Metrics everyone agrees on
Every Monday, Mangrove's leadership sees three different "sales" numbers, and each one is correct for the team that made it. This lesson shows you how to take a metric apart, explain the gap between two numbers with a bridge, agree one definition with the people who own it, handle measures that can't be added up, and write each definition once so every dashboard gives the same answer.
By the end you can
- Break a metric into what is counted, which price, which date, which scope and which records, and build a bridge that explains why two numbers differ
- Write agreed definitions of net sales and out-of-stock loss, with an owner and caveats
- Tell additive, semi-additive and non-additive measures apart, and compute every ratio as a ratio of sums
- Write a metric dictionary entry and the general shape of a metric in a semantic layer
- Set up certification and change control so a definition can't drift without anyone noticing
The everyday picture
Picture a busy food market. A rice trader says a sack weighs 52 kilos. The buyer weighs it at home and gets 50. The truck driver, who is paid by weight, wrote 53 on his delivery slip. Nobody is lying. The trader weighed the sack with its bag, the buyer weighed only the rice, and the driver weighed it on the truck with the pallet. Packaged food solves this with two words on the label: gross weight (with packaging) and net weight (the food alone).
Markets solved the other half of the problem long ago. The town keeps one official set of weights. An inspector checks each trader's scale against them and puts a stamp on it. If the town wants to change the official kilo, it doesn't let each trader adjust his own scale: it announces the change, sets a date, and stamps the scales again.
Business metrics need the same three things: different names for different quantities, one official definition of each, and a stamp that says "this one was checked", with a proper process for changing it.
Why three correct numbers disagree
It is Monday morning, and leadership is looking at the week from Monday 12 to Sunday 18 October 2026. Store operations reports till sales of $31.4 million. The app team reports order value of $7.9 million. Finance reports net sales of $35.2 million. The head of stores adds the first two, gets $39.3 million, and asks why finance is $4.1 million lower. A week of emails follows.
Nobody made a mistake. Each team answers a different question, and each number has a different definition hidden inside it. In Asking a question data can answer you saw that a metric needs an exact definition. For money metrics, five choices make most of the difference:
| Choice | Store ops: till sales | App team: order value | Finance: net sales |
|---|---|---|---|
| Which price | Shelf price of every line rung up; promotions sit on separate discount lines and are not subtracted | Basket value when the order is placed | Price paid, after discounts and returns |
| Tax | Includes VAT | Includes VAT | Excludes VAT |
| Which date | Sale date at the till | Date the order was placed | Posting date in the ledger |
| Scope | 320 stores | App orders | Stores and app |
| Which records | All sales; returns reported elsewhere | All placed orders, including cancelled ones | Completed sales only |
Two terms in that table need a word. VAT is tax collected for the government, so it was never Mangrove's money. The posting date is the date an entry is booked in the accounting ledger. Store takings are posted the morning after, so Sunday's sales land in finance's next week.
The way out of the argument is a bridge (accountants also call it a reconciliation). You start from one number, list every difference with its amount, and end exactly at the other number. If the bridge closes, both numbers are right and everyone can see why they differ. If a gap is left over, that gap is something nobody understands yet, and it is often a bug: a test store still sending data, a filter missing, a join that doubles rows.
Quick check
You build a bridge from till sales to finance's net sales and it ends $60,000 short. Every step you know about is in it. What does the $60,000 most likely mean?
Rounding in cents doesn't produce $60,000. A gap that doesn't close is information: it could be a training till sending sales, a missing filter or a duplicated join. Taking either side's word for it skips the very check the bridge exists for.