New York – October 13, 2026 -- KNOWIDEA, the decision intelligence company, has published a research paper co-authored with Georgia Tech and developed with one of the world's largest asset managers, aimed at making AI-generated financial database queries verifiable before they inform decisions.
The paper, titled "VeriPIE: Checkable Query Plans and Evidence-Guided Selection for Text-to-SQL," describes a method that improves the accuracy of AI-written database queries. Each answer comes with evidence that a risk team can validate before acting on it.
Silent query errors return plausible but wrong numbers
Banks, asset managers and corporate finance teams are adopting AI tools that convert plain-English questions into database queries, a task known as text-to-SQL. According to the paper, the main danger is not the query that fails but the one that runs, returns a reasonable number and is wrong.
The paper's example involves two common ways of writing "no matching payment." Both return identical results on the same database. Once a single payment record with a missing account ID is added, one version silently excludes every unmatched account while the other still returns them. No error message or warning appears, and the report looks complete.
The paper warns that a wrong query can silently omit exposures, double-count transactions or miscalculate financial aggregates while returning plausible results.
Omitted exposures and double-counting hit risk, reporting and operations
- In risk, an omitted exposure understates a portfolio's risk.
- In reporting, double-counted transactions misstate a regulatory filing.
- In operations, a reconciliation that drops unmatched accounts hides the breaks it was built to find.
Majority-agreement selection falls short
Most AI systems settle on the answer that most of their candidate queries agree on. The paper argues this is inadequate because queries can agree on today's data and disagree tomorrow, when the data changes.
KNOWIDEA is offering 30-minute sessions in which prospective users can see VeriPIE applied to their own data, based on one decision they bring.