ServiceTitan's April 2026 survey of 1,000 residential contractors found that about 74% see AI as key to efficiency, that roughly a quarter use it today, and that more than half are unsure where to start. The survey was run by Thrive Analytics for ServiceTitan, and ServiceTitan notes that it provides no assurances about the accuracy of the data, so read it as a sentiment snapshot.
The gap between "this matters" and "we use it" is rarely a shortage of tools. It is usually that the business is not ready to hand anything to a machine. An AI that books jobs needs somewhere to write them, a booking process to follow, and a phone line someone controls. If those are missing, the AI inherits the mess and runs it faster.
This checklist covers the six things we would check first. It is a companion to the where-to-start guide for contractors, which explains which problem to point AI at. This one tells you whether you are ready to point it at anything. Score each item green, yellow or red. You can do it in an afternoon.
1. Clean customer data in your field service software
Every automation reads from and writes to your customer records. If those records are wrong, the AI will text the wrong person, skip the right one, or create a duplicate and send your tech to a customer who already has a job open.
Check these in your software (ServiceTitan, Housecall Pro, Jobber, or whatever you run):
- Duplicates. Search your ten most common last names. Count how many customers appear twice.
- Phone numbers. Pull 50 recent customers and call or match the numbers. How many are missing, landlines, or a spouse's number from five years ago?
- Source tags. Does each job record where the customer came from, or does half of it say "unknown"?
- Job status. Are old estimates still marked open two years later? Dead estimates turn follow-up automation into spam.
Green: under 5% of the sample is wrong. Yellow: 5% to 15%. Red: more than that, or you have never looked. Those cutoffs are our rule of thumb, not an industry standard. Pick your own and write them down so you can compare next quarter.
2. A defined booking process, written down
Ask your best CSR how a call becomes a booked job. Then ask your newest one. If the answers differ, you do not have a process yet, you have two people.
A booking process spells out:
- What questions get asked, in what order
- Which calls are emergencies and what the dispatch rule is for them
- How you decide whether to book, quote or refer out
- What the CSR does when the schedule is full
- What gets logged, and where
An AI receptionist or follow-up system will do whatever you tell it. If you cannot tell it, it will guess. This is the single most common reason a pilot underperforms: nobody ever defined what "booked" means.
Green: one page that two different CSRs would follow the same way. Red: it lives in one person's head.
3. Call tracking you actually look at
You cannot improve a number you do not have. At minimum you need to know, per week: how many calls came in, how many were answered by a person, and how many of the answered ones became jobs.
ServiceTitan describes call booking rate as the share of incoming calls that become jobs, and its 2022 analysis of more than 3,000 trade businesses put the typical shop at 42%. Whatever tool you use to track it, the test is simple: can you tell me last week's booking rate by branch without calling anyone? If not, you cannot tell whether anything you add made a difference.
Green: tracked weekly, by branch. Yellow: tracked but nobody reviews it. Red: no call tracking at all.
4. One named owner of the phone line
Who can change where your main number rings? Who can add a forwarding rule, port the number or cancel the carrier account? In a surprising number of shops the answer is "the guy who set it up in 2019, who left" or "the account is in the founder's personal name."
An AI answering layer usually needs a forwarding rule, a number port or a carrier setting. If nobody has the login, the project stalls in week one. Find out now:
- Which carrier or VoIP provider holds each number
- Whose name and email are on the account
- Whether you have admin access, and where the credentials are stored
- Whether your Google Business Profile and ad numbers point to the numbers you think they do
Green: a named person with admin access to every number. Red: you would have to look it up.
5. Consent records for anyone you will text or call
If AI will send follow-up texts or place outbound calls, you need to be able to show where each contact's permission came from. The FCC's rules at 47 CFR 64.1200 define "prior express written consent" as a signed written agreement that clearly authorizes the seller to send the messages, includes the phone number authorized, and carries a clear disclosure that signing is not a condition of buying anything. Electronic signatures count if they are valid under applicable law. The same rule requires that do-not-call requests be honored within a reasonable time, not to exceed ten business days, and the do-not-call rule at section 64.1200(d)(6) says a request must be honored for five years.
Whether a given message needs written consent depends on what it says and how it is sent, which is a question for counsel. What you can check today is whether the evidence exists:
- Does your web form record the consent language the customer saw, with a timestamp?
- Can you export a list of everyone who opted out, and does every system you use respect it?
- Is there a record for contacts you imported from an old database?
Our guide to texting and calling rules for contractors goes deeper. This section is general information, not legal advice. Confirm your own situation with counsel.
Green: consent captured and stored per contact, with an opt-out list every tool respects. Red: you could not prove it for a single contact.
6. Someone accountable for the result
Every AI project needs one person who is on the hook for a business number, not a technology milestone. "Booked jobs from missed calls went from 6 a week to 11" is a result. "The agent is live" is not.
Pick the person before you pick the tool. In a small shop it is usually the owner or GM. In a multi-location company it is often the ops lead or the call center manager. They need three things: the authority to change the booking process, access to the weekly numbers, and fifteen minutes a week to look at transcripts and fix what is wrong. Without that, the system drifts and people stop trusting it.
Green: a named person with a number and a weekly review. Red: "the office handles it."
How to read your score
Tally your colors:
| Result | What it means | What to do |
|---|---|---|
| Mostly green | You can start a fixed-scope pilot now | Pick the one leak with the highest dollar value and measure it for 30 days |
| A few yellows | Start the pilot and fix the yellows in parallel | Assign each yellow an owner and a date |
| Any red in items 2, 3 or 4 | Fix those first | A booking process, call tracking and phone line access gate everything else |
| Red in item 5 | Do not start outbound texting or calling yet | Inbound answering can proceed while you sort out consent |
Notice what is not on the list: a data lake, a new CRM, a technical hire. A ten-person HVAC shop with a one-page booking script, a clean customer list and a named owner is ready before a 40-person shop with none of those.
What to do with a red
Most reds are a week or two of work, not a project. Dedupe the customer list. Write the one-page booking script from your best CSR's calls. Move the phone account into the company's name. None of that needs AI, and all of it pays off even if you never buy any.
If you want a second set of eyes on the answers, our free AI Readiness Audit walks through these six areas with you and tells you what to fix before spending money. If the checklist comes back green and you want the build, the Booked-Jobs Recovery System is a fixed-scope engagement, live in 30 days and measured in booked jobs. Before you commit to any vendor, it is worth reading what to ask an AI consultant and how to measure AI ROI in a contracting business.