Skip to content
QDNALearn AI, from beginner to expert
FR

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

Eight Copilot Chat prompts for sales teams: follow-up email, pitch, negotiation and sales analysis, from Beginner to Expert level.

In brief. Sales work produces emails, pitches and meeting preparations every day. These eight prompts range from the follow-up email to the meeting preparation agent, with two prompts per level of the course. All follow the same rule: instruction first, attached source last, and an explicit instruction for missing information, so that no performance figure is invented.

When to use it

Before pasting a client history, remove the names, contact details and amounts you would not send to an external contractor. Without a paid licence, Copilot Chat does not read your CRM: export the record or the history and attach the file.

To adapt one of these prompts to your situation, open the composer and choose the "Sales" template in the list of job templates: the fields fill in, and all that remains is to replace the brackets.

Beginner

1. Follow-up email after a quote (Beginner)

BeginnerContext. A client has not replied for two weeks to a quote sent after a trade show.

Write a follow-up email for a client who has not replied for two weeks. Context: quote of [amount] euros for [subject] services, contact made at a trade show. Expectations: 80 words, professional and warm tone, an offer of a phone call with two time slots, standard signature. Answer in English.

Expected result. A short follow-up ready to send, with a subject line under 60 characters and a single call to action.

Mistake to avoid. Asking for "a follow-up" without stating the tone or the length: the reply is long and pushy, and you rewrite it.

Related lesson. Writing a good prompt: the four ingredients

2. Rephrase a technical paragraph for a buyer (Beginner)

BeginnerContext. A product description written by the engineering office must become readable for a non-technical buyer.

Rephrase the paragraph below for a buyer with no technical training. Keep exactly the three figures it contains. Replace every technical term with an everyday expression or an explanation in brackets. Length: 90 words at most, formal address, no exclamation marks.
Paragraph: [pasted text]

Expected result. A clear paragraph where the original figures are intact and every trade term is explained.

Mistake to avoid. Saying "make it simpler" without listing what must be kept: a figure disappears or changes along the way.

Related lesson. Your first useful conversation: summarise, rephrase, translate

Intermediate

3. Benefit-proof pitch from the product sheet (Intermediate)

IntermediateContext. You attach the official product sheet and want a pitch aligned with it, with no invented figure.

You are a sales engineer. From the attached file product_sheet.pdf only, write a pitch in three blocks: client need, benefit, proof. Context: industrial client sensitive to line stoppages. Expectations: one block per priority argument, 120 words at most per block, English, no unverifiable superlative. Each proof cites the page of the sheet it comes from. If the sheet contains no proof for a benefit, write "proof to be provided" rather than inventing.

Expected result. Three benefit-proof blocks, each tied to a page of the sheet, with the gaps flagged.

Mistake to avoid. Not restricting the source to the attached sheet: Copilot fills in with plausible but false performance figures.

Related lesson. Attaching a Word, PDF or Excel file and asking the right questions

4. Answer an objection in the team's style (Intermediate)

IntermediateContext. You provide two objection replies the team considers successful to get a third in the same style.

Write a reply to the following objection from a prospect: "[objection, for instance: your solution is 20 percent more expensive than the one we have today]". Expected style, see these two replies our team considers successful: [pasted reply 1] and [pasted reply 2]. Rules: acknowledge the objection in one sentence, provide a verifiable fact, end with an open question. 100 words at most, formal address. If you need a figure I have not provided, put it in brackets so that I can fill it in.

Expected result. A reply in the house tone, with the missing figures flagged in brackets instead of being invented.

Mistake to avoid. Describing the style with adjectives ("punchy, human") instead of showing two examples: the reply stays generic.

Related lesson. Giving examples and a role to set the style

Advanced

5. Negotiation file in four steps (Advanced)

AdvancedContext. The history of exchanges with the client is attached, anonymised, and the negotiation meeting is on Thursday.

Goal: prepare Thursday's negotiation. Proceed in four steps and wait for my validation after each. Step 1: summarise the client's position in three points. Step 2: list their objections, each with the exact quotation from the file that supports it. Step 3: for each objection, propose a reply and a maximum concession. Step 4: propose a three-part meeting plan. Format: table objections, quotations, replies, concessions, then plan in bullets. If an objection has no quotation, write "not found".
Source: client_history.docx attached, anonymised.

Expected result. A table of objections traceable back to the exchanges, then a meeting plan validated step by step.

Mistake to avoid. Asking for the negotiation plan in a single prompt: the objections are paraphrased without quotation and you answer beside the point.

Related lesson. Breaking a task down into three chained prompts

6. Client portfolio analysis with the Python interpreter (Advanced)

AdvancedContext. An export of the year's sales, prepared as a single sheet with no total row, is attached to the prompt.

Attached file sales_2026.xlsx: one sheet, row 1 = headers (Client, Date in DD/MM/YYYY format, Amount excluding tax in euros), one row per sale, no total row. Before calculating, describe your method in three lines. Then: 1) rank the clients by decreasing revenue, 2) calculate the share of the top ten clients in the total, 3) list the clients down by more than 20 percent compared with the first six months. State the number of rows taken into account. Flag any non-compliant date instead of ignoring it.

Expected result. A stated method, a ranking, a concentration rate and a list of declining clients, with the number of rows processed.

Mistake to avoid. Attaching a raw three-sheet export with totals in the rows: Copilot adds the totals to the sales.

Related lesson. Analysing an Excel table with the built-in Python interpreter

Expert

7. Instructions for a meeting preparation agent (Expert)

ExpertContext. You create an agent with the Copilot Chat agent builder, based on instructions only, hence at no extra cost.

You are the meeting preparer of the [department name] sales team. From the company profile the user pastes or attaches, you produce in this order: three discovery questions, the roles of the likely decision-makers, an approach angle in two sentences. You invent no figure about the company: if the profile does not contain it, you write "not found". You do not write emails: if the user asks, you refer them to the team's standard follow-up prompt. If the profile contains personal data naming individuals, you flag it first and offer to remove it. Answer format: three titled sections, 150 words in total. Sample good answer: [pasted example].

Expected result. An agent with a clear scope, which refuses what is not its role and always answers in the same format.

Mistake to avoid. One-line instructions ("you prepare meetings"): the agent invents figures and changes format with every use.

Related lesson. Creating a simple agent with the agent builder

8. Test set for a lead qualification prompt (Expert)

ExpertContext. You measure a prompt that sorts incoming requests on ten cases whose expected answer you know.

Classify each of the ten requests below into a single category among: Hot, Warm, Cold, Off-target. Rules: Hot if a budget and a deadline are mentioned, Warm if only one of the two, Cold if neither, Off-target if the need does not match our [offer] offer. If you hesitate between two categories, choose the lower one and flag it with an asterisk. Output: table Number, Category, Justification in one sentence.
Requests, anonymised: [ten pasted messages]

Expected result. Ten classified rows; you compare with the expected table, record the score, change a single rule and run again.

Mistake to avoid. Judging the prompt on a single successful example: without a test set or a score over three runs, quality is felt, not measured.

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

Frequently asked questions

Can I paste a client's full history into Copilot Chat?

Paste only what you would send to an external contractor. Replace people's names with initials and remove contact details. Enterprise data protection prevents the models from being trained, but conversations remain logged within your organisation.

Can Copilot Chat read my CRM or my client emails?

The CRM, no: no access without a paid Microsoft 365 Copilot licence. Your emails, yes: in Outlook, Copilot Chat queries your mailbox, your calendar and your meetings through lexical indexing. For the CRM, export the record or the history, then attach the file at the end of the prompt.

Sources