ChatGPT for Procurement: Prompts, Maths and Limits

ChatGPT for procurement is most useful for the reading and writing around a purchase: cleaning and categorising spend lines, drafting RFQs from a specification, tabulating bids, pulling payment, warranty and liability clauses out of contracts, preparing negotiation fact sheets and drafting supplier follow-ups. Claude, Gemini and Microsoft Copilot do the same jobs. It is least reliable at the arithmetic and the judgement: landed cost, EOQ, safety stock and the award decision belong in a spreadsheet and with the buyer. Supplier prices, bids and contracts are confidential, so use your employer's approved AI tool or the AI inside your ERP, never a personal ChatGPT account.

Checked on 11 October 2026 against the sources listed under this post. Product claims are dated October 2026.

What can ChatGPT do for a buyer?

A buyer's week is full of text: specifications, RFQs, supplier emails, contracts, meeting notes. General AI assistants handle text well, so the gains are real. They are weaker on numbers and have no access to your supplier history unless you give it to them.

Prompts that buyers can trust

Procurement prompts need the data, the format and the check. Three examples:

Spend categorisation. "Here are 200 anonymised spend lines (supplier code, description, amount). Suggest a category for each from this list: [your taxonomy]. Add a confidence column and mark low-confidence lines for review. Do not create new categories."

RFQ draft. "Using this specification and bill of quantities, draft an RFQ with scope, technical requirements, documents to submit, delivery terms, evaluation method (technical compliance first, then lowest evaluated landed cost) and a deadline placeholder. Flag any requirement in the spec that is ambiguous."

Contract extraction. "From this anonymised contract text, extract payment terms, liquidated damages, warranty period, limitation of liability, termination rights and governing law, each with the clause number. Write 'not found' where a clause is missing. Do not interpret."

The phrases "do not create", "flag" and "write 'not found'" are what make the output checkable. Without them, the model fills gaps with confident inventions.

This short SCMDOJO lesson, AI in Procurement Basics Course, opens the free AI for Procurement and Supply Chain course; EDWartens is not affiliated with SCMDOJO.

Worked example: EOQ, safety stock and the missing "2"

Inventory maths is where a chat answer can look right and be wrong. Suppose a buyer manages a gasket used on maintenance jobs:

Economic order quantity. EOQ = square root of (2 x D x S / H) = square root of (2 x 12,000 x 75 / 2.40) = square root of 750,000 = 866 units.

Safety stock. z x standard deviation of daily demand x square root of lead time = 1.65 x 12 x 3 = 59.4, rounded up to 60 units.

Reorder point. d x L + safety stock = 40 x 9 + 60 = 420 units.

So the policy is: when stock falls to 420 gaskets, order about 866. Now the trap. Ask a chat assistant for the EOQ and it may return 612 units, which is the square root of (12,000 x 75 / 2.40): it dropped the 2 in the formula. Ordering 612 at a time is not a disaster, but it means more orders a year than needed, and the same kind of slip in a landed-cost or bid table can change an award. Put the formulas in a spreadsheet, and use the AI to explain them, document the assumptions and draft the note to the planner.

Negotiation preparation with AI

Negotiation is where AI is a good sparring partner and a poor negotiator. Give it the anonymised facts (your volumes, the supplier's last three price increases, the market index for the main raw material, your alternatives) and ask for three things: a one-page fact sheet, the five objections the supplier is most likely to raise, and a give-get list of concessions you could trade. Then ask it to role-play the supplier's account manager and push back on your opening position.

What the AI cannot do is set your walk-away point. That comes from your landed-cost model, your switching cost and the risk of a supply gap, which only you and your stakeholders can weigh. Write the walk-away number down before the meeting, not after the AI has made a concession sound reasonable.

The AI inside procurement suites

The ERP and sourcing suites now build these tasks in, working on your real data under your company's agreements.

Microsoft's Procurement Agent FAQ for Dynamics 365 Supply Chain Management (a production-ready preview) describes drafting follow-up emails to vendors on unconfirmed or late purchase orders, reading vendor emails to spot confirmations and change requests, and analysing the downstream impact of a change. It says generated content "should never be used without manual review or supervision", and that emails are not sent automatically unless an admin configures it.

SAP's October 2025 announcement for SAP Ariba described a Joule agent for bid analysis that "automatically compares supplier bid data, including total cost", planned for Q1 2026, an AI summary of supplier questionnaire responses, and intelligent contracting that extracts key terms and searches past contracts for discrepancies. Oracle and Coupa also market AI features for sourcing and contracts; check your own version's release notes for what is switched on.

Keep supplier data confidential

Bids, price lists, contracts and supplier bank details are confidential and often covered by non-disclosure terms. Consumer AI services have their own data rules; Google's Gemini Apps Privacy Hub, for example, says some chats are reviewed by human reviewers and asks users not to enter confidential information. So:

• never paste supplier quotes, price lists or contracts into a public AI tool or personal account;
• use your ERP's AI or the enterprise AI tool your employer approves, under its AI policy;
• anonymise (Supplier A, Part 1, scaled prices) for practice and training;
• keep the award, the contract acceptance and any supplier blacklisting with named people.

A four-week start for a buyer

1. Week 1: read your AI policy; categorise one month of anonymised spend with AI and check 20 lines by hand.
2. Week 2: draft an RFQ for a repeat item and compare it with the last one your team issued.
3. Week 3: rebuild a past bid comparison and see whether the AI's ranking matches the real award, and why not.
4. Week 4: set EOQ, safety stock and reorder point for three items in a spreadsheet, with AI writing the explanation for the planner.

For a grounding in how language models behave, read Generative AI for Engineers: What LLMs Can and Cannot Do, and for other options see our guide to the best free AI courses with certificate.

A free course for buyers and planners

The free AI for Procurement and Supply Chain course covers where AI helps and where it does not, prompts buyers can trust, supplier data and AI policy, spend analysis, supplier risk screening, RFQs and technical-commercial bid comparison, contract red flags, negotiation preparation, demand forecasting, EOQ, safety stock and reorder points, shipment exceptions, and the AI inside ERP and procurement suites, ending with a capstone sourcing event. It is a free course with a verifiable certificate of completion, and anyone can check a certificate on our verification page. Start the AI for Procurement and Supply Chain course at your own pace.

Take the free course

Free

AI for professionals · Beginner · Free

AI for Procurement and Supply Chain

Free

Engineering data · Beginner · Free

Excel for Engineers: Formulas, Data and Dashboards

Free

AI for professionals · Beginner · Free

AI Essentials for Work: ChatGPT, Claude, Gemini and Copilot