What's the bigger risk with AI in finance: a wrong answer or the wrong data?

What worries you more about AI in finance, a wrong answer or an answer based on the wrong data?

Andrew Weiss, Head of Product at Reportworq, answered this one for Built To Know.

"What worries me is a confident answer that nobody checked."

Why is a confident AI answer harder to catch than a wrong one?

A wrong answer that looks wrong gets caught. The risk is the one that looks finished. AI writes in complete sentences, uses the expected terms and ends on a clear number. Nothing in that format tells a reviewer whether the number came from the approved forecast or from a spreadsheet someone emailed last quarter.

Software teams handle a version of this already: code is tested before it ships, and the test results are the evidence. Finance has the same discipline for models, with rules for how they are built and reviewed. An AI answer that skips that discipline is an unvalidated number. An unvalidated number becomes a decision nobody can defend.

What this means for your team

Your stakeholders are already asking AI about your numbers. When the answer reads well, they act on it, and the follow-up lands on your team: where did that figure come from, and why doesn't it match the report you sent?

The fix starts with what the AI is allowed to read. When it answers from the reports Finance has already built and approved, every answer traces back to a source your team signed off on. Your analysts stop reverse-engineering numbers they didn't produce. When someone asks how the AI reached its answer, you can show them the report it read. That is governed AI in practice.