AI for home services companies: a playbook for HVAC, plumbing, roofing, and cleaning operators
In home services, the job usually goes to whoever answers the phone first and shows up when they said they would. AI will not fix a broken compressor, but it can make sure you never miss the call, your dispatcher has the right information, and your techs stop calling the office to ask where the manual is. This is the order I would do it in.
the short version
- Missed calls are the most expensive leak in most home services businesses. Capture and qualify every one, after hours included, before doing anything else.
- Dispatch support works when AI prepares the decision and your dispatcher makes it. Fully automated scheduling breaks on the exceptions that define this business.
- Quotes from customer photos are a triage tool, not a final price. Use them to send the right tech with the right parts.
- Ask every customer for a review the same way, and never write, filter, or incentivize reviews. Automate the ask, not the content.
01where to start: the phone
Ask any owner of an HVAC, plumbing, roofing, or cleaning company where revenue leaks and they will name the same thing within a minute: calls that ring out. A furnace dies at nine at night, the customer calls three companies, and the first one that picks up and gives a time gets the job. The other two never find out it happened.
That makes call capture the obvious first project. It has a clear before and after, it does not require changing how your techs work, and you can measure it with your phone system's own logs. The rest of this playbook builds on it, because every later step needs clean information about the job, which starts with the first conversation. More on how we work with home services operators is on our industry page.
02the use cases, ranked
1. Missed-call capture and booking
An assistant answers after hours and overflow calls, or texts back instantly when a call is missed. It collects the address, the problem in the customer's words, urgency, equipment details if known, and preferred times, then books into open slots or creates a callback for first thing.
What it needs: access to your schedule, your service area, your job types and their durations, and a clear emergency rule. Gas smells, flooding, no heat in freezing weather: whatever your emergency list is, those calls go to a live on-call person, not into a queue.
2. Dispatch support
Every morning your dispatcher solves a puzzle: who goes where, in what order, with what parts. AI can prepare a draft board based on tech skills and licenses you track, location, job duration, parts on each truck, and customer notes, then flag conflicts like a callback that should go to the original tech.
What it needs: structured job types, tech skills, and truck inventory, plus a dispatcher who accepts or changes the draft. The value is in the preparation. The decisions stay human, because this business runs on exceptions.
3. Quotes and triage from customer photos
Asking customers to text a photo of the unit, the leak, or the roof section is already common. A model can read those photos alongside the description to identify equipment type, visible damage, and likely job category, then attach a triage note to the job: which tech, which parts, whether it likely needs a second visit.
What it needs: a library of your past jobs with photos and final outcomes so the triage can be checked against reality, and a rule that the customer gets a price range only when your team approves it. For roofing and remodeling, photo triage can also speed up the estimate appointment by letting the estimator arrive prepared.
4. Follow-up, maintenance reminders, and review requests
The money in home services is often in the second job: the maintenance plan, the seasonal tune-up, the replacement quote that was "not right now" six months ago. Automated follow-up after each job can confirm the work, offer the maintenance plan where it fits, and set reminders for the next seasonal visit.
Review requests belong here too. Ask every customer after every completed job, the same way, with a direct link. Do not filter who gets asked, do not offer incentives for positive reviews, and never write or post reviews on a customer's behalf. Those practices break platform rules and can break consumer protection law. Automate the ask; leave the content to the customer.
What it needs: job completion data, customer contact preferences and consent for texts, and templates your team approves once.
5. Technician knowledge lookup
Your senior techs carry a lot in their heads. Your newer techs call the office or the senior tech to ask about a fault code, a wiring diagram, or how you handled a similar job last year. A knowledge assistant that searches your equipment manuals, internal notes, and past job records lets a tech ask in plain language from the truck and get an answer with the source attached.
What it needs: manuals and notes gathered in one place, a retrieval setup that cites the exact page it used, and a rule that it says "I could not find that" instead of guessing. On safety-critical topics like gas and electrical work, the answer should always point to the manufacturer document.
| Use case | Main risk | Control |
|---|---|---|
| Call capture | Emergency stuck in a queue | Hard escalation list to a live person |
| Dispatch support | Bad assignment on a complex job | Dispatcher approves every board |
| Photo triage | Wrong price expectation | Prices only after team review |
| Review requests | Non-compliant practices | Same ask for everyone, no incentives |
| Tech knowledge | Invented answer on a safety topic | Cited sources, "not found" allowed |
03what to avoid
- A bot that pretends to be a person. Customers are fine talking to an assistant that books them quickly. They are not fine finding out later they were misled.
- Over-automating the schedule. Fully automated booking that ignores drive time, job complexity, or the customer who needs the tech who speaks their language creates callbacks.
- Firm quotes from a photo. You will eat the difference on every job where the real problem was not visible.
- Texting without consent. Marketing texts need opt-in. Keep job updates and marketing on separate consent tracks.
- Ten disconnected apps. Call capture in one tool, dispatch in another, follow-up in a third, and nobody owns how they connect. I compare that path with a custom build in no-code vs custom automation.
04a 90-day sequence
- Days 1 to 15: pull your phone logs and count missed and after-hours calls by week. Write your emergency list and job types with durations. Confirm what your field service platform exposes through its API.
- Days 16 to 40: launch missed-call text-back and after-hours capture. Review every conversation weekly for the first month.
- Days 41 to 65: add photo intake and triage notes to new jobs, plus the post-job follow-up and review request.
- Days 66 to 90: start the dispatch draft board and the tech knowledge assistant with a few senior techs testing it first. They will tell you quickly what it gets wrong.
05how to measure it
- Missed and unanswered calls per week, and how many became booked jobs.
- Time from first contact to booked appointment.
- First-visit completion rate, which improves when techs arrive with the right parts.
- Callbacks to the office from techs in the field.
- Maintenance plan signups and seasonal rebookings.
- Review requests sent per completed job.
06where Insomnia Club fits
We connect the pieces your current tools leave apart: call capture that feeds your schedule, triage notes that feed dispatch, follow-up that runs off completed jobs, and a knowledge assistant your techs can use from the truck. That is AI workflow automation and AI agents, with internal tools when your office needs a screen to run it. Fixed budget agreed before work starts, working software every two weeks, and we stay after launch. If your field service platform already does the job, use it and keep the money.
common questions
Can AI answer phone calls for an HVAC or plumbing company?
Yes. A voice or text assistant can answer after hours and overflow calls, collect the address, problem, and urgency, book into open slots, and escalate true emergencies to an on-call person immediately. It should sound clear about being an assistant and should hand off whenever a caller asks for a person.
Can AI give a quote from a photo?
It can give a useful first read: identifying the type of equipment, visible damage, and likely job category so you can send the right tech and parts. A final price should still come from a technician on site or a person reviewing the photos, because the costly details are often behind the wall.
Will AI replace our dispatcher?
No, and I would not try. AI can prepare the board each morning, suggest assignments based on skills, location, and parts, and flag conflicts. The dispatcher makes the call, especially when emergencies, callbacks, and customer preferences collide.
Is it okay to use AI to get more reviews?
Using automation to ask every customer for honest feedback after a completed job is fine and common. Writing reviews, posting them for customers, only asking happy customers, or offering incentives for positive ones can break platform rules and consumer protection law. Automate the request and keep it the same for everyone.
Do we need custom software or does our field service platform cover this?
Start with what your field service platform already offers. Custom work makes sense when you need the pieces to talk to each other, such as call capture feeding dispatch feeding follow-up, or when your pricing and scheduling rules are specific enough that generic tools fight you.