Creating quotes with AI: a faster first draft without losing your margin

Writing a quote is largely assembling text you've written before. Reading the request, finding the right blocks, filling in the customer details, putting it all in your house style. That's exactly the kind of work AI is good at. There is one part AI is bad at: arithmetic. And that's the part that becomes binding once the quote is with the customer.

The rule for this whole article: the model writes, the system calculates, a person sends.

Where does the time go in a quote?

Take your quoting process apart and you usually see six steps.

• Reading and understanding the request. Little time. AI summarises and spots what's missing.
• Asking for missing information. Moderate. AI drafts the questions.
• Choosing the right text blocks and components. A lot of time. This is where the gain is.
• Setting prices and discounts. Moderate. Not AI: this comes from your system, with fixed rules.
• Writing the draft and applying your house style. A lot of time. Easy to automate.
• Checking and sending. Little time. Stays human work.

The two biggest items are where language models are strong. The item where things go wrong is where they're weak. That split is the whole design challenge.

How to set it up

1. One entry point for requests

Requests that arrive by email, form and phone and land in three inboxes can't be automated. First make sure everything arrives in one place: a shared mailbox, a form or a field in your CRM. How to let a mailbox sort itself is in automating email processing in Outlook.

2. A summary that shows the gaps

Have AI summarise each request in a fixed format and state what's missing: quantities, deadline, location, specifications. That alone saves a round of back-and-forth emails.

3. A library of text blocks

Your best quotes contain the text you need. Pull out the reusable blocks, such as service descriptions, approach, terms and references, and put them in one place. That's one-off manual work, and it determines the quality of everything that follows.

4. Prices from the system

AI chooses which price rules apply. Your price list, CRM or a fixed calculation sets the amounts. If your prices only live in someone's head or an old spreadsheet, that's the real problem. Solve that first.

That's how the quoting tool we built for De Duurzame Jongens works: the calculation model from their Excel sheet now lives in code. What a visitor sees comes from their own calculation rules.

5. A draft, not a final version

The result is a draft quote that someone reviews. Make that a mandatory step in the workflow, not a good habit.

6. A checklist at approval

Are the amounts right? Is the correct customer name in it? Are the terms the current version? Are there no commitments you can't keep? Four ticks.

How an AI employee prepares such a draft is shown with the Quote Assistant. For chasing quotes you've sent, there's the Quote Follow-up Assistant.

Four mistakes that cost money

Letting the model calculate prices

The most expensive mistake. Ask a language model to apply a 12% discount to €14,750 and you get an amount that looks plausible and may be slightly off. In a quote you've sent, that's your problem. Why this happens is explained in preventing AI hallucinations.

Working from an outdated price list

Automation multiplies whatever is in your source. A price list from last year then produces not one wrong quote, but every quote.

Commitments nobody checks

Models like to fill in the gaps: delivery times, guarantees, "including aftercare". If nobody watches for that, your quote contains promises you never made. Put this explicitly on the checklist.

Automating the sending

It saves the last minute. But the difference between a mistake in a draft and a mistake at the customer is the entire risk of this process.

What to measure

Time saved is the easiest number and the least interesting. Also measure:

• Lead time: from request to quote sent.
• Win rate: if quotes go out faster but are won less often, quality has dropped.
• Corrections after sending: your quality gauge. If this rises, the check is too light.

Record the baseline before you start. Otherwise the evaluation becomes a discussion about feelings. How to set up that calculation is in calculating the business case for AI.

When it doesn't pay off

Already working with a configurator that produces a quote in a few clicks? Then there's little to gain. The gain is in quotes with a lot of variation, not in repetition.

Is your pricing structure not fixed, and do you set prices per customer on gut feeling and history? Answer that question first. Automating a structure that doesn't exist only creates chaos. That applies more widely; see which business processes are suitable for AI.

Set it up yourself or have it built?

For the simple version you don't need a build project. A shared assistant in a business AI subscription, fed with your text blocks and house style, turns a request into a draft. Most of the time goes into getting the blocks in order. See ChatGPT for business.

An integration that picks up requests automatically, pulls data from the CRM, fetches prices and puts the draft back in your quoting system is a real project. It only pays off with a substantial number of quotes a week. What such a solution costs is on AI solution costs.

Not sure whether your quoting process is the best place to start? In the AI Opportunity Audit we compare it with your other processes.