One question, three right answers: define the metric before you count

Three colleagues message the analyst on the same morning. Each asks, more or less, “how many customers do we have?”. Wrong Number, our practice game, runs on the accounts table of the same fictional marketplace, Meridian: 7,136 accounts, including sellers, staff and test accounts. Here is what each of them actually needs:

None of these numbers is wrong. Each one is right for exactly one question. Swap them and each colleague makes a bad decision: product overstates its buyers, marketing leaves four in ten of its audience out, operations never sees the failed checkouts.

This is the case behind Wrong Number, our free three-question game: you can make an optional guess on the first question, ask each colleague up to two clarifying questions, then choose the filters. The challenge is deciding who counts.

Short version: a metric is a definition plus a number. Write the definition first — who, what, when, excluding what — and only then count.

Why definitions go wrong

• The same word means different things in different teams. “Customer”, “active”, “revenue” and “order” each have several reasonable meanings.
• The data doesn’t know your intent. The table has every account; it can’t tell whether you care about buyers or visitors.
• Filters are invisible in the result. A number doesn’t show that test accounts were excluded — or that they weren’t.
• AI assistants guess silently. Ask an assistant for “customers” and it may pick an unstated definition. See how to ask AI about your data.

A one-line template

[Measure] of [population], [condition], [period], excluding [exclusions].

For example: Count of buyer accounts with at least one paid order, 1–31 August 2026, excluding test accounts. If you can’t fill in the line, the question isn’t ready for a query — or for an AI.

The same trap with money

On the Meridian practice marketplace (the SQL-course data, 31 August – 22 September 2026), four money measures that can get confused are the goods value of paid orders (€552,645), what customers paid after discounts (€529,269), that minus completed refunds (€518,808) or the platform’s commission income (€67,040). These are not interchangeable definitions of accounting revenue. Name the measure explicitly; customer payments and marketplace GMV are not the platform’s own income. More in how to check numbers an AI gives you.

FAQ

What is a metric definition?

A precise statement of what is counted: the measure (count, sum, share), the population (which accounts, orders or events), the condition, the period and the exclusions. Two people with the same definition and the same data should get the same number.

How do I define “active users”?

Decide what action makes someone active (a visit, a purchase, a login), over which window (7, 28 or 30 days), and which accounts are excluded (test, staff, bots). Then write it down where others can see it. There is no universal definition; there is only the one your decision needs.

Why do different reports show different numbers for the same metric?

They may use different definitions, periods or data snapshots; one query may also be wrong. Compare definitions and cutoffs, then trace the queries and source records. For a reusable brief, see the metric definition template.

Is there a way to practise this?

Yes: Wrong Number has three questions, needs no SQL and no account, and ends with a result card you can share.