Solution

The agent that chases the follow-up your team never gets to

AI agents for home services work the follow-up that never makes it to the top of anyone's list — deciding which unsold estimates are still worth a second nudge, catching a maintenance-plan renewal before it quietly lapses, reactivating a customer whose system is due for trouble, and sorting an inbound text into the right job type before anyone calls back. None of it is a phone call and none of it runs a fixed script. It is judgement applied to a queue of follow-up that always loses to whatever job is on fire today.

· Reviewed by Artur Horimoto, Founder & CEO

The money sitting in the follow-up queue

Most of the attention in this industry goes to the phone — who answers it, how fast, whether a caller gets a real person before dialing the next name on the search results page. That is a real problem, and it is a different one from what this page covers. This is about everything that happens after the call already ended: the estimate that went out and was never followed up, the maintenance plan that quietly lapsed because nobody flagged the renewal date, the customer who had a great experience eighteen months ago and has heard nothing from you since, and the inbound text that sat in a queue for two hours because nobody was sure if it was urgent.

None of that is a single dramatic failure. It is a dozen small handoffs that depend on someone remembering to act, on a day when the person who would remember is standing in a customer's attic instead of at a desk. An estimate sent on Tuesday and never followed up is not a lost job in any one visible moment — it is a job that quietly went to whoever called the homeowner back first, and in this industry that is rarely the company that sent the first quote. A membership that lapses does not trigger an alarm; the revenue and the relationship just stop, a few months apart, and nobody notices until a competitor's van is already in the driveway. This is the layer of revenue that voice AI and a booking flow were never built to catch, because it does not happen on a call — it happens in the gap between calls, where nobody is looking.

What the agent does, day to day

A custom AI agent scoped to this work sits on top of your CRM, field service platform, and messaging channels, and runs a defined set of judgement calls continuously rather than waiting for someone to open a spreadsheet.

  • Unsold estimate follow-up. Every open estimate is scored against signals like job type, quoted value, how long it has sat, and what's happened in the account since — then the agent decides which ones are worth another touch, when to send it, and what the message should say, instead of every estimate getting the same generic "just checking in" three days later.
  • Membership and maintenance-plan renewals. Renewal dates are tracked against the plan terms and the customer's actual service history, and outreach goes out with enough lead time to book the visit before the plan lapses — not a mass reminder blast the week it expires.
  • Dormant customer reactivation. Customers are segmented by what they had done, when, and what the season typically brings for that job type, so a homeowner whose AC was serviced two summers ago hears from you again right as cooling season starts, not at a random point in the calendar.
  • Inbound enquiry triage. A web form submission or text message is read for job type, urgency signals, and property details, then routed to the right queue — a same-day emergency, a routine booking, or a quote request — before a person has even opened it.
  • Callback-risk flagging. Completed jobs are checked against patterns that tend to precede a return visit — a repair on an aging system, a part substitution, a job that ran longer than the type usually takes — and flagged for a proactive check-in before the customer has to call back annoyed.

Walkthrough: deciding which unsold estimates are worth chasing

A homeowner gets a quote for a full system replacement on Monday and goes quiet. The agent doesn't treat every unsold estimate the same way. It looks at the job value, the job type's typical close-out window, whether the homeowner opened the estimate email, and what the account history shows — is this a first-time enquiry or an existing customer who has bought from you twice before.

For a high-value replacement from an existing customer who opened the estimate but hasn't responded, the agent schedules a personal follow-up on day three: a short message referencing the specific system quoted, not a generic nudge, offering to answer any questions before the promotion pricing window closes. For a low-value repair estimate that was never opened, sent to a first-time enquiry from three weeks ago, it does not chase at all — the signals say this one has gone cold, and a follow-up here is more likely to read as spam than to win the job back. Where the agent isn't confident either way — a mid-value quote with mixed signals — it flags the estimate to a sales coordinator with its reasoning attached, rather than guessing silently in either direction.

Walkthrough: a renewal window and a dormant account, same logic applied differently

A customer's annual maintenance plan is due for renewal in six weeks. The agent checks the service history attached to the plan — two completed visits, no missed appointments, one prior year of renewal — and sends a renewal reminder timed to land a few weeks before the expiry, referencing the specific visits completed rather than a form letter. If there's no response after the first message, it sends one more a week before the plan actually lapses, then stops and hands the account to a person rather than repeating itself into annoyance.

At the same time, a different customer who had a one-off repair fourteen months ago and was never signed up for a plan comes up in the dormant-customer review. The agent checks what that job was — an AC repair — cross-references the season now starting, and sends a light-touch message offering a pre-season check before the heat arrives, not a hard sales pitch. One customer gets a renewal reminder because they're already a member; the other gets a seasonal nudge because the service history and the calendar line up. The same underlying logic — service history plus timing — produces two different messages because the two accounts are in different states, which is the actual difference between this and a scheduled mail-merge.

Walkthrough: a text comes in, and a finished job gets a second look

A homeowner texts the business number at 6:45pm: "hot water heater making a weird noise, is this something you can look at." The agent reads it, matches "hot water heater" and "noise" against job types, checks whether the message suggests an active leak or just a sound, and — finding no urgency language — logs it as a routine repair enquiry, drafts a same-day acknowledgment asking two clarifying questions, and adds it to the booking queue for the next available slot. A different text later that evening — "water heater leaking all over the garage" — gets routed as urgent and surfaced to whoever is on call immediately, with the message and the customer's history attached so nobody starts the conversation from zero. This kind of enquiry triage is the same underlying judgement call a good dispatcher makes reading a message, just applied to everything that comes in through text and web forms, not only the phone.

Separately, a technician closes out a furnace repair where the fix was a part substitution because the original part wasn't in stock. The agent flags that job type and part combination as one that has historically led to a return visit within ninety days, and schedules a proactive check-in call two weeks out — not a repair visit, just a "how's it holding up" message that catches a problem before the customer is frustrated enough to leave a bad review over it.

The cost of a wrong nudge versus a missed one

Judgement means the agent will sometimes be wrong in both directions. It will occasionally chase an estimate that was never going to convert — the homeowner already booked with someone else, or decided to wait a year, and the follow-up message lands as noise instead of a nudge. It will also, occasionally, decide not to chase one that would have closed with a better-timed message. Neither failure is dramatic on its own, and no scoring model built on real accounts avoids both entirely.

The actual design question, worked out during scoping rather than assumed, is which mistake costs your business more. For a residential repair business with thin margins on small jobs, an unnecessary follow-up message costs almost nothing — a slightly annoyed homeowner, easily ignored — while a missed renewal on a recurring maintenance contract costs real, ongoing revenue. For a business that's sensitive about brand tone with high-value replacement customers, the calculus can run the other way, and the agent gets tuned to chase less aggressively and escalate sooner instead. There's no universal right answer here — it's a threshold your team sets, reviews against what actually happened, and adjusts, not a setting the agent decides for itself.

Not a phone system, and not a fixed pipeline

It's worth being direct about what this isn't. Voice AI for home services answers and makes calls — the conversations that happen live, over the phone. This system runs in the background between calls, deciding who to reach out to and when. And it's a different kind of build from workflow automation for home services, which handles the deterministic half of the admin tail — turning a completed job into an invoice, firing a reminder the day before an appointment, sending a review request on a fixed schedule. Those are fixed paths: the same trigger produces the same action every time, and that's exactly right for work that doesn't need a decision. Deciding whether a specific estimate is worth chasing, which dormant customer to reach this month, and how urgent a specific text message actually is — that's a different shape of problem, and it's why this is a separate system built around judgement rather than a rule. The AI agent glossary entry covers the underlying mechanics of what makes something an agent instead of a script, if you want the plain-language version. Most home service businesses that adopt one of these eventually run all three — voice, workflow automation, and an agent — because they solve three different halves of the same operational gap, not because one replaces the others.

Where it connects into your systems

Field service platforms. ServiceTitan, Housecall Pro, and Jobber are common examples of what a build connects to across the trades — mentioned here as tools businesses already run, not as partners or certified integrations. Where a platform exposes an API, estimate status, service history, and membership records are read directly and any outreach the agent triggers gets logged back as a real activity on the account, not a note in a separate spreadsheet.

Texting and email. Outreach for estimate follow-up, renewals, and reactivation runs through whatever channel your customers already respond on, sent from a number or address they recognize rather than a new tool they have to learn to trust.

Calendar and dispatch. Where an enquiry gets triaged into a booking, the agent checks live availability the same way a dispatcher would, rather than offering a slot that turns out to be already taken.

Data handling. The agent gets read access to service history and account data and write access scoped narrowly to logging outreach and updating status fields — not blanket access to billing or payment systems. Every message it sends and every decision it makes is logged, so a manager can see exactly why a given estimate was or wasn't chased.

Escalation. Anything the agent isn't confident about — a mixed-signal estimate, an ambiguous text, an account with a recent complaint on file — gets handed to a person with the reasoning attached, rather than resolved by a guess in either direction.

Frequently asked questions

What does an AI agent for home services actually do?

It works the follow-up that usually falls through the cracks: deciding which unsold estimates are worth chasing and when, tracking maintenance-plan renewals against real service history, reactivating dormant customers by job history and season, triaging inbound texts and web enquiries by urgency, and flagging jobs likely to generate a callback before the customer complains.

How is this different from your voice AI or workflow automation for home services?

Voice AI answers and makes phone calls. Workflow automation runs fixed, deterministic steps like invoicing and reminders the same way every time. This system makes judgement calls — which estimate to chase, which customer to reactivate, how urgent a message is — and most businesses that adopt one eventually run all three together, since each covers a different kind of work.

Will it chase leads that were never going to buy?

Sometimes, yes. Scoring every estimate and every dormant account involves real uncertainty, and an agent working on judgement will occasionally send a message that lands on someone who was never going to convert. The alternative — not following up at all — costs more in missed renewals and estimates than the occasional unnecessary nudge, but the exact threshold is set and tuned during scoping, not assumed.

Does it work with ServiceTitan, Housecall Pro, or Jobber?

In most cases, yes. Where the platform has an API, estimate status, service history, and membership data are read directly and outreach gets logged back to the same record your office already checks. Bring the software you run to the first call and we'll tell you what's possible.

How long does it take to get one of these live?

Most agents are live within weeks, starting with the single queue costing your team the most revenue — usually unsold estimates or lapsed memberships — rather than every workflow on this page at once. Scope expands once the first one is proven against your real accounts.

This is one piece of the wider operating picture for home service businesses — worth reading alongside this page if you're weighing where to start. Bring the list of open estimates and lapsed memberships sitting in your system right now to a free 30-minute strategy call, and we'll tell you honestly whether an agent is the right fix for that queue and roughly what it would take to build one.

See what this looks like for your business

Thirty minutes, your actual workflow, and a straight answer on whether this is worth building.

Free 30 minutes. No pitch deck. You leave with a plan either way.