A weighted average multiplies each value by its appropriate weight, sums those products and divides by total weight. For a pooled order-value average, order counts are the weights. Publish the weights and population; averaging branch averages equally answers a different question about branches.
Identify the observation the average should describe
Ask whether the target is an average order, average customer, average branch or average day. Each implies a different unit and potential weight. An average order value should give each eligible order one contribution, regardless of branch. An average branch performance measure may deliberately give each branch equal influence. State that choice before applying a formula, because both calculations can be mathematically correct while supporting different interpretations.
Verify that the component values use compatible definitions. One branch cannot include taxes and refunds while another excludes them without qualification. Use the same reporting period, currency basis and eligibility rules. If the input average has been rounded, request underlying totals where possible. Multiplying rounded averages by counts can introduce a small reconciliation difference that should not be mistaken for newly discovered sales.
Show the weights and their denominator
For values x and non-negative weights w, the weighted average is the sum of w times x divided by the sum of w. Explain what one unit of weight represents. Counts of orders, customer exposures, staff hours and sampling weights are not interchangeable. A missing weight is not automatically one or zero. Resolve its meaning or present the affected population as unmeasured.
Where underlying totals and counts exist, sum those directly. Pooled order value equals total eligible order amount divided by total eligible orders. This avoids unnecessary rounding and makes reconciliation straightforward. Exclude a group only under a stated rule and disclose how much population was removed. A branch with no eligible orders contributes zero orders to the denominator; its average is undefined rather than a measured zero.
Explain composition and uncertainty
A pooled average can move even when every branch average stays constant, because the order mix shifts between branches. Show group counts alongside group averages when that composition matters. Do not infer that one branch improved or declined from the overall movement alone. A decomposition or stable-mix comparison can help, provided it is labelled as an additional analytical view rather than the observed pooled result.
NIST's location material supplies background on means; the grouped weighting example here is original. Descriptive weights do not automatically establish representative sampling or statistical precision. If the metric is intended to estimate a wider population, examine the sampling design separately. A transparent weighted calculation should let a reader identify the observation unit, reconstruct the denominator and understand which missing groups could alter the interpretation.
Do not carry forward an old group average when its current count is missing. That combines values and weights from different populations. Show the group as unavailable or use a clearly labelled scenario with the assumed weight and period stated.
Fictional branches: a small premium branch has less volume
A fictional scenario to explain the method, not a customer result or a market benchmark.
A fictional branch A records ten eligible orders totalling 1,000 units, for an average of 100. Branch B records ninety orders totalling 4,500, for an average of 50. The simple mean of the two branch averages is 75. That describes equally weighted branches, not the average order across the business.
The pooled order average is 5,500 divided by 100 orders, or 55. Equivalently, ten times 100 plus ninety times 50, divided by ten plus ninety, gives 55. The report shows both branch counts so the larger influence of branch B is understandable rather than hidden in a formula.
If the next period has more orders at A, the pooled average can rise even if both branch averages remain unchanged. These invented values demonstrate a composition effect and appropriate order weights. They do not claim that either branch is more efficient or that the order population represents all potential demand.
TAKE THIS INTO YOUR NEXT CONVERSATION
A practical next-step checklist.
- Name the observation unit being averaged.
- Check component definitions and periods match.
- Publish weights and total eligible denominator.
- Prefer underlying totals over rounded averages.
Common questions.
Should every branch average receive equal weight?
Only when the intended measure is the average across branches. Pooled order value needs order-count weights.
What is the average for a branch with no orders?
It is undefined under this definition. Zero orders can contribute to counts without inventing a zero average.
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