On a contractor's website an AI chatbot can realistically do seven jobs: turn vague requests into a job brief, screen by job type and area, book site visits, explain how a job runs, update existing customers, hand out insurance and accreditation papers, and log aftercare problems. It cannot price unseen work, judge a defect or promise dates.
AI chatbot technology is, underneath, a language model reading your approved answers and the visitor's message, so what it can do depends on what you connect it to and what you forbid. The seven jobs below share one trait: each ends with a person holding better information than before, or a customer holding a fact the firm has already approved. None ends with the bot making a commitment. Generic lists of chatbot uses describe customer service in the abstract. A contractor's site has its own visitors: people with a leak tonight, people planning a project for spring, past customers chasing a snag, and property managers who need a certificate before Monday.
1. Turn "I need a quote" into a job brief an estimator can use
A visitor who types "need a quote for my roof" has told you almost nothing, and the phone call that follows is ten minutes of questions you ask every time. A chatbot can ask them first and hand over a brief. For a roofing contractor the questions are specific: what is wrong or wanted (leak, missing tiles, full re-cover, flat roof), roof type and rough age, number of storeys, whether scaffolding can reach from the street, how urgent it is, and three photos: a wide one from the road, a close one of the damage, and the underside in the loft if water is coming in. Say the firm gets 30 web enquiries a month and the first call used to run 12 minutes of questions. With the brief in hand it becomes a 4-minute confirmation, which saves 8 minutes an enquiry, or 240 minutes a month. These figures are assumptions for the sum.
This is what the estimator receives, illustrative and filled in:
NEW ENQUIRY Roofing Priority: leak (same week)
Customer: [first name], prefers text message
Roof: pitched, concrete tiles, about 25 years old, two storeys
Problem: water mark on a bedroom ceiling after heavy rain,
noticed 2 days ago
Wanted: repair, or advice on a full re-cover
Access: attached house, street scaffold likely, side passage blocked
Photos: 3 received (road view, ceiling stain, loft underside)
Budget stated by customer: not stated
Timing: wants someone this week
Told by the bot: an estimator will contact you [today]; no price given
Two design choices matter. First, answers come from buttons wherever possible: "under 10 years, 10 to 30, over 30, not sure" gets a usable answer where a free-text box gets "old". Second, anything the customer did not say reads "not stated", never a guess, because a brief that fills gaps looks complete and sends the estimator to the wrong job. Keep the question count to six to eight, and let people skip. The full intake script for this job is in capturing job details before you call back. The limit is that photos are for the estimator: the bot never says "that is just condensation" or "that looks fixable".

The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy now2. Say no early: screen by job type, size and area
Every contractor has jobs they do not take: under a minimum value, outside the working area, trades they do not offer, or work that needs a licence they do not hold. Each wasted site visit costs a morning. A chatbot can say no early and politely, but only on rules the firm has written down. This is the rule sheet a groundworks and landscaping contractor might hand it (illustrative).
| Rule | Written as | The bot's move |
|---|---|---|
| Job type | "We lay driveways and patios. We do not fit artificial turf or fencing." | Declines turf and fencing with the polite wording below |
| Minimum value | "We do not take jobs under about $3,000." | Declines, but only if the firm chose to publish a minimum. Otherwise it passes the enquiry on |
| Working area | "The towns on this list, plus anywhere within a 40-minute drive of the yard for jobs over $15,000." | Looks the town up in the list. Unknown towns go to a person |
| Timing | "The next available start is about eight weeks out." | States the lead time from one line the firm updates every week |
Those four rules do the work of a first phone call, and the wording of the refusal matters as much as the rule. For example: "Thanks for the details. Artificial turf is not something we fit, so we would be the wrong firm for this one. If you have a patio, driveway or groundwork project, we would be glad to hear about it."
Borderline cases go to a person: a job just under the minimum, a town just outside the list, an unusual material. The bot declines only on what is written down. For a "maybe" it captures the brief and says someone will reply, because the right answer is sometimes yes. Now the sum. Suppose six enquiries a month fall outside the rules, and each one gets a site visit of two and a half hours including travel because nobody checks until the visit. If the bot stops four of them, that is 10 hours a month, or 600 minutes, and at an assumed $45 an hour it is $450 a month. Plumbing firms use the same pattern, covered in answering questions, collecting photos and refusing jobs.
3. Book the site visit with the right person and the right notice
Booking a visit is more than offering a slot. The length of the visit, who attends and what the customer must have ready all depend on the job, and a chatbot that holds the table below can set expectations before anyone drives anywhere. These are example rules for an electrical contractor, so replace the numbers with yours.
| Job | Visit length | Who attends | Customer has ready |
|---|---|---|---|
| Extra sockets or lighting | 30 minutes | One electrician | Rooms marked, furniture clear of the walls |
| EV charger point | 45 minutes | One electrician | A photo of the main electrical panel (fuse board) and a chosen charger location |
| Full rewire | 90 minutes | Estimator and electrician | Loft and under-floor access, a room list, and whether the house will be empty |
The bot offers two arrival windows of two to three hours from the diary, not exact times, because the visit before may overrun. It then sends a confirmation that lists what to have ready, a reminder the day before with a one-tap reschedule, and a note to the estimator with the brief from job one.
Here is how it breaks. A bot offered Monday at 9 am because the electrician's calendar showed the previous job ending at 8:30, and that job was a 40-minute drive away. The fix is a travel buffer built into the calendar itself, 45 minutes to start with for local work, so the bot cannot offer back-to-back visits however the slots are generated. The bot books surveys and estimate visits only. It does not book the work, because that depends on an accepted quote and materials being ordered. The calendar and job-software side of the flow is laid out in qualifying, booking and confirming site visits.
4. Explain how a job runs without promising dates
"What happens after you quote?" is one of the commonest questions a home-improvement contractor gets, and it is where a friendly chatbot does the most damage by being reassuring. Left to improvise, it says "we can usually finish a kitchen in two weeks", and the firm has just made a promise it never wrote. The cure is a process card, written once, that the bot quotes. This one is for a kitchen and bathroom fitter, with brackets for the firm's own figures.
HOW A KITCHEN JOB RUNS (approved wording)
1. Survey: we visit, measure and talk through layout and appliances.
About [60] minutes.
2. Quote: written, valid for [30] days, covering what it lists.
3. Deposit and ordering: your deposit confirms the booking and lets
us order units and worktops. Lead times vary by supplier and are
stated in your quote.
4. Start date: confirmed once the materials are on site.
5. On site: usually [number] working days for a kitchen of this
size, plus finishing visits.
6. Sign-off: a walk-through, a snag list, then the final payment.
| The customer asks | Approved answer in brief | The bot never |
|---|---|---|
| "How long will it take?" | On-site time from step 5, and the start date depends on supplier lead times shown in the quote | Gives a finish date, or says "usually two weeks" without the card |
| "Do I pay everything up front?" | The stages in your written terms, with amounts in the quote | Invents a percentage or says "it's flexible" |
| "Can I change something halfway?" | Changes are priced in writing before the work is done | Agrees to a change or names a price for it |
| "Do I need to be at home?" | The access arrangements your rule sets for each stage | Guesses from what other trades need |
One rule catches most slips: no sentence containing a duration, a date or a price unless it appears in the approved card. Test it by asking the bot the same question six different ways, and keep loose promises off your marketing pages too, because the bot treats everything on the site as its own voice.
5. Answer "where are you up to?" from a status sheet
Customers with a job under way ring the office to ask what is happening this week. A chatbot can answer, provided it reads from a record the site manager keeps and not from inference. The simplest record is a sheet with one row per live job, updated every Friday. This is a row for a refurbishment (illustrative).
| Job ref | Stage | This week | Next week | Waiting on customer | Planned completion | Updated |
|---|---|---|---|---|---|---|
| R-2417 | Second fix complete | Plastering and tiling | Decorating starts | Paint colours by Wednesday | Week of 23 November (planned) | 6 November |
The bot's answer to "where are you up to?" comes straight from that row: "As of 6 November, second fix is complete. This week is plastering and tiling, and decorating starts next week. Planned completion is the week of 23 November, which is a plan and not a promise. One thing is waiting on you: paint colours by Wednesday. Would you like the site manager to call you?"
- Every answer carries the "updated" date, and a sheet older than seven days produces "I don't have a current update, so I have asked the site manager to call you".
- Dates are always "planned", never "promised".
- The bot checks the job reference and the surname on the contract before showing any detail, and never discusses another customer's job.
- A status flag of "delayed" sends the conversation to the site manager, so bad news arrives from a person.
The arithmetic is short. Take 15 live jobs that bring about 60 status calls a month at six minutes each, which is six hours of office time. If the bot answers six in ten from the sheet, that returns 3.6 hours. Firms that send weekly written updates to buyers will recognise the same input, and tools and cost for weekly buyer progress updates shows how to produce it.
6. Send current insurance and accreditation papers on request
Property managers and commercial clients ask for the same documents before a contractor starts: a current insurance certificate, proof of accreditation and the firm's standard safety paperwork. They tend to ask late on a Friday for a Monday start. A chatbot can send these from a controlled folder at any hour, which makes this one of the simplest wins for a firm that does commercial work. The folder is described by a register like the one below, written for a commercial painting and decorating contractor (illustrative).
| Document | Bot may send | Expires or reviewed | Owner |
|---|---|---|---|
| Liability insurance certificate | Yes, once the requester gives a name, company and email address | Expires 31 March 2027 | Office manager |
| Employers' liability or workers' compensation certificate | Yes, same rule | Expires 31 March 2027 | Office manager |
| Trade accreditation proof | Yes | Reviewed every 12 months | Director |
| Health and safety policy statement | Yes | Reviewed every 12 months | Director |
| Job-specific method statements and risk assessments | No: a person writes and sends them | Per job | Contracts manager |
The expiry column is the point. Every firm with a document folder meets this failure once. A policy expires on 31 March and renews on 1 April, the new certificate arrives by email, and nobody replaces the PDF in the folder. In April the bot sends the old file under the words "our current certificate", and the client's compliance team either rejects it or, worse, assumes cover that nobody checked. The fix has three parts. The bot reads the expiry date from the register and not from the file name. It refuses to send anything expired or within 14 days of expiry, and hands the request to the office instead. And the owner replaces the file on the day the renewal arrives.
A good reply is short: "Here is our current liability insurance certificate, valid until 31 March 2027." Log every send with the requester's name, company, document and time, because clients sometimes ask who has received a copy of their paperwork. Job-specific method statements stay with a person, since they describe how this firm will do this job in this building.
7. Log snags and aftercare problems with photos and urgency
After a job closes, customers write when something looks wrong: a cracked tile, a gap between boards, a squeak, damp at a skirting board. The chatbot's job is not to explain it. It is to log it so the aftercare coordinator can act without a second call: job address, completion date from the record, location in the room, when it was noticed, and photos with a coin in the frame for scale. Three categories decide the response, shown here for a flooring and tiling contractor (illustrative).
| Category | What customers say | Bot does | Goes to, and when |
|---|---|---|---|
| Urgent | "water coming up through the floor", "a tile edge cut my foot", "floor lifting, someone tripped" | Shows the firm's approved urgent message and alerts the coordinator by phone or text | A person phones back within [your promised time] |
| This week | "tile cracked", "grout missing", "board has lifted at the door" | Logs details and photos, tells the customer a person will review | Coordinator, within [number] working days |
| Routine | "slight gap between boards", "squeak", "small chip" | Logs it and thanks the customer | Weekly aftercare round |
AFTERCARE LOG Category: this week
Job: [job ref], completed 14 May (from the record)
Customer: [first name], prefers phone
Issue: two tiles cracked near the bathroom door, noticed last Tuesday
Location: bathroom floor, doorway end
Photos: 2 received, coin in frame
Changes since the job: none mentioned
Told by the bot: a person will review within [number] working
days; no view given on cause or cover
The bot never says whose responsibility something is, never says "that is normal movement", never promises a repair or a date, and never says whether the job is under warranty. It may state a fact from the record, such as the completion date. Turning a log like this into a numbered list for the site team is covered in writing client updates and snag lists with AI.
What to leave to a person, every time
The seven jobs work because they stay on the safe side of a line. This table is the line, and it is worth printing next to whoever configures the chatbot.
| Topic | Why the bot must not take it | What the bot does instead |
|---|---|---|
| A price for work nobody has seen | Access, condition and materials decide it, and a figure from a bot gets quoted back at you | Explains how and when quotes are given, and repeats a published "from" price with its conditions if one exists |
| Whether something is safe or structural | It needs a site visit and a qualified person | Hands off at once and says no more |
| Design and specification choices | They depend on the building and carry responsibility | Hands off to the estimator |
| Extra work ("while you are here, could you...") | A variation changes price and timing | Logs the request for the site manager and never agrees |
| Complaints, disputes and payment arguments | Tone and judgement matter | Hands off to a director or manager, with no admission and no denial |
| Injury, damage, claims or legal threats | Insurance and legal consequences follow | Hands off at once and stops answering |
| Dates beyond the status sheet | A date from a bot becomes the customer's expectation | States what the sheet says and who will call |
Every hand-off uses one plain line: "I am passing this to [role], and they will reply by [a time you really keep]". Promise only times the firm already meets. Whether a bot's statement can bind your firm is a question for your adviser, and who is liable when your chatbot gets it wrong sets out the issues. The practical protection is the same either way: the bot only says what you have approved.
Which of the seven to switch on first
Rank the jobs by minutes returned and data needed, using your own numbers. This is the arithmetic for a six-person building firm, with every figure an assumption to replace.
| Job | Times a month | Minutes saved each | Minutes a month | Data needed |
|---|---|---|---|---|
| 1. Job briefs | 30 | 8 | 240 | Your question list |
| 2. Screening | 4 visits avoided | 150 | 600 | Written rules |
| 3. Site-visit booking | 20 | 5 | 100 | A shared diary |
| 4. Process questions | 25 | 4 | 100 | One approved card |
| 5. Job status | 36 of 60 calls | 6 | 216 | A weekly status sheet |
| 6. Paperwork requests | 12 | 10 | 120 | A document register |
| 7. Aftercare logging | 8 | 10 | 80 | Job completion dates |
| Total | 1,456 |
That is about 24 hours a month. Jobs 1 and 2 return 840 of the 1,456 minutes, 58% of the total, and need no data feed beyond rules the firm can write in a day. That is the decision rule: start with the two that need only your own words, run them for a month, then add the ones that need a sheet or a folder to be kept up to date. Allow a week of part-time effort for the first two, mostly spent writing and testing.
The software is the cheap part. On the vendors' own pages in November 2026, Tidio's Lyro AI Agent plan starts at $32.50 a month billed yearly ($39 month to month) for 50 conversations a month, Chatbase's entry plan is $40 a month ($32 billed yearly) for 700 message credits, and Intercom Fin charges $0.99 per outcome. Against 24 hours of office time those numbers hardly matter. What decides success is whether the rules are written, the sheet is kept current and someone reads the transcripts each week.
A number to watch for each job, and the sign it is going wrong
A chatbot that is never reviewed drifts: the sheet goes stale, a rule gets tighter than intended, a certificate expires in the folder. Pick one number per job and look at it on the first Friday of every month, with ten transcripts read alongside. None of these needs special software, only a tick box on the hand-off record or a quick count.
| Job | The number to watch | The sign it is going wrong |
|---|---|---|
| 1. Job briefs | Share of briefs the estimator can quote from without a second call | Estimators keep phoning back about the same missing item, so make that question required |
| 2. Screening | Declined enquiries that someone later says they would have taken | Even one is a prompt to loosen the rule that fired |
| 3. Site-visit booking | Visits where the customer or the access was not ready | Rising no-shows mean the reminder is not being read, so check what it says and when it goes |
| 4. Process questions | Transcripts where the bot used a duration, date or price that is not on the card | A single one: fix the wording that allowed it the same day |
| 5. Job status | Share of answers that carried an "updated" date within seven days | A falling share means the sheet is going stale |
| 6. Paperwork | Documents sent within 14 days of expiry | Any at all means the register or the rule is not working |
| 7. Aftercare logging | Urgent logs that reached a person within the promised time | One late urgent log is a process failure, so test the alert path today |
Keep a short list of what you changed and why. After three months the list shows which jobs are earning their place and which were switched on because they sounded good. Drop the second group without ceremony: a chatbot with four jobs done well beats one with seven done loosely, and every job you add widens the set of things that can quietly go wrong on a site you are not looking at.
Questions contractors ask before adding a chatbot
Does a contractor need a chatbot or just a contact form?
If most enquiries already arrive complete, a form is enough. If your first phone call is spent extracting basics such as the job type, the address and a photo, a chatbot earns its place, because it can ask follow-up questions and answer a few approved ones first. Count how many minutes your last ten first calls spent on questions the customer could have answered in writing.
Should the chatbot give ballpark prices?
Only if you already publish them, as a 'from' price with its conditions, and the bot repeats your exact wording. A ballpark invented by a chatbot gets quoted back at you on site, and for most trades the price depends on access, condition and materials that nobody has seen. Let the bot explain how your quote works and when it will arrive instead.
What if customers say they dislike chatbots?
Keep a visible 'talk to a person' button and your phone number next to the chat, and let anyone skip the questions. Many contractors' customers prefer a call, and that is fine: the chatbot is for the visitor at 9 pm or on a lunch break. Watch how many chats end in that button, and fix the flow if it is more than a third.
How long does a first version take to set up?
Allow roughly a week of part-time effort for the first two jobs, as a rough planning figure and not a promise. Most of it is writing: the questions, the rules for jobs you decline, and the hand-off wording. Configuring the software is the short part, and testing with real past enquiries is the part people skip.
Further reads
- Roofing Chatbot Platforms With AI Compared: Qualifying Storm Enquiries — Chatbot platforms compared for a roofing firm qualifying storm enquiries.
- AI Chatbot for CRM Systems: Send Website Enquiries to Job Software — Send chat enquiries into your job software.
- AI Chatbot for a Construction Company: Quote Requests and Service Areas — Service areas, project types and quote requests for a larger firm.
- AI Voice Agent for Contractors: What It Handles and What It Hands Off — The phone version: what a voice agent handles.
- How to Measure Whether Your AI Chatbot Is Actually Working — The numbers that show whether the chatbot is working.
- Heating Engineer Website: What the AI Chatbot and Form Should Ask — What a booking chatbot and form should ask for a heating job.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Tidio, Chatbase and Intercom pricing pages (read November 2026).