periods, especially after a tracking change or an organisational restructure.
Separate changes within groups from changes in composition
Display each group's count, rate and share of the denominator for both periods. A stable subgroup rate with a changing share suggests composition matters. A changing rate with a stable share points toward a within-group movement, though measurement and selection can still explain it. If both move, show both. A single headline such as channel performance deteriorated is too broad to describe all these possibilities.
One descriptive comparison holds the earlier mix fixed and applies the later subgroup rates. Explain which period supplies the weights and why that reference is useful. This standardisation is a calculation under an explicit convention, not the performance that would necessarily occur if the business changed its mix. Customer behaviour, capacity and channel quality might respond to the change. Avoid presenting the arithmetic exercise as a causal forecast.
Carry the finding into a specific operational question
A mix explanation does not mean there is nothing to investigate. A lower-converting channel may bring customers with different needs, lower acquisition costs or longer decision cycles. Check the intended outcome before recommending that it be reduced. Conversion, retention and contribution answer different questions. Review the maturity of each cohort so newer customers are not penalised simply because their observation window is shorter.
Publish enough detail for the aggregate to be reconstructed without exposing unnecessary individual records. For sparse segments, combine categories only when the combination remains meaningful and explain any suppression. Decide in advance which differences deserve investigation. The next task may be validating attribution, reviewing audience fit or waiting for outcomes to mature. Assign that task explicitly rather than letting the overall average become a verdict on a team. Keep the original channel assignment rules available so that a later attribution correction can be separated from a real mix movement.
A fictional shift between two channels
A fictional scenario to explain the method, not a customer result or a market benchmark.
A fictional subscription service has 200 eligible visits from referrals and 200 from paid search in April. Referrals generate 40 signups and search generates 10, giving rates of 20% and 5%. The total is 50 signups from 400 visits, or 12.5%.
In May, 100 referral visits yield 20 signups and 300 search visits yield 15. The denominator remains 400 visits, but the referral share falls from one half to one quarter. Referrals still convert at 20% and search at 5%. The resulting 35 of 400 is 8.75%, with both channel rates unchanged. The report shows both channel denominators, labels the numbers fictional and attributes the aggregate difference descriptively to the changed weights. It does not claim that buying more referral traffic would preserve the same rate at a larger scale. The next review checks whether the mix shift was intended and whether the new customers meet the service eligibility definition.
TAKE THIS INTO YOUR NEXT CONVERSATION
A practical next-step checklist.
- Rebuild the overall rate from subgroup counts.
- Show subgroup shares alongside subgroup rates.
- Retain unknown classifications in the reconciliation.
- Label the weighting convention used for comparison.
Common questions.
Should I average the two channel percentages?
Only equal denominator weights justify that simple average. Otherwise sum successes and eligible observations first.
Does standardising the mix predict the effect of a channel change?
No. It holds selected quantities fixed for comparison and does not establish how customers would respond.
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