Automate quotes with AI: what an agent does, step by step

Written by the editorial team · 12 November 2026 · Updated12 November 2026

A quote agent reads the customer request, finds the closest past quotes, builds the lines from your current catalogue and writes the quote in your own template, citing where each price came from. A person in the company checks prices, discounts and assumptions, then sends it. To start, pick one type of quote, gather recent quotes and an up-to-date catalogue, and name the person who validates.

A request for a quote lands in a mailbox. Someone opens it, remembers that a similar job was quoted a few months ago, and starts looking. This page describes what a quote agent does from that point on, and what you need in place for it to be useful.

Where does the time go today?

Very little of the time goes into the pricing decision itself. Most of it goes into the work around it:

• searching mailboxes and shared folders for the closest past quote;
• reading the customer’s request twice to extract quantities, options and deadlines;
• checking the current catalogue against the price in the old quote;
• retyping everything into the quote template;
• waiting for someone to validate a discount.

You can measure this on your own business before deciding anything. Note, over two weeks, how many minutes each quote takes from request to sent. Then multiply: minutes per quote, times quotes per month, times the loaded hourly cost of the people involved. That figure is yours, and it tells you whether the project is worth starting.

What does a quote agent read?

Three kinds of material, in read-only mode:

1. Past quotes, in Word, Excel or PDF, stored in SharePoint, OneDrive or Dropbox.
2. The catalogue or price list, wherever it lives today.
3. The customer’s history: earlier mails, earlier orders, special conditions already agreed.

The agent does not need a clean database. It needs the documents your team already uses, in folders you point it to.

What does it prepare, step by step?

1. It reads the request and lists what is asked: items, quantities, options, deadline, constraints. It also lists what is missing, for example a delivery address or a reference that is ambiguous.
2. It finds comparable work: the closest past quotes and what this customer was offered before.
3. It builds the lines from the current catalogue. When the price in an old quote differs from the catalogue, it shows both instead of choosing silently.
4. It writes the quote in your Word template, plus a draft reply to the customer.
5. It cites its sources: each line points to the catalogue entry or the past quote it came from.

The output is a document your team recognises, in your layout, with the assumptions written at the top.

What does the person validate?

Everything that engages the company. Prices, discounts, margin, payment terms, validity period, and every assumption the agent listed. The person edits the draft like a colleague’s draft, and the person sends it. Nothing goes to the customer without that step, and the agent’s source citations make the check fast because the reviewer can open the original line.

Which mistakes should you avoid?

• Letting the agent invent a price. When no source exists, the line must say so and stay empty.
• Mixing customers. A discount agreed with one customer must not leak into another customer’s quote. The agent should use a customer’s special conditions only for that customer.
• Feeding old prices without dates. A two-year-old quote is a fine example of structure and a poor source of prices. Show the date next to every price.
• Starting with the exceptions. Begin with the most repetitive type of quote, the one your team could almost write from memory.
• Ignoring your template. If the output does not look like your quotes, people will retype it, and the gain disappears.
• Automating the sending. The risk is small on one quote and large on a hundred.

What should you prepare on the company side?

• One type of quote to start with, chosen by the people who write it.
• A handful of recent quotes that your team considers good, and the folder where the rest are kept.
• A current catalogue or price list in a file or a folder, with someone responsible for keeping it up to date.
• The rules that are currently in people’s heads: discount thresholds, who approves what, when a quote needs a second pair of eyes. Write them down in a page.
• One person on your side who decides what is good enough to send.

And with us?

We are a forward deployed agency: three engineers from Télécom Paris, specialised in AI, who work on your documents with you. A quote agent is one of the agents we build to measure. This page describes what such an agent does, not a result obtained at a client.

The first call lasts 30 minutes, is free and carries no commitment. You authorise access to your tools in two clicks, without a password. We read SharePoint, OneDrive and Dropbox in read-only mode, on the folders you choose, and keep no file. A mailbox or a business application can be connected on demand. Two weeks after the call, a first agent runs, then you get a dedicated contact. The agents cite their sources, write Word and PowerPoint files, and can also be used from Claude, ChatGPT or Copilot. Nothing is sent without the validation of a person in your company.

On security: one encryption key per company, servers in France, the database in the European Union, European models (Mistral AI) under a contract that forbids training on your data. It can be installed at the client’s and erased on request. Pricing is a set-up fee based on the volume of documents, then a monthly subscription for the whole company with unlimited users. The amounts are given during the call.

More on this use case: automating quotes with an AI agent. On data handling: data security.