Can AI do this? Straight answers for procurement.
Every page here answers one question a procurement person actually asks — in forty words or fewer, then the evidence. The evidence is real: dated runs of the same task on ChatGPT, Claude, Gemini and Copilot, with screenshots on file. No benchmark scores, no vendor claims — what happened when the task met the tool.
Can AI do this?
- Can AI build a supplier risk register?Yes, but it still needs field-level verification. In PAIR-20's July 2026 T04 run, most platforms flagged hidden source-status risks, while Copilot either buried Weiss or fabricated a Dravec expiry.
- Can AI calculate landed cost correctly?It can build the landed-cost stack, but do not trust the final number. In PAIR-20's July 2026 run, all four platforms treated the headline freight quote as incomplete, and none matched the EUR7,940 case answer.
- 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.
- Can AI check a contract before your meeting?Yes, for a first pass. In July 2026, all four platforms found the renewal clauses, but three needed direction before connecting the 60-day price notice to the 90-day cancellation deadline.
- Can AI compare supplier quotes reliably?Not reliably. In PAIR-20's July 2026 run, AI found several quote issues, but one platform carried a unit error into the ranking and another ranked the highest consulting bid first after doing the math.
- Can AI find hidden fees in a quote?Yes, sometimes. In PAIR-20's July 2026 run, platforms found the SaaS mandatory fee and surfaced consulting hidden-cost items, but later failures still changed the recommendation and blended-rate presentation.
- Can AI merge and deduplicate a parts catalog?Yes, with a narrow review step. In PAIR-20's July 2026 T06 run, most platforms separated the glove false duplicate, but Copilot filled Atlas's blank box quantity with "Likely 100ct."
- Can AI rank bids correctly once the math is done?Sometimes. In PAIR-20's July 2026 T05.5 run, ChatGPT calculated the consulting-bid math correctly, then ranked the highest all-in total first in its summary table.
- Can AI review a procurement contract?Yes, but a clause list is not enough. In PAIR-20's July 2026 run, all four platforms found the renewal clauses; three needed direction before connecting the 60-day price notice to the 90-day cancellation deadline.
- Can AI tell true duplicates from false twins in item data?Sometimes, but only if the model keeps blank fields visible. In PAIR-20's July 2026 T06 run, ChatGPT, Claude, and Gemini flagged the glove false twin; Copilot guessed a missing box count.
- Can you trust AI's arithmetic in procurement files?Not as the final check. In PAIR-20's July 2026 T03 run, four platforms rebuilt the landed-cost stack, but every final number differed from the EUR7,940 case answer.
- Do procurement teams already use AI?Yes. Microsoft Work Trend Index 2024 says 75% of knowledge workers use AI and 78% bring their own tools, while only 39% have received company AI training.
- Does a better prompt fix AI's procurement mistakes?No. In PAIR-20's July 2026 run, structured briefs made outputs easier to audit, but did not prevent a unit miss, wrong landed-cost numbers, a ranking contradiction, or stripped hidden costs.
- Does generic AI training work for procurement teams?Generic AI training is a weak bet by itself. Section AI Report 2026 found no measurable productivity gain; CIPD says only 7% of organizations have a learning-transfer process.
- How often does AI fabricate facts in procurement files?PAIR-20 does not give a stable frequency. In one July 2026 set of six single-run tasks, two baited tasks recorded fabricated facts: an invented Dravec expiry and a guessed Atlas box quantity.
- Is free-tier AI good enough for procurement work?Enough for triage, not enough for unsupervised decisions. In PAIR-20's July 2026 run, all findings came from free or standard web tiers, which produced useful drafts and still made arithmetic, ranking, and fabrication errors.
- Which AI platform is best for procurement work?There is no single winner from the July 2026 PAIR-20 run. ChatGPT, Claude, Gemini, and Copilot each passed some tasks and missed others; the useful signal was where they disagreed.
- Will AI invent numbers in procurement work?Yes. In the July 2026 PAIR run, Copilot fabricated a backup-contract expiry in a risk register and guessed an unstated box quantity in a catalog merge. Under the rubric, each fabrication failed the task.
Procurement-AI glossary
- What do Incoterms mean for a landed-cost comparison?
- What is a blended rate in consulting bids?
- What is a freight SLA?
- What is a statement-of-work audit?
- What is a structured brief (vs. a prompt)?
- What is a supplier risk register?
- What is AI fabrication in procurement documents?
- What is an MSA (Master Service Agreement)?
- What is BYOAI (bring your own AI)?
- What is landed cost?
- What is learning transfer in AI training?
- What is PO price variance?
- What is quote normalization in procurement?
- What is savings realization (reported vs. realized)?
- What is total cost of ownership (TCO)?
The answers come from the course
Twenty lessons, each one a real procurement task worked end to end on four AI platforms. Lesson 2 is free to read.