Can you use Microsoft Copilot for procurement work?

Yes — you can do real procurement work with Microsoft Copilot on the free or standard web tier, the kind of access most teams have without filing an IT request. In the July 2026 PAIR-20 run, we gave the same case files and prompts to Copilot, ChatGPT, Claude, and Gemini across procurement tasks like contract review, supplier-risk briefing, cost breakdowns, and catalog cleanup. Copilot produced useful first drafts, and the same kinds of errors as the others: wrong arithmetic, invented facts, and tables that contradicted its own text. It is a drafting tool to check, not a source to trust.

Copilot's convenience is the risk. Because it sits inside Word and Excel, its output shows up in documents you are already working on. A standalone AI tool feels like a separate step: you paste in, you review, you paste out. Copilot's suggestions appear in place, which makes it easy to accept a number without checking where it came from. In the July 2026 run, that pattern showed up in four tasks:

• Supplier-risk brief (T04): Copilot invented a backup-contract expiry date for a supplier called Dravec. The case file never stated that date. The invented detail survived into a VP-ready summary — leadership would have been acting on a fact that does not exist.
• MRO catalog cleanup (T06): MRO is maintenance, repair, and operations — the bolts, gloves, and filters a company buys in bulk. A box-quantity field in the case file was blank; it needed confirmation from the supplier. Copilot filled in "Likely 100ct" and used that guess as the denominator for unit pricing. Every unit cost calculated from that field would be wrong.
• Landed-cost breakdown (T03): A landed cost is the full price of getting a product to your warehouse — freight, duties, insurance included. Copilot's summary table and its own written explanation gave different numbers for the same line items. When the table says one total and the text says another, you cannot send either number forward. The run recorded a partial verdict: partly right, partly contradicting itself.
• Contract review (T02): Copilot found the relevant clauses, but only after a more specific prompt that pointed it toward what to look for. The first, broader prompt missed provisions a reviewer would need to flag.

Watch the blank fields hardest. Both fabrications in this run landed on fields the source file left open — a contract expiry and a pack quantity. Before Copilot's output shapes a supplier register, a cost comparison, or a recommendation you send upward, check every critical field against the original source.

This is one dated run, not a verdict on the product. These were single runs on the July 2026 free and standard web tiers, not averages across many tests. An error Copilot made here it might not make on a re-run, and a task it handled well it could stumble on next time. For how all four platforms compared, see which AI platform is best for procurement work. The takeaway that holds: the verification habit, not the platform brand, is what makes the work safe to send forward.