The monthly trading review opens with a slide. Within two minutes someone says, “That’s not the number I have.” The rest of the meeting goes on reconciling spreadsheets.
The cause is rarely bad data. It’s that one metric has quietly turned into four.
Four teams, four numbers
Take active customer contribution: what each active customer is worth after the cost of serving them. In a business trading across several markets, every team works it out differently, and every team has a good reason.
Finance
Converts every market at a fixed monthly exchange rate, so it ties back to the accounts.
Commercial
Shares out fulfilment costs in the way that fits its pricing decisions.
Operations
Shares out the same costs another way, to fit how it runs fulfilment.
Product
Needs a governed source its new AI agent can call, not a spreadsheet.
None of them is wrong. But all four go by the same name, so the meeting argues about the data when it should be choosing a definition.
The fix, in three steps
- 01
Separate plumbing from logic
Data engineering lands clean, tested data. Analytics engineering builds the business definitions once, in code.
- 02
Agree one definition
Write it in plain English, settle each disputed choice once, and give it one owner who signs off changes.
- 03
Serve it from one place
Build it into a tested semantic layer that every report, analyst and AI agent reads from.
What changes
The number on the first slide is the number everyone has. When it moves, the talk is about the business, not whose spreadsheet is right. New reports and AI tools start from the same definition, so nobody creates a fifth version.
Do you have this problem?
- Two reports show different values for a metric with the same name.
- Someone updates a reconciliation spreadsheet every month.
- Nobody can say who signs off a change to your headline metric.
If two of those sound familiar, the fix is one definition and a semantic layer, not another dashboard.