Copilot prompts for buyers: 8 ready-to-copy examples

Eight Copilot Chat prompts for buyers: request for quotation, offer comparison, negotiation and supplier panel, from Beginner to Expert level.

By QDNA · Published on September 10, 2026 · Updated on September 10, 2026

In brief. The buyer requests, compares, negotiates and monitors their panel. These eight prompts range from the request for quotation to the extraction test set, with two prompts per level of the course. They share three rules: the criteria are announced before the offers are received; missing cells are written "not specified" instead of being guessed; Copilot builds the grid and does the sums, the buyer chooses.

When to use it

Before attaching an offer, check the confidentiality clauses of the tender and the internal policy, anonymise the supplier and attach only the useful pages. Copilot Chat accepts PDFs, Word documents and Excel workbooks; prepare a clean sheet before any calculation.

To adapt one of these prompts, open the composer and choose the "Buyer" template in the list of job templates.

Beginner

1. Request for quotation comparable from one supplier to the next (Beginner)

BeginnerContext. Renting ten laptops for six months, with the evaluation criteria announced from the request onwards.

Write a request for quotation for the rental of ten laptops for six months. Context: classroom training, standard office configuration, on-site delivery on [date], collection of the equipment at the end of the period. Expectations: 90-word email, professional tone, then the numbered list of the information expected from the supplier: monthly price excluding tax, delivery lead time, warranty and replacement in case of failure, collection terms, reply deadline on [date].

Expected result. A precise request whose replies will be comparable on the same criteria.

Mistake to avoid. Leaving the evaluation criteria out of the request: the replies arrive in formats that cannot be compared.

Related lesson. Writing a good prompt: the four ingredients

2. Summarise terms and conditions with the article cited (Beginner)

BeginnerContext. A supplier's general terms and conditions of sale, a public document, are attached and you want the five points that bind you.

From the attached file supplier_terms.pdf only, list the five points that bind the buyer: term, notice period for termination, late penalties, price revision, warranty. For each point: a one-sentence summary, then the word-for-word quotation and the article number. If a point is not covered in the document, write "not found" instead of filling in with a general rule.

Expected result. Five points summarised and quoted, each checkable in the document, with the absences flagged.

Mistake to avoid. Asking "summarise these terms": Copilot produces a fluent summary where a missing clause is filled in with an invented rule.

Related lesson. Checking an answer: sources, figures, dates

Intermediate

3. Comparison grid for three offers (Intermediate)

IntermediateContext. Three anonymised offers are attached; you want a common grid and a neutral summary, with no recommendation.

From the three attached files only, offer_A.pdf, offer_B.pdf and offer_C.pdf: Step 1: extract the data into a common grid: total price excluding tax, lead time, warranty, payment terms, penalties. Write "not specified" for any cell absent from an offer. Step 2: flag the incomplete offers and the elements without figures. Step 3: calculate a total cost over three years from the amounts present only and show the formula. Format: comparison table, then a neutral five-line summary, with no recommendation. Sample expected row: | Lead time | 3 weeks | not specified | 15 days |.

Expected result. A homogeneous table ready for the purchasing committee, with the missing cells visible and the calculation shown.

Mistake to avoid. Asking for a recommendation: the choice belongs to the buyer, and Copilot readily recommends the best-written offer.

4. Web search on a price revision index (Intermediate)

IntermediateContext. A revision clause refers to a public index and you want its latest value with the official source.

Search the web for the latest published value of the [index name] index used in price revision clauses in France. Use only official sources (insee.fr, ministry websites, public bodies). Give the value, the reference period, the publication date, the exact link to the page consulted. If you find several values depending on the base or the period, present them all. If no official source publishes the index, say so instead of citing a commercial website.

Expected result. A dated index value, with an official link to open before applying it in the calculation.

Mistake to avoid. Asking "what is the current index?": Copilot cites a commercial website and an outdated value.

Related lesson. Web on or off: when Copilot should search

Advanced

5. Renewal negotiation file (Advanced)

AdvancedContext. The current contract, the renewal proposal and the year's volumes are attached; the mandate is built in four steps.

From the attached files current_contract.pdf, renewal_proposal.pdf and volumes_2026.xlsx only, proceed in four steps and wait for my validation after each. Step 1: identify the clauses that worsen for the buyer, with the quotation from each version. Step 2: quantify the annual impact of each worsening from the volumes, showing the calculation. Step 3: formulate a reasoned counter-proposal per clause. Step 4: propose a concession limit per clause, marked "to be validated by the head of purchasing". Format: table clause, quotations, impact, counter-proposal, limit.

Expected result. A file quantified clause by clause, with a concession mandate to validate before the meeting.

Mistake to avoid. Negotiating without figures: the better-prepared party wins, and an impact that has not been calculated cannot be defended.

6. Critical dependencies in the order database (Advanced)

AdvancedContext. Two years of orders are attached as a clean sheet; the analysis goes through the Python interpreter and is checked.

Attached file orders_2025_2026.xlsx: one sheet, row 1 = headers (Date in DD/MM/YYYY format, Supplier, Purchasing category, Amount excluding tax in euros), one row per order, no total row. Before calculating, describe your method in three lines. Step 1: calculate the spend and the share per supplier. Step 2: identify the purchasing categories with a single supplier, that is, the critical dependencies. Step 3: check the totals by cross recalculation (by supplier, then by category) and flag any inconsistency in the database. State the number of rows taken into account.

Expected result. A map of dependencies based on actual orders, with the totals cross-checked and the anomalies listed.

Mistake to avoid. Confusing high spend with risk: the risk lies in the impossibility of substituting, not in the amount.

Expert

7. Have the team's request for quotation prompt improved (Expert)

ExpertContext. You submit the request for quotation prompt the team shares to have it rewritten block by block, then you arbitrate.

Improve the prompt below for Copilot Chat. Rewrite it according to the blocks Role, Goal, Context, Sources, Format, Constraints, Verification. For each modification, explain in one sentence the problem it fixes. Add a robustness rule: what the prompt must do if the described need specifies neither quantity nor date. End with the questions to ask me to complete the prompt.
Prompt: [the team's current prompt pasted]

Expected result. A prompt rewritten block by block, with a rule for incomplete needs and every change justified.

Mistake to avoid. Accepting every modification without testing it on a real request: a change can break what was working.

Related lesson. Asking Copilot to improve your prompt

8. Test set for extracting data from an offer (Expert)

ExpertContext. You measure an extraction prompt on ten offer excerpts whose expected values you know.

For each of the ten offer excerpts below, extract: total price excluding tax, delivery lead time in days, warranty period in months, payment terms. Rules: copy the values as they are, convert lead times into days and warranties into months showing the conversion, write "not specified" if the data is absent, never infer one value from another. Output: table Number, Price excl. tax, Lead time, Warranty, Payment, with ten rows.
Excerpts, anonymised: [ten pasted excerpts]

Expected result. Ten rows extracted; you compare with the expected table, count the score over three runs and fix one rule at a time.

Mistake to avoid. Judging the prompt on a well-written offer: the test set must contain ambiguous excerpts and missing data.

Related lesson. Evaluating a prompt on a set of ten cases

Frequently asked questions

Can Copilot Chat recommend a supplier?

It can build the grid and flag incomplete offers, but the choice belongs to the buyer and to the purchasing procedure. Ask for a neutral summary with no recommendation, then decide with the committee.

Can I attach the offers received from suppliers?

Offers often contain information the supplier regards as confidential. Check the internal policy and the confidentiality clauses of the tender, anonymise the names and attach only the useful pages. Attached files stay in your work OneDrive.