Level 5 · Two AOVs from the same orders

Two ways to average a ratio

AOV is a ratio: net sales divided by orders. There are two ways to get “an average AOV”:

• Ratio of totals: add all net sales, add all orders, divide once. Every order counts equally.
• Average of ratios: compute the ratio per group (per customer, per day, per channel), then average those numbers. Every group counts equally, however many orders it has.

net sales / orders (2025)              535.78
mean of 1,466 per-customer AOVs        467.56

Where the $68 goes

The groups are not the same size. In 2025, 254 business customers placed 1,165 orders averaging $1,060.15, while 1,212 consumer customers placed 3,244 orders averaging $347.46. In the ratio of totals, business orders are a quarter of the orders. In the average of ratios, business customers are a sixth of the votes. The big orders lose weight, so the number falls.

Which one is “the” AOV

The KPI sheet needs a number that adds up: AOV × orders = net sales. Only the ratio of totals does that. The per-customer average is not wrong; it answers “what does a typical customer’s order look like?”, which is a fair question with a different name. Two numbers with one label is the bug.

The same trap everywhere

Any rate averaged over groups of different sizes does this:

• Monthly conversion rates. The average of 2025’s twelve monthly conversion rates is 2.93%; total conversions / total sessions is 3.08%. November and December carry far more sessions, and convert better, so equal month weights pull the average down.
• Store or region rates. A tiny store with a 40% rate counts as much as the flagship.
• Averages of averages in BI tools. A grand total row that averages the rows above it is an average of ratios. Check how the tool aggregates before you trust the total.

Common mistakes

• Averaging a column of percentages. Re-derive the rate from summed numerators and denominators.
• Assuming a gap means a bug. Both queries can be correct; they weight differently.
• Mixing them over time. A dashboard that switches method makes a trend out of nothing.