What an AI Agent Can Do for a Plumber After the Call Ends

After the call ends, an AI agent can send the booking confirmation, ask the customer for photos, chase an unanswered quote, remind them about an unpaid invoice and ask for a review. A person should still approve anything involving price, blame, a refund or a promise about time.

Most talk about AI agents for plumbers is about the voice that answers the phone. The money, though, is often lost in the hours and days afterwards: a customer who never got a confirmation, a quote nobody chased, an invoice that sat for three weeks. The catch is that one wrong automated message, such as a payment reminder sent to a customer whose tap is still dripping, costs more goodwill than ten good ones earn, so the approval rules matter as much as the messages.

Follow me on Instagram@sagnikteaches

The quiet stretch between the call and the payment

A plumbing job has two busy moments, the phone call and the visit. Between them, and after them, sits a stretch of small admin tasks that nobody owns. The plumber is under a sink. The office person, if there is one, is on the phone. So the tasks wait, and each one has a cost: a customer who is unsure whether anyone is coming, a quote that goes cold, an invoice that is paid late because nobody asked.

Connect on LinkedInSagnik Bhattacharya

An AI agent, in the sense that matters here, is software that takes an event (a call finished, a quote sent, a job marked complete), decides what message or action comes next from rules you set, drafts the wording, and either sends it or hands it to a person. It is different from a plain automated text because it can read the notes on the job, adapt the wording, and notice when a reply does not fit the script. It is also different from a receptionist, which works while the caller is on the line. If you want that side covered, read the tutorial on AI receptionists for plumbers and burst-pipe rules first, then come back here.

Subscribe on YouTube@sagnikteaches

Six jobs make up the follow-up, and they run in this order for a typical call-out: confirmation, photo request, quote chaser, visit reminder, invoice reminder and review ask. Each one below gets a trigger, a timing, a sample message and a statement of who approves.

A four-person firm and what the gaps cost

As an illustration, take a plumbing firm with two engineers, an owner who also works on site, and a part-time office person. They handle about 22 call-outs and 6 quotes a week. Nothing here is a measured result; the figures are a worked sum so you can swap in your own.

The owner reviews a typical month and finds four leaks in the follow-up. Roughly 5 of 22 weekly call-outs book with no written confirmation, and two of those customers phone back to ask whether anyone is coming, which costs about six minutes each. Of six weekly quotes, perhaps two are never chased, and the owner estimates one of those would have converted at an average value of $620. Invoices go out at the end of the week, and the average payment arrives 17 days later. And only one customer in ten is ever asked for a review.

Put sums to it. The phone-backs: 8 or 9 calls a month at six minutes is close to an hour. The unchased quotes: about 8 a month, and if one in two would have converted, that is 4 jobs at $620, or about $2,480 of work a month that never became a booking. The slow invoices: if 22 jobs average $240, that is about $5,280 a week outstanding for an extra stretch of days, which is a cash-flow cost rather than a loss, but it matters when a van needs fuel and parts are bought on account. And the reviews: of roughly 1,100 customers a year, only about 110 are ever asked, so whatever share would leave a review is applied to a tenth of the people who could.

That is the target. Nobody should expect an agent to recover all of it. But the estimate shows why the follow-up deserves an hour of attention before the next piece of software gets bought, and it also gives you a baseline to check against after a month.

Confirmations that say what the customer wants to know

A confirmation has three jobs: tell the customer the job is booked, tell them what to do before you arrive, and give them a way to change it. The agent builds it from the job record, so the quality of the confirmation depends on what was typed into the booking. If the note says "leak", the message will be vague. If it says "kitchen sink, cold tap drips when off, stopcock under sink", the message can be specific.

A prompt you can paste into your agent or an assistant while you test:

You write booking confirmations for a small plumbing firm.
Use only the job details below. Do not add prices, arrival promises
or advice about causes. Keep it under 70 words, plain and friendly.
Always include: date and arrival window, what to clear or locate,
how to rearrange (reply to this message), and that a person will
confirm if the window changes.

Job details: [paste the booking record]

Illustrative output for a bathroom tap repair booked for Thursday between 1pm and 3pm:

Hi [first name], you're booked in for Thursday between 1pm and 3pm to look at the dripping bath tap. Please clear the space under the bath panel if you can, and tell us where your stopcock is if you know. If the time stops working, reply here and we'll move it. If our window changes, a person will message you before then.

What to fix in a draft like this: remove any line that guesses at the cause, take out "our engineer will arrive on time" if it appears, and check that it does not tell the customer to turn the water off unless your firm has a standard instruction for that. Safety advice is yours to write once, then pasted in as fixed text, never improvised by the agent.

Who approves: nobody, once the template has been tested. A confirmation that follows a fixed structure and uses facts already on the booking is the safest thing an agent sends.

Asking for photos before the visit, and what the answers do

A photo request is the step most firms never automate and most benefit from. A picture of the boiler's data plate, the stopcock, the cistern or the leak can change the van stock, the job length or whether the visit is even needed. The agent's job is to ask in the right words and to file what comes back against the job.

The request should be short and give the customer a task they can finish in a minute. Compare these two versions.

Before: "Please send some photos of the problem so we can see what is going on."

After: "To bring the right parts, could you reply with two photos: one of the whole radiator valve end, and one close-up of any drips or marks on the floor beneath it? A photo of the label on the valve helps too. Thanks, [your name]."

The second gets better photos because it names the shot. When the images arrive, the agent can do two things safely: attach them to the job and tell the engineer they are there. It can also do one thing that needs a person: describe what it sees. An assistant reading a photo of a corroded joint might call it "limescale" when it is a weeping fitting, so treat any description as a note for the engineer, never as a diagnosis sent to the customer.

If a customer sends a photo of something that looks like an immediate danger, such as water near a socket or the smell of gas, the rule must be a person calls within minutes, not that the agent replies. Write that rule into the setup; do not trust the agent to work it out.

A quote chaser that sounds like you, not like a bot

For work that needs a quote, the follow-up has most commercial weight. A customer who has a quote and has not replied is rarely saying no. More often they are comparing, busy, waiting on a partner, or unsure what they are agreeing to. A chaser that asks one specific question does better than one that says "just checking in".

A three-step sequence for a replacement bathroom-suite quote might look like this, with timings you can change:

• Day 2: a message offering to answer questions and asking if the scope matches what they expected.
• Day 6: a short note that mentions the earliest start date you can currently offer, if that is true.
• Day 12: a polite closing message saying you will hold the price until a stated date and then release the slot.

The agent writes each from the quote itself. An illustrative day-2 message:

Hi [first name], I sent your bathroom quote on Monday. If anything in it is unclear, such as what the price includes for the old suite removal or tiling, reply here and I'll go through it. If you've decided to go another way, no problem, just let me know and I'll close it off. [Your name]

Who approves: for quotes under a threshold you set, say $500, nobody. For larger quotes, the owner sees the day-6 message before it goes, because that one sometimes mentions price or availability. Never let the agent offer a discount. If a customer replies asking for one, that reply should stop the sequence and land with you. For the quoting side itself, see how plumbers use AI to write quotes.

One thing to check before you switch on automatic chasing: whether your job software already does it. Jobber's quote follow-ups, for instance, are on its Connect plan and above rather than Core, and ServiceM8 lists an AI quote and invoice feature on every plan. Housecall Pro sells its CSR AI for calls separately and publishes no price for it. Check the plan you are on before paying for a separate tool.

Invoice reminders: where the cash is

For most small firms the invoice reminder is the highest-value automation of the six because it moves money you have already earned. The goal is not to be aggressive. It is to make sure no invoice goes quiet because nobody remembered.

The simplest workable schedule has four touches, and the wording should get gradually firmer without ever becoming rude:

• Same day the job is marked complete: invoice sent with a payment link, and a line saying how to pay.
• Three days before the due date: a gentle reminder with the link again.
• One day after the due date: a plain message noting it is now overdue and asking if there is any problem.
• Fourteen days overdue: a firmer message, which a person reads before it goes.

An illustrative one-day-overdue message:

Hi [first name], invoice 1048 for the boiler service was due yesterday and I can't see a payment yet. Here's the link again: [link]. If something about it doesn't look right, reply and I'll sort it out. Thanks, [your name]

Here is the quick sum for why this matters. If the firm's 22 weekly jobs average $240 and the average payment moves from 17 days to 10, that is about 7 days earlier on roughly $5,280 a week, so around $5,000 less sitting outstanding at any moment. It is an illustrative figure, not a promise, but it explains why the invoice reminder is worth doing before the review ask.

The agent must never send a reminder for an invoice the customer has disputed. This is the single most common failure: a customer replies "the tap is still dripping", the invoice remains overdue, and the next automated message arrives like a slap. The safeguard is a dispute flag on the job that pauses all reminders. For the full reminder ladder in detail, see chasing unpaid invoices with AI for trades.

Review requests only after a clean job

Asking for a review is the lowest-risk and the easiest to damage. Sent to a customer whose job went wrong, it invites a public complaint. Sent too soon, it asks for an opinion before the water has been run through the new fitting for a day.

A sensible rule: send the review request 24 to 48 hours after the visit, only when the job is marked complete, the invoice is paid or at least not disputed, and no reply from the customer in the last two days contained a negative word. If the engineer left a note such as "customer unhappy about mess", the request is held for a person to decide.

An illustrative request that works for most trades:

Hi [first name], thanks for having us out for the tap repair. If you were happy with the work, a short review helps other people find a plumber they can trust: [link]. If anything wasn't right, tell me directly and I'll look into it. [Your name]

Do not offer anything in return for a review, and do not word the request so that only happy customers are directed to the public page. Review platforms have rules against both, and ignoring them can lead to reviews being removed. For the longer version, getting more reviews after every job with AI follow-ups goes through timings and responses.

What the person approves, and what the agent never does

The approval line is the whole design. Below is a table to adapt. The left column is the task, the middle says whether the agent may act alone, and the right column says what flips it to a person.

Three further rules sit outside the table. The agent never states a price that is not on the quote. It never says what caused a fault. And it never makes a promise about when someone will arrive beyond the window already booked. Put those as three lines at the top of the agent's instructions and check each in testing by trying to make it break them.

A realistic slip and how it shows up

Suppose the firm switches on the agent and, within a week, a customer receives a message saying "we can fit your new boiler on Saturday morning". Nobody offered Saturday. The agent saw an empty slot on the calendar and filled in the quote follow-up with a helpful-sounding availability line.

It looks innocent, and it is exactly the kind of error that damages trust, because the customer now expects Saturday. The fix is not a better prompt alone. It is a structural change: the agent is allowed to read job records but not the live calendar, and any message mentioning a day or date has to come from a fixed field (the booked slot), never from the model's own wording. Test for it by asking the agent to follow up a quote on a day the calendar is empty and checking whether it invents availability.

Setting it up in the software you already have

Most job software already covers parts of this. Before buying anything, make a one-page list of which of the six jobs your software does natively, which it does badly and which it does not do at all. The match usually looks like this:

• Confirmations, visit reminders and invoice reminders are the most common built-in features. Turn them on and rewrite the default wording in your own voice.
• Quote follow-ups are often on a mid-tier plan. If you are on a lower tier, decide whether the upgrade costs less than the quotes you lose.
• Photo requests are usually manual, a text with a link. A simple reply template is enough to start.
• Review requests are either built in or handled by a separate review tool. Do not run two at once, or customers are asked twice.

As of October 2026, Jobber's Receptionist add-on is $29 a month with 30 conversations included and $0.79 for each after that. That concerns answering the phone, not the follow-up, so do not buy it for the follow-up alone. If you want to compare answering tools, AI answering services for plumbers compared covers the options.

If your software cannot do the step, a general assistant can still draft the wording while a person sends it. That is a legitimate first stage: paste the job notes, without the customer's contact details, into ChatGPT, Claude or Gemini, get the draft, and send it yourself. It takes about 90 seconds a message. If you find yourself doing it ten times a day, that is the signal to automate.

Four more situations that test the setup

A bare list of triggers hides the awkward cases. Here are four to try on your own setup before you trust it.

The customer who books for a relative. A daughter books a call-out at her mother's home. The confirmation goes to the daughter's number, which is right, but the visit reminder must also reach the person who will open the door. The booking needs two contact fields, and the agent should know which receives which message.

The job that turns into a bigger job. A tap repair becomes a quote for a replacement kitchen supply. The agent has a completed job and a new quote on the same customer. Without a rule, it may send an invoice reminder and a quote chaser on the same morning. Give the agent one rule: never send two different follow-ups to the same customer within 24 hours; if two are due, send the invoice reminder and hold the other.

The landlord paying for a tenant's repair. The person on site is not the person paying. Reminders must go to the account holder, and a review request should go to the person who was present. If your software cannot separate these, switch off the review request for landlord jobs.

The repeat customer with an annual service. A boiler service customer is a good candidate for a rebooking message eleven months after the last visit. This is the one follow-up that is about the next job, not this one. Keep it separate from the six above and do it with your fixed wording; a rebooking message that sounds like a sales pitch costs more goodwill than it earns.

How to tell it is working

After four weeks, compare against the baseline from the worked example. These four checks are enough:

1. Phone-backs asking if anyone is coming. Count them for a week before and after. They should fall close to zero.
2. Quote conversion on chased quotes. Note how many quotes got a reply after the chaser, and how many of those booked. Do not count replies that say no as failures; a clear no is useful.
3. Days from invoice to payment. Most job software reports this. If it has not improved, check the payment link works on a phone and the wording asks for payment plainly.
4. Complaints about messages. Keep a tally of any customer who mentions a message was odd, wrong or came at a bad time. Even two in a month justify a review of the rules.

Spot-check ten sent messages every Friday for the first month. Read them as the customer would. You are looking for wrong names, messages sent after a dispute, anything that sounds like a promise, and any tone slip. Once you have four clean Fridays, reduce the check to once a month.

What the follow-up costs a firm already on job software

For a firm already on job software, the cost is mainly time: one afternoon for the first two or three tasks, a further day for the rest, then about 20 minutes a week of spot-checking. Software costs vary widely. If your plan already covers confirmations and invoice reminders, the extra is nothing. If you need a plan upgrade for quote follow-ups, compare the upgrade with the quotes you are not chasing today. If you decide to add a separate agent product, get its price in writing and ask what happens when volume spikes in a cold snap, because per-conversation charges climb when calls do.

If you would rather map this with someone who has set it up before, that is what my AI implementation consultation covers; the details are on the contact page.

A week-one plan that fits around a working week

A plumber cannot stop work for a week to set up admin. Here is a plan that takes about two hours spread across four days, using the order of value rather than the order of the job.

On day one, spend 30 minutes on the invoice reminder. Write the four messages, set the timings, and put the dispute flag in place. It earns the most, so it goes first. On day two, spend another 30 minutes on confirmations: write the template, test it with a booking in your own name, and fix whatever sounds off. On day three, write the quote chaser for quotes under your threshold and test it with a dummy quote. On day four, add the photo request wording as a saved reply, and decide whether you want the review request at all.

Give yourself one rule for the first fortnight: nothing goes to a real customer that you have not first sent to your own phone. Reading a message on a phone screen, as a customer would, catches most problems in seconds. Wrong tone, a missing space, a link that opens a login page instead of a payment page, a message that arrives with a placeholder still showing. The last is the classic one; "[first name]" arriving as literal text is a sign the template was not tested. Fix it before anyone else sees it.

After the first fortnight, add one more rule: write down every time you override the agent. If you stop a message from going out, note why. After a month you will have a short list, and that list is your rulebook. Common entries on a plumber's list include customers who prefer calls to texts, commercial landlords with their own payment process, and repeat customers on a standing arrangement who should not get the standard sequence. Each entry becomes an exception in the setup.

The broader point is that the agent is as good as the rules you write for it, and those rules come from knowing your own customers. The software supplies the typing speed; you supply the judgement about when to stay quiet. If you want the wider picture of where AI fits in a plumbing firm before you decide how much to automate, the AI guide is the place to start, and how a plumber can save five hours a week on admin puts the same idea against a weekly timetable.

Questions plumbers ask before handing over the follow-up

Is an after-call agent the same as an AI receptionist?

No. A receptionist answers the phone and books the job. An after-call agent starts when the call ends: it confirms the slot, asks for photos, chases the quote, reminds about the invoice and asks for a review. Many firms use both, and some job software bundles them, but the follow-up work is a separate task.

Can the agent send invoices without me checking?

It can, but most small firms should not allow it at first. Let it draft the invoice from the finished job and show it to you or your office person. Once the drafts have been right for a month, switch to automatic sending for standard call-outs and keep manual approval for anything with parts or extras.

What if a customer replies to an automated text with a complaint?

Set a rule that any reply containing words such as leak, damp, flooded, still not working, refund or complaint stops all automation on that job and alerts a person. An agent should never argue, apologise on your behalf or promise a free return visit.

Do I need new software to do this?

Often not. Job software such as Jobber or ServiceM8 already sends confirmations, quote follow-ups and invoice reminders, and some plans add an AI layer. Start by switching on what you already own, then add a separate agent only for the gaps.

How long does it take to set up?

A first version covering confirmations and invoice reminders is an afternoon of work: write the wording, set the timings, test it on your own phone. Photo requests, quote chasers and review asks take another day or two, mostly deciding what a person must approve.

Further reads

• AI Quoting Tools for Plumbers: What They Cost and What They Do — What quoting tools for plumbers cost and what they actually do.
• AI Software for Plumbers: What a Small Firm Pays Each Month, by Van Count — Monthly software cost by number of vans.
• Best AI for Plumbers: What a Solo, Two-Van and Ten-Van Firm Should Buy — What a solo, two-van and ten-van firm should buy.
• Can AI Triage Emergency Plumbing Calls Out of Hours? — Handling emergency calls out of hours, before this follow-up starts.
• AI Automation for Plumbers: Package Prices and What Should Be Included — What a done-for-you automation package should include and cost.
• Angi Leads for Plumbers and HVAC: Worked Cost per Job and AI Follow-Up — Follow-up speed matters most on bought leads; see the cost per job.
• Housecall Pro CSR AI vs Jobber's AI Receptionist: Which Suits a Plumber? — Housecall Pro CSR AI against Jobber's receptionist for the call itself.
• How Can AI Help Plumbers? Start With Calls, Quotes or Follow-Up — Decide whether calls, quotes or follow-up is your first AI job.
• AI Voice Agent for Contractors: What It Handles and What It Hands Off — What an AI voice agent can finish alone for a contractor, which calls it must transfer, and how Jobber, Housecall Pro and others handle the hand-off.
• 10 ChatGPT Prompts for Plumbers: Call-Backs, Quotes, Complaints — Ten prompts for call-back texts, quotes and complaint replies, each with a filled example, a pass mark and a note on what never to paste in.
• Plumbing Change Orders: Four Worked Examples Drafted With AI — Four plumbing change orders (hidden pipework, access, code upgrade, customer extras) drafted from site notes with AI, with approval wording.
• Can an AI Dispatcher Replace Your Plumbing Office Dispatcher? — What an AI dispatcher can take over in a small plumbing company, and the list of calls a person must still make.
• AI Voice Agent Trained on Your Business: Setup and Running Cost — Build the knowledge base, run test calls, set handoff rules and price the running cost per minute for an AI voice agent.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Jobber pricing and help pages, ServiceM8 pricing page, Housecall Pro pricing page (checked October 2026).