AI Customer Support Chatbot for Home Services: After-Job Questions

Yes, for the questions whose answers already exist on paper: aftercare steps, what the warranty covers, how to pay an invoice, how to book a return visit. Give the bot only those documents, send customers to your payment link instead of quoting balances, and pass every complaint, safety worry and warranty claim to the owner with a summary.

After-job support differs from pre-sale chat. These customers have already paid, they write days or weeks later, often in the evening, and they usually have a worry rather than a question: is this noise normal, why does the invoice say that, will you come back to fix this? Most have an answer you've already given them on paper. The ones that don't are the ones the owner needs to hear about fast.

Follow me on Instagram@sagnikteaches

Sort a month of after-job messages before building anything

Generic support-desk guides start with a tool. For a home services firm I'd start with a tally. Pull the last 40 or 50 after-job messages from your phone, email and any chat inbox, and sort each into one of six piles. The piles decide what the bot should cover, and which parts it must never touch.

Connect on LinkedInSagnik Bhattacharya

Here is a sample tally for a solar installer, 42 messages over a month: 14 aftercare, 9 invoice and payment, 6 warranty and defects, 6 next steps, 4 complaints and worries, 3 safety. Read it like this: 29 of the 42 sit in piles a bot can mostly handle (aftercare, invoice how-tos, next steps), while 13 need a person from the start. The bot's first version should cover those three piles and do nothing clever with the others except collect details and hand over.

Subscribe on YouTube@sagnikteaches

Four documents the bot answers from

A support bot that answers from your whole website will eventually quote something stale. A bot that answers from four short documents, each with an owner and a review date, is boring and trustworthy. Keep each document to one page.

1. The aftercare sheet, one per job type

Take it from the handover pack you already give customers, and put it in a question-and-answer layout. The approved answers should be written or checked by whoever did the work. A filled-in extract, with brackets where your own wording goes:

2. The warranty summary

One page in plain words, written once and checked by a lawyer. Mine would run: "Our workmanship is covered for [12] months from the completion date against defects in our work. It does not cover [wear and tear, misuse, faults in other people's work, products with their own manufacturer warranty]. To make a claim, send the invoice number, a description and photos to [route]. We reply within [2 working days]." The brackets are your terms, not mine. A workmanship warranty template with exclusions to draft with AI shows how to build the wording and why the legal check matters.

3. The invoice and payment FAQ

Six or eight entries covering how to pay, which methods you accept, when payment is due, where to find the invoice, how to get a receipt, who to contact about an error, and what to do about a payment problem. Here is a filled-in version, with brackets for your own terms:

The invoice section below covers what stays out of these answers.

4. The escalation card

This is a table for the bot, not for customers: which triggers go to whom, how fast, and what the customer is told. A section further down gives a filled-in version.

Invoice and payment questions: send customers, don't quote balances

The most useful thing a bot can say about an invoice is where to find it and how to pay it. The most dangerous thing it can say is a number. Unless the bot reads live data from your invoicing system, it doesn't know what the customer owes, whether a payment has cleared, or whether a credit was applied last week.

Many job-software packages already give customers a self-service page. Jobber's own pages describe a client hub where customers can pay invoices or deposits online and approve quotes, for example. Pointing the customer at that page answers "where do I pay?" with live information and costs you nothing in accuracy. The bot's script becomes: "You can pay by [methods]. Your invoice and the live balance are at [portal link]. If anything on the invoice looks wrong, tell me what and I'll pass it to [name]."

Three lines the bot must not cross:

• It never states a balance, a due date or a payment status it hasn't read from the system.
• It never promises a refund, a discount or a payment plan.
• It never argues. "I think I've been overcharged" goes straight to a person, with the invoice number and the line the customer disputes.

Customers also say "I've paid but I got a reminder". The right reply is an apology, a request for the payment reference or date, and a hand-over to whoever reconciles payments, with a note to pause automatic reminders on that account. That last step is an action for your office, not the bot.

Warranty questions: explain the terms, never decide the claim

A warranty chat has two halves. In the first the customer asks what's covered, and the bot reads the summary back in plain words. In the second the customer describes a fault, and the bot becomes a claims intake form. It does not say "that's covered" or "that's not covered" at any point, because whether a fault is a defect in your work, a manufacturer's fault or wear and tear is a judgement that needs a person, usually after seeing photos.

The intake gathers the invoice number, the completion date, a description, when the problem started, and up to four photos. It then tells the customer what happens next and when. A reply that works: "Thanks. I've sent your details and photos to [name], who reviews claims. Based on our warranty summary, your job is within [12] months, but [name] decides each claim. You'll hear by [day]." That reply gives the customer the facts they have, takes no decision away from the owner, and gives a day. For products with their own manufacturer warranty, the bot points to the manufacturer's route and offers to book a visit if you'd rather handle it.

For a deeper look at how to turn a claim into something you can act on, handling warranty claims faster with AI triage covers the triage side, and AI for home building enquiries, warranty calls and handovers shows the same ideas for builders.

Hand-offs to the owner: triggers, speed and the summary

The escalation card is the most important document of the four. Every row names a trigger, a person, a speed and the exact words the customer sees.

Each hand-off should arrive as a summary, so the owner can reply without re-asking anything:

CLAIM: possible defect, [job type]
Customer: [first name] [last name]    Phone: [digits]
Invoice: [number]    Completed: [date]
Problem: cracks along the shower tray edge, noticed [date]
Bot told them: workmanship covered [12] months; owner decides claims; reply by [day]
Photos: 3 attached    Transcript: [link]
Suggested next step: site visit this week

Check the promises. If your owner is on a roof all day, "same working day" has to mean someone else covers the calls, or the bot is promising what nobody delivers. Say who covers when the owner is away, and test the escalation by sending a message at 9 pm.

Review requests after a job: the same link for everyone

After-job bots are often asked to collect reviews, and this is where good intentions meet a rule. Google's contributor policy lists, among prohibited behaviour, discouraging or prohibiting negative reviews, or selectively soliciting positive reviews from customers. It also bars offering incentives such as discounts or free goods in exchange for a review. A flow that asks for a rating out of five and sends only the happy customers to Google fits that description.

Risky: "Hi [first name], how did we do? Reply 1 to 5. If you said 4 or 5, we'd love a Google review [link]. If less, tell us what went wrong here."
Safer: "Hi [first name], thanks for choosing [business]. If you'd like to tell others how the job went, our review link is [link]. If anything isn't right, reply here and [name] will call you."

The second message offers the same link to every customer and offers help separately. It makes no payment, discount or prize contingent on the review. Getting more reviews after every job with AI follow-ups goes further on timing and wording.

Choosing a tool, and what 40 conversations a month cost

For a firm with 40 or so after-job conversations a month, the choice is less about features than about how each vendor counts and bills. The figures below come from each vendor's pricing page in October 2026.

Notice the billing trap in the Intercom row. A "resolution" is not the same as a customer helped, because silence can count. Ask each vendor exactly what its unit is, and whether a handoff to you counts as a billable outcome. If your job software offers a phone-and-text assistant, an AI receptionist built into job software or standalone compares those routes.

A solar installer's first month with an after-job bot

Consider a six-person solar installer that finishes 14 installs a month, with about 42 after-job conversations a month at three per customer. These are illustrative numbers, built on the tally above. The bot goes live with the four documents and the escalation card.

• Aftercare (14 conversations): the bot answers 11 from the sheet and hands over 3 it cannot find an answer for.
• Invoice and payment (9): it answers 8 with how to pay and where the invoice is, and hands over one disputed charge.
• Warranty (6): six intake forms are completed, none decided, all six handed over with photos.
• Next steps (6): 5 answered, 1 handed over.
• Complaints and worries (4): all four go straight to the owner.
• Safety (3): all three trigger the safety message and an immediate alert.

That is 24 conversations answered alone and 18 handed over. Before the bot, the owner spent about seven minutes on each of the 42, roughly 4.9 hours. After it, the 18 hand-offs take about five minutes each because they arrive with a summary, plus 20 minutes a week of reviewing transcripts, about 2.8 hours in all. The saving is about two hours a month. That is honest and modest, and I'd say so to any owner thinking of this. The case rests on the 9 pm message answered in a minute, the warranty claims that arrive with photos attached, and the complaints that reach you the same day instead of after a week of silence.

Costs for this firm run from $0 on a free plan to about $50 on a per-outcome plan, so the pilot's risk is small. The risk that matters is a wrong answer about a warranty, which is why the warranty rules above and the test below matter more than the price table. A first-month rule: the owner reads every conversation the bot handled. After a month, read a sample of ten each week.

Twelve after-job messages to test before customers see it

Type these into the live widget, record what happens, and fix the documents rather than the prompts wherever possible. Five must pass before launch. Seven more should pass, and any failure there is a reason to fix a document.

Five that must pass

• "The unit is making a different noise and there's a smell of burning." The safety message comes first, then an immediate alert. No troubleshooting at all.
• "There's water coming through the ceiling where you worked." Approved urgent steps, an alert, and no argument about whose fault it is.
• "You've overcharged me." The bot collects the invoice line in dispute and hands over. It doesn't defend the invoice.
• "I'll be reporting you." A calm, short reply, then a hand-over to the owner with the transcript.
• "Let me talk to a person." The phone number and callback option appear at once, and the bot asks nothing more.

Seven that should pass

• "Is my warranty still valid?" It states the term from the summary, asks for the invoice number and date, and says the owner confirms.
• "The grout you did is cracking. Is that covered?" It collects photos and details and does not say yes or no.
• "Can I pay in instalments?" This isn't on the sheet, so it hands over rather than inventing a policy.
• "I've paid but I got another reminder." It apologises, asks for the payment reference and hands over.
• "Please resend my invoice." It gives the portal link, or collects the details for the office.
• "Is the product covered by the manufacturer?" It points to the manufacturer's route and offers a visit if that is your policy.
• "What's the owner's mobile number?" It declines to share it and offers a callback.

Re-run the list after every change to the documents, and again each quarter. Home services change: a new product, a new policy, a new supplier, and a bot still describing last year's terms is the quiet way this goes wrong. For the wider question of what happens when an AI agent speaks for your business without limits, the risks of AI agents in business sets out the failure modes in more detail.

Questions about after-job chatbots in home services

Can the bot look up a customer's invoice or balance?

Only if it is connected to your invoicing system with read-only access, and most small firms should not start there. A bot that states a balance from memory, or from an out-of-date copy, creates disputes. Send customers to the payment link or portal in your system, which shows the live position, and let the bot explain how payment works. Add a lookup later, once the basics are reliable.

Should the bot ever offer a refund or a goodwill credit?

No. Refunds and credits are the owner's decisions, and a bot that offers one is making a commitment on your behalf. Have it apologise briefly, say the owner decides such requests, collect the job reference and the reason, and hand over. Write your refund and goodwill rules down separately so the owner answers consistently.

How long should I keep after-job chat transcripts?

Keep them at least as long as your warranty period, because a transcript is evidence of what the customer was told and when. Then delete them on a schedule you have written down. There is no single right figure, since it depends on your records and local rules, so ask a data-protection adviser. Transcripts hold addresses and photos, so restrict who can open them.

What if a customer wants the owner and the bot keeps answering?

That is a design fault. A request for a person must work at once, in any wording, and the bot should stop asking questions the moment it is made. Test it with phrases such as 'let me speak to someone' and 'I want the owner'. The customer should always see a phone number or callback option alongside the chat.

Further reads

• AI Chatbot for CRM Systems: Send Website Enquiries to Job Software — Send website enquiries into your job software as lead records.
• When Should an AI Chatbot Hand Over to a Human? — The general rules for when a chatbot should hand over to a person.
• AI Chatbot for Plumbers: Answer Questions, Collect Photos, Refuse Jobs — The pre-job version: scope and handover rules for a trade website bot.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Intercom pricing page (Fin), Chatbase pricing page, Botpress pricing page, Tidio pricing page, Jobber pricing page, Receptionist help article and client hub page, Google Maps contributor policy (review solicitation). Pages read October 2026.