Using AI for Data Analysis Without Fooling Yourself

AI is genuinely good at data work — writing SQL, explaining a chart, spotting patterns in messy CSVs. It is also genuinely good at inventing plausible-sounding numbers that do not exist. If you present a fabricated statistic to your CFO, you own that mistake, not the model. Here is how to use AI for analysis without walking into that meeting.

Great uses

• Translating a business question into a SQL or spreadsheet formula.
• Explaining an unfamiliar dataset column by column.
• Suggesting the right chart type for the story you want to tell.
• Writing the plain-English summary of a finding your team actually understands.

The one rule that saves you

Never let AI produce a number without also producing the query or the calculation behind it. If you cannot re-run the calculation yourself, you do not have a number — you have a rumour. This single rule prevents most AI-in-analytics disasters.

A repeatable analysis workflow

1. Write the question in one sentence.
2. Ask AI for the query and the assumptions it is making.
3. Run the query yourself against real data.
4. Ask AI to sanity-check the result — is the number plausible given the business?
5. Write the summary in your own voice; ask AI to critique it, not replace it.

Warning signs

• The model quotes a specific percentage but cannot show you the calculation.
• The chart looks compelling but the axis labels are vague.
• The narrative is very confident and slightly too convenient for the person who asked.

In analytics, the fastest way to look smart is to double-check the number before you send it.