Before you act on a number
A number in a report has authority it did not earn. Four questions give it back the humility it should have arrived with.
Reports, exports and analytics all produce figures that look equally solid. The difference between a decision and a mistake is usually four questions asked before the number goes into a slide.
What period is this, exactly?
Most surprising numbers are a date range problem. A month that includes a public holiday, a comparison of a 31-day month to a 28-day one, a "last 30 days" that quietly includes today's incomplete data.
Say the range out loud whenever you present a figure — "March, so four full weeks" — and half of the confusion never happens. If two people quote different numbers for the same thing, the range is where to look first, before anyone starts questioning the tools.
What is the number divided by?
Percentages hide their denominator, and the denominator is where the meaning lives. A 40% conversion rate means something very different at 10 visitors than at 10,000, and the report prints both the same way.
Always carry the raw count next to the percentage. It is one extra column and it stops the entire class of decisions made on a rate calculated from twelve people.
Who is counted, and who is quietly excluded?
Every dataset has a filter someone applied and forgot to mention. Test accounts included or not. Your own team's traffic. Cancelled orders. Rows that failed validation and silently never arrived.
Before a number becomes a decision, find out what it excludes. The answer is often more interesting than the figure — when internal traffic turns out to be a fifth of the total, the change you were about to make is no longer the right change.
When two systems disagree
They will, and the instinct is to decide which one is right. Usually neither is wrong: they are counting different things, at different moments, with different rules about what counts as one event.
Do not average them and do not pick the friendlier one. Pick the one whose definition matches the decision you are making, write down which you chose and why, and use that same source next time. Consistency over a year is worth far more than accuracy in one report, because trends survive a systematic error and do not survive a change of source halfway through.
Questions people actually ask
Why do two people quote different numbers for the same thing?
Almost always the date range. A month with a holiday, 31 days against 28, or a “last 30 days” that includes today's incomplete data. Check the range before questioning the tools.
What is wrong with reporting a percentage?
It hides its denominator, and the denominator carries the meaning. A 40% rate means something very different at 10 visitors than at 10,000. Always carry the raw count next to the percentage.
Two systems disagree — which one should I use?
Usually neither is wrong; they count different things. Pick the one whose definition matches your decision, write down why, and keep using it. Consistency over a year beats accuracy in one report.
