Before you present the number
Four checks that take ten minutes and stop you defending a wrong figure in a room full of people.
The query ran. The number looks reasonable. This is the most dangerous moment in analytics, because a reasonable-looking wrong number gets acted on.
Ten minutes, four checks.
1. Does the total reconcile?
Find something the business already counts — invoiced revenue, the count in the source system, last month's published figure — and compare.
If they differ, understand why before you present. There is often a legitimate reason: different date basis, cancellations, tax. "It differs because we count from dispatch, not from order" is a fine answer. "It differs" is not.
2. Where did the rows go?
Count the rows at the start of your query and at the end. Every filter and every join moved that number. Can you account for each drop?
An inner join that silently discarded 4% of orders because their customer record was missing is the classic. Nothing errors. The number is just quietly too small, and consistently so.
3. What is the denominator doing?
Rates mislead most easily. A conversion rate that jumped from 2% to 6% might be success — or the denominator collapsed because tracking broke.
Always look at the numerator and denominator separately before believing the ratio. Plot all three if you can.
4. Is the sample big enough to say anything?
A segment with eleven customers has a retention rate that will swing wildly on one person's decision. The chart draws it at the same size and confidence as the segment with eleven thousand.
Put the counts on the chart. If a bar is built on a small number, say so on the chart, not in a footnote — the footnote does not survive the screenshot.
Saying what you are unsure about
There is a real fear that caveats undermine your credibility. They do the opposite, and the mechanism is simple: the person who flagged the weakness themselves is the person still trusted after it is found by someone else.
Two sentences is usually enough. What the number is; what would change it.
Try this
Take the last number you presented and run these four checks on it now, with no pressure. If one of them turns something up, that is not a bad day — it is the cheapest possible version of finding out.