AI job intake for a service business is not one product. It is a chain of seven handoffs between the moment a homeowner first reaches out and the moment a truck is assigned, and a job can die at any one of them. That is the through-line of everything below: intake is a chain, not a funnel. A funnel narrows and you accept the loss. A chain has links, and one weak link means the whole thing drops the load — which is why owners who buy an AI receptionist and see almost no change usually fixed link one and left links two through seven exactly as broken as they were.
I have walked this path with HVAC, plumbing, electrical, and roofing shops in the $1M-$5M range. The leaks are remarkably consistent. Here they are in order, with what the AI layer actually does at each point and what it cannot do.
1. The ring that nobody picks up
Link one is the only one most owners measure, and even then they measure it wrong. The number that matters is not "missed calls." It is missed calls during billable hours from numbers that have never called before. A missed call from an existing customer is a nuisance. A missed call from a first-time caller is a lost job, because that caller has your competitor's number in the same search results and dials it within about ninety seconds.
Pull your call log for last month and segment it three ways: answered, missed-and-called-back, missed-and-never-recovered. Most shops running three to twelve trucks find that the never-recovered bucket is somewhere between 15% and 30% of inbound volume, and it clusters hard between 11am and 1pm and after 4pm — lunch and the end-of-day rush when the office is closing out tickets.
The AI layer here is voice answering that picks up on ring two, every time, with no queue. What it does well: capture the name, the number, the address, the service type, and the urgency. What it does not do: sell a difficult job, calm a genuinely angry customer, or make a judgment call about whether to squeeze in a friend of the owner. Build the escalation path before you build anything else. If you are weighing this against a human answering service, the tradeoffs are laid out in this comparison of AI receptionists and live answering services.
2. The form fill that sits for four hours
Web forms leak differently than calls. Nobody drops them — they just sit. The homeowner filled out your form at 8:40pm, and your CSR sees it at 7:15 the next morning. By then that homeowner has also filled out two other forms, and one of those companies texted back in ninety seconds.
This is the single cheapest link to fix and the one owners most often skip because it feels less urgent than a ringing phone. It is not less urgent. The response-time curve is brutal and well documented; we broke down the underlying math in the piece on speed to lead for service businesses. The practical build is an instant acknowledgment text that names the specific service requested — not "Thanks for contacting us," but "Got your request about the water heater at 1420 Oak. Are you without hot water right now?"
That second sentence is doing real work. It is a qualifying question disguised as a courtesy, and it is what turns a form fill into a routable job in link four.
3. The intake that captures a name but not a job
Here is where most AI intake deployments quietly underperform. The bot answers, it is polite, it grabs a name and number — and the CSR still has to call the customer back to find out what is actually wrong. You did not remove a step. You added one.
A complete intake record for a trades job needs the fields that determine dispatch, and they are trade-specific:
- HVAC: system age, unit location (attic, closet, roof), whether it is blowing but not cooling versus not running at all, whether the home has anyone medically vulnerable during a heat advisory.
- Plumbing: whether water is currently shut off, whether it is a supply or drain issue, fixture count affected, and whether the customer has already had someone out.
- Electrical: panel brand and amperage if visible, whether a breaker is tripping repeatedly, burning smell yes or no. That last one is a same-day dispatch trigger, no exceptions.
- Roofing: active leak or not, and — the branch no other trade shares — whether an insurance claim has been filed and with which carrier. That single field routes the job into a completely different sales process.
Write these questions down as a script before you configure any tool. The AI is only as good as the intake form you hand it, and a generic out-of-the-box script will produce generic out-of-the-box records.
4. The routing decision nobody encoded
Now you have a complete record. Link four asks: who can actually do this job, and when?
Most owners carry this matrix in their head. Miguel does not do commercial rooftop units. Only two techs are Type II certified under EPA Section 608, so refrigerant work routes to them. The new guy can pull permits but is not signing off on service changes. Dispatch works because the owner or a ten-year dispatcher holds all of it.
An AI scheduler cannot infer any of that. It has to be told. This is the step where booking automation goes wrong in a way that creates a compliance problem rather than just an annoyance — an intake system that books an uncertified tech onto a refrigerant job or an unlicensed helper onto work requiring a journeyman card is not a scheduling error, it is an exposure. Encode the skill-and-license matrix explicitly: job type to required certification to eligible tech list to service area.
ServiceTitan, Housecall Pro, Jobber, FieldEdge, and Service Fusion all expose booking through an API or a connector, but almost all of them will only book into an open capacity window by default, not to a named technician, unless that matrix is configured first. If you want the deeper version of this, the buyer's guide to AI dispatch software for 3-20 truck shops covers what to ask vendors.
5. The booking that never gets confirmed
A job on the calendar is not a job. It is an intention. Between booking and arrival you have three failure modes: the customer forgets, the customer books someone else who can come sooner, or the tech arrives at an empty house.
Confirmation automation is boring and it is the highest-return link in the chain relative to effort. The sequence that works is three touches: an immediate confirmation with the arrival window and the tech's name, a day-before reminder with a one-tap reschedule link, and an en-route text when the tech leaves the previous stop. That last one cuts no-shows more than the other two combined, because it converts a four-hour window into a twenty-minute one.
One compliance note, because this is where automated intake sequences get owners in trouble. The FCC's revocation rule that took effect on April 11, 2025 requires you to honor an opt-out made in any reasonable manner — "stop," "quit texting me," a spoken request to a tech — and to process it within ten business days. Your automation has to have a working suppression list that a human can add to. Read the FCC's own guidance rather than taking a vendor's word for it.
6. The job that gets rescheduled into oblivion
Reschedules are the quiet killer. A customer moves an appointment once and it stays booked. They move it twice and it enters a limbo where nobody owns it. In most shops the second reschedule is where the job silently exits the pipeline — it is no longer on the board, no longer in the follow-up queue, and no longer in anyone's head.
Give reschedules a state of their own. Any job rescheduled twice should trigger a human call, not another automated text. There is usually a real reason underneath — price anxiety, a spouse who has not agreed, or a competitor's estimate on the counter — and that reason is closeable by a person and not by software. This is the boundary I would draw hardest: automation handles volume and timing, the owner and the CSR handle ambivalence.
The same logic applies to estimates that go quiet. A structured follow-up cadence for unsold estimates is the highest-margin automation most trades shops never build; the lead follow-up system for contractors walks through the cadence itself.
7. The intake data you never look at
The last link is the one that makes the other six improve over time, and it is why intake and marketing are the same system.
Every AI intake tool worth using produces a transcript. Those transcripts answer questions your CRM cannot: which service the callers actually ask for versus what you advertise, how many callers were outside your service area, how many wanted a trade you do not offer, and what objection comes up before booking. Read twenty transcripts a month. You will find at least one recurring request you could profitably fulfill and at least one script problem costing you bookings.
Transcripts also have a direct financial use. If you run Google Local Services Ads, you are charged per lead and you can dispute charges for leads that were spam, out of area, or for a job type you do not list — but disputes have a filing window and they succeed or fail on evidence. A transcript showing the caller asking for a service you have never offered is the cleanest evidence there is. Shops that dispute systematically instead of occasionally recover a meaningful share of wasted lead spend.
How to find your own weakest link this week
You do not fix seven links at once. Run this sequence over five business days:
- Monday: export last month's call log, segment into answered / recovered / never-recovered. Count the never-recovered from new numbers.
- Tuesday: take your last 25 web form submissions and record the timestamp gap between submission and first outbound contact. Find the median, not the average — the average lies.
- Wednesday: pull 15 booked jobs and check whether the intake record had enough information to dispatch without a callback. Score each yes or no.
- Thursday: count last month's no-shows and count how many had an en-route text.
- Friday: list every job rescheduled more than once in the last 60 days and find out what happened to each.
Whichever day produced the ugliest number is where you start. Fix that link only, run it for four weeks, then move to the next. Owners who try to install a complete intake system in one push almost always end up with a half-configured tool and a CSR who has gone back to the old way. If you want a paced version of this, we mapped one out in the 90-day rollout plan for contractors.
What AI job intake does not do
It does not replace your dispatcher's judgment about which tech handles a difficult homeowner. It does not diagnose. It does not decide whether to eat a warranty call to protect a relationship. It does not close a $14,000 system replacement — a person in a kitchen does that. What it does is make sure that every opportunity that reaches your business survives all seven handoffs and lands in front of that person.
That is a smaller claim than most AI marketing makes, and it is the one that holds up. At Turnkey AI we build these intake chains link by link for trades shops, and the pattern is always the same: the win is not intelligence, it is the elimination of drops.
Questions owners ask about AI job intake
Will customers know they are talking to AI?
Many will, and that is fine as long as the system is fast, accurate, and hands off to a human on request. What customers object to is not the machine — it is the machine that cannot answer their question and will not let them reach anyone who can. Build an escalation path that reaches a real phone within one request.
Do I need to replace my field service software?
No. The intake layer sits in front of ServiceTitan, Housecall Pro, Jobber, or whatever you run, and writes into it. Replacing your system of record to add intake automation is the wrong order of operations and an expensive mistake.
Frequently asked questions
Will my customers know they are talking to an AI during intake?
Many will, and that is acceptable as long as the system is fast, accurate, and hands off to a human on request. What customers object to is a machine that cannot answer their question and will not connect them to anyone who can. Build an escalation path that reaches a real phone within one request.
Do I need to replace my field service software to add AI job intake?
No. The intake layer sits in front of ServiceTitan, Housecall Pro, Jobber, FieldEdge, or Service Fusion and writes records into it through an API or connector. Replacing your system of record just to add intake automation is the wrong order of operations.
Can an AI intake system book directly to a specific technician?
Only if you first encode the skill and license matrix — job type to required certification to eligible techs to service area. By default most platforms book into an open capacity window, not to a named tech. Skipping this step risks routing refrigerant or service-change work to someone not certified to perform it.
Is automated text follow-up on intake legal?
Yes, with consent and a working opt-out. The FCC rule effective April 11, 2025 requires honoring a revocation made in any reasonable manner and processing it within ten business days, so your automation needs a suppression list a human can add to immediately.
Which intake leak should I fix first?
Whichever produces the worst number when you measure. Segment last month's calls into answered, recovered, and never-recovered; then measure the median gap between web form submission and first contact. Start with the ugliest of the two and run it four weeks before touching anything else.