LogoLucyWorks
  • Lessons
  • Answers
  • Pricing
  • About
  • Ask Lucy
  • Book a call
LW·A03·Can AI do this?← All answers

Can AI catch a unit-of-measure error in a quote comparison?

Sometimes — and you cannot predict which platform will trip. In PAIR-20's July 2026 run, Gemini caught a per-100-seats unit trap unprompted; Claude carried the wrong number into its ranking even after being told the unit.

A unit mismatch is the classic quote-comparison trap: one supplier prices per seat, another per 100 seats, and the cheapest-looking bid is only cheapest because the units don't line up. It is exactly the kind of mechanical check people assume AI handles well.

The July 2026 PAIR-20 run says: not reliably. The task planted a "per 100 seats" price among per-seat quotes. Gemini caught it on a generic prompt — no hint, no structured brief. ChatGPT caught it once the brief named the fields to normalize. Claude was told the unit explicitly and still carried the flat figure into its final ranking; the structured brief did not fix it.

Two things follow for a working procurement team:

The failure is platform-specific and unpredictable. The same file, the same day, produced a clean catch on one platform and a silent miss on another. Whichever platform your team uses, you cannot assume this class of error is covered — and you cannot know in advance whether yours is the one that trips.

A structured prompt helps, but is not a guarantee. Naming the fields to normalize turned one miss into a catch, and left another miss untouched. Treat the prompt as a control that raises the floor, not a cure — the unit column still needs a human eye before the comparison leaves your desk.

Where this comes from

  • PAIR-20 July 2026 run of record, task T01 — single dated runs, screenshots on file

Last checked Sat Jul 04 2026 00:00:00 GMT+0000 (Coordinated Universal Time). Evidence comes from dated, single-run platform sessions with screenshots on file — read each finding as “this happened,” not “this always happens.”

Work this yourself — from the course

Normalize Messy SaaS Vendor Quotes — Unit Conversion, Hidden Fees, Cross-Platform VerificationShows how to feed AI three messy, differently-formatted SaaS vendor quotes (plus a distractor) so it normalizes them onto the same basis instead of comparing sticker prices that hide a per-seat trap.

See what the platforms caught — and missed

Twenty procurement tasks, four AI platforms, real dated runs. Lesson 2 is free to read, no account needed.

Read the free lessonTraining for your team
LogoLucyWorks

AI training that ends with a working agent.

Product
  • The 20 Lessons
  • Answers
  • Ask Lucy
  • PAIR-20 benchmark
  • Reality Check
  • Pricing
  • Book a call
Company
  • About
Legal
  • Cookie Policy
  • Privacy Policy
  • Terms of Service
© 2026 LucyWorks. All Rights Reserved.