Automating quotes with AI: from enquiry to a draft ready to send

Why a quote takes so long

Ask a salesperson how long a quote takes and you hear half an hour. Measure the time between the enquiry arriving and the quote going out and you often land on days. The writing is not slow; the waiting is:

• Looking up prices: a product price from the latest list, an hourly rate that differs per customer, a surcharge agreed once and written down nowhere. And digging out what you charged for a comparable job last year.
• Waiting for the estimator. The enquiry arrives at sales, and the hours come from somebody out on site.
• Chasing and retyping. The customer did not say how many units. Then the same details sit in the email, the package and the CRM.

Template, configurator or AI: three different things

These three get used interchangeably while they solve different problems.

• A template is fixed text with fields to fill in. It costs almost nothing and works as long as the enquiry fits. It only saves on layout: somebody still fills in and looks up everything.
• A configurator is a set of rules: choices in, price out. Excellent when your offering fits into options — a machine with variants, a window in sizes, a subscription with bands. It always calculates the same way, and that is its value. But every exception has to be written out up front.
• AI reads the enquiry as the customer wrote it — an email in their own words, a PDF specification, a phone note — and turns it into a draft. That works where no list of options exists. But it does not calculate: the amounts come from your own system.

What the chain looks like in practice

It is not “AI writes the quote”, but a chain in which two steps use AI. The rest is integrations and rules.

1. 01Enquiry arrives. Email, web form, PDF specification or a phone note land in one place. As long as enquiries sit in five inboxes, there is nothing to automate.
2. 02Extract the details. This is where you use AI: customer, location, what is asked for, quantities, date. Anything missing is flagged as an open question, not filled in with a guess.
3. 03Pull in the customer and history: agreed terms and earlier quotes from your CRM. An integration, not a language model.
4. 04Fetch prices from your own system: product prices, volume bands, rates and customer discounts from the source where they already sit. Fixed, checkable arithmetic; the model does not touch the amounts.
5. 05Produce the draft in your own tone, with the assumptions and open questions at the top. That list matters more than the text below it.
6. 06A human checks and sends, looking at price, margin, tone and exceptions. Not an interim step you strip out later.
7. 07Follow-up: a reminder after an agreed number of days, and a recorded reason when you lose one. The dullest part, often the most rewarding.

Three examples: installation, wholesale, construction

At an installation company the knowledge sits in job sheets and costings from earlier work. A draft can take the hours and materials of a comparable job as its starting point and pull the rates from the package; the work planner corrects it. Never let a model estimate hours — make it name the job the figures came from.

At a wholesaler it is price lists and volume bands. A customer asks for forty units while the band starts at fifty. The system may flag that, but not stretch the band.

At a construction company the money is in variations. They get agreed verbally on site and go on paper weeks later, if at all. From a message from the site manager you can have a draft variation order produced, referring to the item in the contract. Not a saving on typing, but work that actually gets invoiced.

Why the human sets the price and the tone

A quote is legally an offer: whatever is in it, the customer can accept. Different from an internal summary where a slip has no consequences.

The price is a decision rather than a sum. Whether you badly want this customer, whether a competitor is looking, whether you have room in March but not September — a person weighs that. A system that produces the cost price and standard margin is useful; one that sets the selling price decides what it cannot justify.

The same goes for tone. A draft that reads too smoothly is a risk: it sounds certain about lead times nobody has looked at. So the assumptions sit at the top, and nothing goes out unless somebody puts their name to it.

Where it goes wrong, and when not to do it

Automating quotes rarely fails on the technology. This goes wrong in practice.

• Prices that are not right. A list that was not updated, a band applied wrongly, a discount that only applies to three customers. A wrong amount costs you money or the job, and you notice late.
• A draft that looks too good. The most dangerous one: a neatly formatted quote invites somebody to forward it unread. The mistake is not in the layout but in an assumption nobody checked.
• Exceptions that are half the work. If nearly every enquiry is different, you automate the minority and keep the hard cases by hand — while still maintaining a system.
• Incomplete enquiries. The real work is often ringing to ask what is not there. A better enquiry form helps more than AI does.
• More quotes, not more orders. Quoting faster means you say yes to enquiries you used to let go. If your win rate drops, you work harder for the same result.

What it costs, and what it connects to

We do not replace your packages: the quote should come out of where your administration already sits, or you get a second version of the truth. Exact, AFAS, Moneybird and HubSpot offer integration options, but what your version and licence allow differs and changes. Check with your supplier, or let us find out first.

Our rates are the same for everyone. What differs is the hours: how many systems are involved and how clean your prices are. So we give no fixed project price before we have watched the work.

• AI scan: free, five minutes online, with a score and a price indication.
• Core session on your shop floor: a half-day (4 hours), €596 excluding VAT and travel (€0.45 per kilometre). Full day €1,192; advisory rate €149 per hour.
• Building: €110 per hour for straightforward work, €165 for complex development. A fixed number of hours per month, paid in advance from an AI budget or build deposit.
• Monthly running cost: the model cost scales with the number of enquiries, plus licence or integration fees from your own package.
• Preparation on your side: cleaning the price list, merging duplicate product codes, making customer agreements explicit. Usually the largest part of the first weeks.
• Maintenance: prices and terms change, a supplier adjusts an integration. Budget a few hours a month.

Start with one type of quote

The most common mistake is wanting all of it at once: six types of quote, so a system for six.

1. 01Pick the type that occurs most often and varies least. Not the biggest, but the dullest.
2. 02Measure one number first: the days between the enquiry arriving and the quote going out. Without it you will never know whether it helped.
3. 03Walk twenty enquiries through the intended route by hand. How many genuinely fit? That is your answer to the exceptions question.
4. 04Get the price source in order: one place with the current prices, bands and customer agreements. Dull, and the most important step.
5. 05Build the chain for that one type, with the human check firmly in it. Do not drop it “because it goes fine anyway”.
6. 06For the first weeks, have every quote reviewed and record what gets changed. That is your improvement list. Only then the second type.

Frequently asked questions

What we do for this

Further reading

Other topics

Start with one type of quote

Take the free five-minute AI scan, or book a core session on your shop floor. We follow one type of quote from enquiry to sending and tell you honestly what there is to gain — including when that is a better enquiry form.