A cell is the summary of the rows matching both labels

Reading a cell aloud

Switch the two-way pivot to Average of amount:

prov    Online   Store   Grand Total
AB      1320.00  980.00      1150.00
BC       360.00  640.00       453.33
ON       465.00  532.50       498.75
QC        90.00  197.50       161.67
Total    510.00  513.33       511.67

Every cell has the same sentence behind it: the [function] of [Values field] over the source rows whose [Rows field] equals this row label AND whose [Columns field] equals this column label. ON × Store is the average amount over the rows where prov = ON and channel = Store — orders 1003 (760) and 1009 (305) — (760 + 305) / 2 = 532.5. The and matters: a cell never means ON or Store.

The margins drop one condition. The ON grand total, 498.75, is the average over all four ON rows regardless of channel — AVERAGEIF by province. The Online column total, 510, is the average over the six online rows regardless of province. And the corner, 511.67, is the average over all twelve rows: 6140 / 12.

Worked example — the average of averages

A report averages the four province figures in the Grand Total column: (1150 + 453.33 + 498.75 + 161.67) / 4 = 565.94. That is not 511.67, and the difference is not rounding. The four province averages rest on different numbers of orders — AB's 1150 is two orders, ON's 498.75 is four — and a mean of means gives each province one vote regardless. The corner cell weights every order equally, which is what "the average order" means. In R:

> pm <- aggregate(amount ~ prov, data = orders, FUN = mean)
> mean(pm$amount)
[1] 565.9375
> mean(orders$amount)
[1] 511.6667

The pivot's Grand Total is always computed from the source rows, never from the cells above it. That is also why a pivot of Averages has a Grand Total that is not the average of its row totals, and why a Sum pivot's corner is the sum of its row totals — adding is the one summary where the two routes agree.

Fair comparison

The same trap runs the other way. From the Sum pivot, AB is our best province: 2300 against ON's 1995. But AB's total is two orders and ON's is four; per order, AB averages 1150 and ON 498.75 — AB really is larger per order — while per store, per rep or per capita the ranking may flip again. A total across groups of unequal size is not a fair comparison; the fair basis is a rate — per order, per store, per resident — and the question decides which. The pivot gives both: Sum for the total, Average or Count for the basis, and the analyst's job is to say which number the claim needs. "Alberta had the highest total sales" is true; "Alberta is our strongest market" needs a denominator.

Read two cells, then choose the basis

The numeric items ask for the ON × Store average and for the unweighted mean of the four province averages — with the true overall mean stated in the explanation. The questions turn on and versus or, weighted versus unweighted, and totals versus rates.

CRISP-DM: reading the pivot is data understanding → explore data; deciding whether a comparison is fair is evaluation → evaluate results.