A plumber calls back a website enquiry. "Hi, you asked about a bathroom?" Then comes the interview. Is it a full refit or just the suite? Upstairs or downstairs? Do you want the tiles done too? Is the room empty? Where are you based? Fifteen minutes later, they have enough to say "roughly this much, I'd need to see it", and the customer has already had two other trades ring with the same questions.
The callback was not a quote. It was data entry done over the phone.
An AI assistant for quotes flips the order. The visitor describes the job in the chat, the assistant asks the questions you would have asked, and the request lands with you already specified. Your callback opens with a number, or at least a sensible range and a date to visit.
The arithmetic of the callback interview
This is worth doing on the back of an envelope with your own figures. The numbers below are an example; replace them.
Say you get 25 quote enquiries a week. Each callback runs 15 minutes, of which about 10 minutes is collecting details. That is 250 minutes a week, just over four hours, spent asking the same questions. Some of those calls go to voicemail and need a second attempt, so the true figure is higher.
If the details arrive in writing, the call shrinks to 5 minutes of discussing the actual job. You get roughly three hours back each week. More usefully, you can call the promising jobs first, because you can see which ones they are before you pick up the phone.
The chatbot ROI calculator will do a fuller version of this sum if you want to include missed enquiries too.
Find the questions that change your price
The mistake is asking everything. A quote request chatbot that fires twelve questions feels like a tax return. The trick is to find the handful of details that actually move the number, and skip the rest until the call.
Ask yourself: if I knew only five things about this job, which five would let me quote? Here is what that tends to look like in a few trades.
Electrician
Type of work (fault, new circuit, rewire, EV charger), property type and age, number of rooms or points involved, whether the consumer unit has been updated, and postcode.
Landscaper
What they want done (patio, turfing, fencing, full redesign), rough size of the area in metres, current state of the ground, access for machinery (side gate width matters), and when they want it finished.
Domestic cleaner
One-off, regular or end of tenancy, number of bedrooms and bathrooms, how often, whether pets live there, and preferred days.
Painter and decorator
Interior or exterior, which rooms or elevations, whether ceilings and woodwork are included, condition of the surfaces, and whether the rooms will be furnished.
Postcode or town comes up in almost every trade. If you cover a defined area, the assistant can tell people outside it straight away, which saves both of you a wasted call. Answering service area questions covers how to phrase that.
What the conversation should feel like
The assistant should talk like a person taking a job over the counter, not a form with a chat skin. One question at a time, acknowledging what the visitor said, and accepting "not sure" without fuss.
Visitor: Hi, how much for a new fence?
Assistant: Happy to help get you a quote. Fence prices depend mostly on length and type, so a couple of quick questions. Roughly how long is the run, in metres or number of panels?
Visitor: About 8 panels I think
Assistant: Thanks. Is this replacing an existing fence, and do you know if the posts are concrete or wooden?
Visitor: Replacing. No idea about the posts.
Assistant: No problem, that can be checked on a visit. What's your postcode, and what's the best email or number for the quote?
Notice it did not give a price. For work that depends on what is on the ground, it should not. If you publish fixed figures, like a minimum charge or a survey fee, the assistant can repeat those. Everything else ends with captured details. What an AI receptionist must never promise is worth reading before you write its instructions.
Setting it up
- List your price-changing details. Four to six per type of job. If you offer several services, write a short list for each.
- Write them into the assistant's instructions. Plain language works: "When someone asks for a quote on fencing, ask for length, whether it replaces an existing fence, postcode and contact details, one question at a time."
- Teach it your fixed facts. Service area, call-out or survey fees, lead times, and anything you never do. This stops dead-end requests.
- Switch on lead capture. The captured enquiry should hold the job details, not just a name and number.
- Send it somewhere you will see. A webhook can push each request into your CRM, a spreadsheet through Zapier or Make, or a Slack channel your team watches. See integrations for the options.
- Test with last month's enquiries. Type in real requests you received and check whether the result gives you enough to quote.
The sixth step catches most problems. You will quickly see where a question is missing or where the assistant asks something you never use.
Where this does not work
Some jobs cannot be specified in a chat. A damp problem, a structural crack, a boiler making a noise: the customer does not know what they are looking at, and neither will the assistant. For those, the useful outcome is a site visit request with an address, availability and a description, not a pseudo-specification.
It also does not replace speed. A beautifully detailed request that sits in an inbox until Thursday is still a lost job. Decide who checks new requests and when, and read following up on chatbot leads for templates.
Your next step
Take a sheet of paper and write down the questions you asked on your last five quote callbacks. Cross out the ones that did not affect the price. What remains is your assistant's question list. Put it into the instructions, run through three test enquiries yourself, and switch it on for your quote or contact page first.