Can AI find hidden fees in a quote?

AI can find hidden fees in a quote, but that is only one part of making the comparison safe.

In the July 2026 PAIR-20 SaaS quote task, the case included a buried mandatory compliance module that would have appeared on the invoice even though it was not in the headline price. The run notes say the platforms found that mandatory fee and ignored the expired 2025 quote left in the file. That is the good news: buried add-ons are a reasonable use case for AI review.

The same task also showed the limit. A unit error still survived on one platform after the structured brief named the unit to normalize. Finding the fee did not make the all-in comparison correct.

The consulting-bid task makes the same point from another angle. The case put important costs outside the simple price table: Hawthorn's coordination fee, Pinnacle's stream-overage exposure, and Bridgepoint's travel-and-expense treatment. In the Reality Check write-up, ChatGPT, Claude, and Gemini surfaced that hidden-cost structure on the generic prompt. Copilot passed the structured brief run, though it missed the scope cap on the generic prompt.

Then the presentation broke. ChatGPT computed the all-in math correctly and ranked the most expensive bid first. Gemini showed Bridgepoint's correct all-in figure in the narrative, but listed the effective blended rate as if those extra costs were not part of the rate.

Use AI to widen the fee search. Ask it to look outside the pricing table: footnotes, terms, addenda, required modules, pass-through costs, scope caps, and expired or superseded versions.

Separate fee detection from award logic. A model can find the hidden fee and still mishandle the unit, blended rate, or final ranking.

Verify the final table. For procurement teams, the practical check is simple: every hidden cost found in the notes should appear in the all-in total, and the ranking should match that total. If the table drops the fee, the decision has already drifted.