Imagine a U.S. bakery asking AI to write its weekly report. Three order cards show $20, $0 and a blank amount. The assistant reports an average of $6.67. The arithmetic can work, but only after silently treating an unknown amount as zero. This synthetic exercise is about reporting definitions, not a customer case or a revenue forecast.

PostgreSQL documents that COUNT(*) counts input rows, COUNT(amount) counts rows with a non-null amount, and AVG averages non-null inputs.[1] Our example has three rows and two known amounts, whose average is $10. That is the average of known amounts—not a known average for every order. The third amount still needs verification.

Zero is a known numeric value; null means the value has not been supplied here; no input rows is a different state. PostgreSQL SUM returns null when there are no input rows, rather than zero.[1] That result alone does not prove there were no sales. Date filters, failed synchronization or permission scope could also leave a query empty. Investigate the business cause instead of asking the writer to guess.

Replacing nulls with zero changes the statistical question. We recommend having the business owner define when zero is legitimate, how missing values appear, and which records qualify. Supply total rows, valid amounts, missing amounts and the reporting scope together. Label the metric “average of known amounts,” not just an impressive-looking number.

Our proposed result card records dates and time zone, filters, data freshness, total rows, valid-value count, aggregate result and completeness status. AI turns those fields into prose; missing amounts require a coverage caveat. For this synthetic example: two of three records have known amounts, averaging $10; one amount awaits verification.

Before rollout, test four sets: all zeros; zero mixed with null; all nulls; and no rows. Check counts, aggregates and the final wording separately. The last two must not automatically become “no revenue this week.” These are proposed tests, not tests run on a customer system. Review the denominator and missing-data state before judging the prose.