AI Chatbot for a Construction Company: Quote Requests and Service Areas

A construction company's chatbot should answer four things from approved fact sheets: where you work, which project types and sizes you take, how to request a quote and what happens next, and which roles are open. It should hand over for prices, tender documents, questions about live jobs and anything contractual, after capturing the basics.

Unlike a one-van trades business, a construction company gets several kinds of visitor in the same inbox: homeowners wanting an extension, developers and property managers with a brief, main contractors sending tender invitations, job seekers, and suppliers. A bot that treats them all as "a customer wanting a quote" will give the wrong answer to most of them. The first design decision is therefore routing, and the second is that the bot never sets, hints at or promises a price.

Follow me on Instagram@sagnikteaches

Write the bot's knowledge as four short fact sheets

A chatbot is only as accurate as the text behind it, and construction websites tend to hold that text in marketing language: "we work across the whole area" or "projects of every size". Replace it with four one-page fact sheets written in the visitor's words, each with a named owner and a review date. The bot answers only from these sheets. A question that no sheet covers gets "I don't know, but I can pass it to someone who does".

Connect on LinkedInSagnik Bhattacharya

The owner column is the part firms skip, and it is why chatbots go stale. A role filled in March that is still "open" in June sends a candidate to a closed door. If you want the wider method for turning company knowledge into approved answers, building the FAQ your chatbot needs before launch covers it. This tutorial applies it to the four sheets a construction company needs.

Subscribe on YouTube@sagnikteaches

Service areas: zones, edge cases and a town you have never heard of

"Do you cover my town?" sounds like a yes-or-no question, and a chatbot will treat it as one. In practice a construction company's coverage has layers: places it works routinely, places it will travel to for the right project, and places it will not go. Write those layers as zones, each with a list of place names. A radius ("within 40 minutes") sounds exact, but a bot cannot measure it and will guess from a name. This is a filled-in sheet for an illustrative mid-sized building firm, with bracketed names standing in for your own towns and every figure a placeholder.

The most important rule on that sheet is that a place not on the list is never an automatic no. A firm may well travel 90 minutes for a large commercial fit-out, and a bot that says "sorry, we don't cover that area" has thrown the lead away at the door. It says it cannot tell and lets the estimator decide. An illustrative exchange:

Visitor: Do you build extensions in [Town X]?
Bot: I don't see [Town X] on my list of places we work, so I can't say yes or no from here. Larger projects sometimes take us further out. If you tell me what you are planning, I'll pass it to our estimator, who will reply to say whether we can help. Is it an extension, a conversion or something else?

Two design choices stop most coverage errors. First, ask the visitor to pick the town from a list of your zone towns plus "somewhere else" wherever the software allows it, because free-text matching trips on misspellings, abbreviations and two places sharing a name. Second, make the sheet the only source. A typical failure is a line on the About page such as "serving the whole region and beyond". The bot reads the page, repeats the line to a visitor two hours away, and the visitor arrives at the quote stage believing coverage was confirmed. Remove or reword the line, and tell the bot that the zone sheet overrides any wording elsewhere on the site.

Before launch, test with ten place names: three from Zone 1, two from Zone 2, two just beyond the edge, one misspelt, one that exists twice, and one village inside a Zone 1 town's catchment that is not on the list. Expected answers are "yes" for the first three, "larger projects" for the next two, "can't tell, taking details" for the edge cases, "did you mean...?" for the misspelling, "which one?" for the duplicate, and, for the village, "can't tell, taking details" plus a note to add the village to the list if you do serve it. Re-run the ten whenever the list changes.

Project types: what you build, what you decline and the size you take

The second sheet answers "do you do X?" and matters more than it looks, because construction websites mix the work a company does itself, the work it subcontracts and the work it once did. A bot reading the whole site cannot tell the difference. This is an illustrative sheet for a general building firm. Replace every row with your own.

Take the roofing row. Imagine a firm that fitted a new roof once as part of an extension and put the photograph on a case-study page. A bot reading the site tells a visitor "yes, we do roofing", and the estimator spends a morning on a job the firm has no crew for. Sheet wording such as "roofing only as part of a larger project, never on its own" fixes it, but only if the bot is told to read the sheet first and the site second. The emergency row has fixed wording of its own: "We don't offer emergency call-outs. If anyone is in danger, please call the emergency services."

The decline needs wording as much as the yes does: "Thanks for the details. We only take on roofing as part of a bigger project, so we can't help with this one. If you're planning an extension or a refurbishment, we'd be glad to hear about it." It puts the refusal down to the firm's scope rather than the visitor's project, and invites the larger job. Publish a minimum project value only if you want visitors to see it. Otherwise let the bot pass small jobs on with "I'll ask the estimator whether this is something we can take", which is slower for the estimator but loses fewer good small jobs that lead to larger ones.

Quote requests: explain the steps, collect the brief, name no price

Visitors ask three things about quotes: how it works, how long it takes and what it will cost. The bot can answer the first two from a card the estimating manager writes, and must decline the third. Here is the card, with brackets for the firm's real figures. Use only turnaround times you actually keep, because the bot will repeat them to every visitor.

HOW WE QUOTE (approved wording)
1. You tell us about the project: what, where, any drawings, timing.
2. For anything beyond a small repair we visit the site first
   (about [one hour]).
3. We prepare a written quote that lists what is included and what
   is not. Allow [10 working days] after the visit for a project
   of this size.
4. The quote is valid for [30 days].
5. If you accept, we agree a start date once materials are confirmed.

The price rule is simple and absolute: no figure, no "from", no "roughly", no rate per square metre, and no "usually costs" unless your own website already publishes that exact sentence. The reason is practical. A price from a chatbot is a price from your company, and the estimator's figure arrives later, after the ground conditions, the steel beam and the specification are known. An illustrative exchange with a visitor who pushes:

Visitor: Just roughly, what does a single-storey extension cost?
Bot: Size, ground conditions, structure, specification and finishes all change it so much that any figure I gave would mislead you, and I'd rather not set an expectation we can't meet. What I can tell you is how we quote: we visit, then send a written quote that lists what is included. Would you like to send the details so our estimator can arrange a visit?

Compare the ungrounded version. It answers with a range it picked up from a general web page. The visitor writes the range down, the estimator's figure comes in 40% higher because of the ground conditions and a steel beam, and the first call is a negotiation about a number nobody at the firm gave. The 40% is an illustration, but the pattern is common enough that it is worth a rule. If a visitor asks a third time, the bot hands over to a person and says so, without apologising for the firm's process.

The fields every enquiry should arrive with

A bot that only says "send us your details" gets "Hi, need a quote, call me". Required fields, asked one at a time with "skip" available for the optional ones, produce a record the estimator can use. Use the same fields on your website form so every lead lands in one format.

A completed record, as it lands in the estimating inbox (illustrative):

NEW ENQUIRY  Construction  Zone 2  Priority: normal
Visitor: [first name], prefers email, weekday mornings
Property: detached house, [Town F], owner-occupier
Project: single-storey rear extension with a new kitchen
Stage: drawings done, permission granted two weeks ago
Timing: wants to start in spring; no hard deadline
Budget (visitor's words, unchecked): "about [amount], flexible"
Documents: drawings uploaded via the link (4 files)
Asked and declined: a price; told how quotes work and what follows
Next step offered: site visit; estimator to reply by the end of
  the next working day

Say the firm gets 40 web enquiries a month and each used to cost the estimator a 12-minute call to establish the basics, which is 480 minutes. With a record like the one above, the estimator reads each in 3 minutes (120 minutes) and calls only the 25 enquiries that fit, for 6 minutes each (150 minutes). That is 270 minutes against 480, a saving of 210 minutes, or three and a half hours a month. All figures here are assumptions. The saving is real but modest, and the larger gain is that enquiries arriving at 9 pm get a complete record and a promised reply time instead of a voicemail. For routing phone calls to the same estimators, AI receptionist costs and call routing for construction companies covers the other front door.

Six kinds of visitor, six routes

Most of the confusion on a construction website comes from treating every visitor as a homeowner wanting a quote. The bot needs to recognise who it is talking to in the first two messages, and the table below is the routing sheet. Anything it cannot place goes to the office with the transcript attached.

Tender invitations deserve the most care, because they are the most valuable messages the chatbot will ever receive and the least likely to be typed into a chat window. The bot's job is to avoid losing them: capture the return date first, because it decides everything, and pass them to a named person the same day, by phone notification if the return date is close. Treat any tender documents as confidential, use your own storage for uploads, and never let the bot read them to judge whether you will bid. That decision belongs to the estimator, and a bid/no-bid screen is the next step once the invitation has arrived.

The last row needs a firm rule. An existing client asking about their own job is the visitor the chatbot is least equipped for, and one that cares most about a wrong answer. The bot never shows job details, never speculates about dates or delays, and never discusses anyone else's job. It takes the name and job reference and says the project manager will call. Status answers from a sheet are possible for some trades, as in seven contractor chatbot jobs, but on larger construction projects the programme changes daily and a person should answer.

Careers: roles open now, how to apply, what the bot never discusses

Careers questions are different from customer questions because the person typing may be deciding where to work. The sheet is short and is owned by whoever does the hiring. It holds the roles open now with an "as of" date, how to apply, and apprenticeships if you run them, which an apprentice training plan can describe in more detail. An illustrative exchange:

Visitor: Are you hiring bricklayers?
Bot: Our careers sheet shows an open bricklayer role as of [date]. The advert has the details and the way to apply: [link]. I can't discuss pay beyond what the advert says, and applications are read by [role]. Shall I ask them to get in touch?

Four things the bot never does here. It never promises an interview, a reply date or a start date. It never discusses pay or terms beyond the published advert. It never asks about age, health, family plans or background, and it never reacts to what a candidate volunteers about them. And it never screens, ranks or comments on candidates: it points to the application route and passes the conversation on. If you later add AI to hiring decisions, that is a separate project with its own rules and a fairness check, so keep it out of the chat.

The "as of" date is the safeguard. When a role is filled, the owner removes it that day, and a bot answering from a sheet dated three months ago is visibly out of date. If the sheet is old, the bot's reply should say "I don't have a current list, so I've asked [role] to get back to you".

The minute after "I'll pass this on"

The handover is where most chatbots lose the value they created. A bot that says "someone will be in touch" and a team that does not check the inbox until Monday is worse than no bot, because the visitor was promised a reply and then waited. Four rules keep it honest.

1. Promise only a time you meet. If the estimating team replies by the end of the next working day, the bot says exactly that and never "shortly" or "within the hour".
2. Alert a named person. The record goes to a named estimator or a monitored shared inbox, with a notification to a phone for tender invitations, and with a deputy for holidays.
3. Say what happens to the details. One line at the start of the capture: "I'll use the details you give me to reply about your enquiry." Check what your own data-protection rules require for the information you collect, and decide how long transcripts are kept.
4. Track the reply. Add one tick box to the record: replied within the promised time, yes or no. It is the single number that shows whether the chat is helping or hurting.

The general rules are in when an AI chatbot should hand over to a human. The construction-specific addition is that handover has several destinations, and the visitor type decides which. A tender invitation that lands in the quote queue and waits for the weekly estimating meeting is a lost tender, and that is a mistake you only make once.

Twenty-five test messages before the bot meets a customer

Write 25 test messages: fifteen drawn from real enquiries in your inbox (copy the wording, mistakes included) and ten from the traps below. The chat's greeting does two jobs at once, admitting it is automated and saying what it covers, for example "This is [company]'s automated assistant. It can tell you where we work and what we build, and pass your project to the right person, but it can't give prices." Score the run on three counts: zero answers containing a price, a coverage promise or a start date; at least 9 in 10 messages routed to the right person; and a complete handover record every time.

Repeat all 25 after any change to a sheet, a rule or the software, since tightening one answer often loosens another. Picture a software update that alters word matching so that "tender" no longer raises the same-day flag: a repeat run shows it within the hour, while skipping it means learning from a main contractor who wants to know why nobody replied.

After launch the routine is short. Once a week, go through every handover record plus a sample of about 20 other chats, and list the questions that left the bot stuck. The owner of the relevant fact sheet turns each into an added line, a routing change or a firmer refusal, and dates the change. The same owners review their sheets every quarter, and a change to your price list, your zones or your vacancies means a change to the bot on the same day. Keep that habit and the chatbot stays an accurate front desk. Skip it and it will go on cheerfully describing a company that no longer exists.

Questions construction firms ask about website chatbots

What should the bot say when visitors insist on a price?

Explain that a price depends on drawings, access, specification and timing, then offer what you can: how the quote is prepared, what it will include and when the estimator replies. Do not offer a 'from' figure unless your website already publishes one. Ask one easy question about the project to keep the chat moving, and hand over if the visitor presses a third time.

Can the bot accept plans or tender documents?

It can point visitors to a secure upload, capture the return date, and pass a notice to a named estimator. Treat drawings and tender packs as confidential: use your own storage and avoid leaving them in a chatbot vendor's chat history unless you have checked how long it is kept. The bot should never read them to judge whether you will bid.

How do we stop the bot claiming we cover places we don't?

Give it an explicit list of the place names you serve, split into zones, and tell it that anything not on the list means 'I can't tell from here' and a hand-over, never a yes or a no. Then test it with ten towns, including near misses and misspellings, and remove any marketing line such as 'we cover everywhere nearby' that contradicts the list.

Should the chat replace our quote form?

Usually it should sit in front of it. Use the same required fields in both so every lead lands in one format, and keep the plain form for visitors who dislike chat. If your form already produces complete enquiries, the chat adds little. If estimators keep phoning back for the same missing details, the chat's follow-up questions are the reason to add it.

Further reads

• AI Chatbot for CRM Systems: Send Website Enquiries to Job Software — Send chat enquiries into your job software.
• How to Measure Whether Your AI Chatbot Is Actually Working — The numbers that show whether the chatbot is working.
• How to Test a Customer Chatbot Before It Goes Live — A general method for testing a bot before customers see it.
• How to Add an AI Chatbot That Captures Leads on Your Website — Adding lead capture to a website, step by step.
• AI Marketing for a Construction Company: Case Studies, Bids and LinkedIn — Website case studies a chatbot can point visitors to.
• AI Chatbot Disclosure: What to Tell Customers at the Start of a Chat — Wording for telling visitors they are talking to AI.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.