How to Use AI in a Trades Business: What Works, What Doesn't

Trades businesses get the most from AI on office work: answering and booking calls, texting back missed callers, following up estimates, drafting messages and sorting bookkeeping. AI can't see the job, so it shouldn't diagnose, price complex work or handle angry customers alone. Start with your biggest money leak, add guardrails, and measure results.

Every software vendor now says it has AI, and every owner has a cousin who swears ChatGPT runs his business. The truth is narrower and more useful. AI in 2026 is very good at a handful of office jobs that trades shops do badly when they're busy: picking up the phone, replying fast, following up, and writing things nobody has time to write. It's not good at the work in the field. This guide separates the two, shows the math on what's worth paying for, lists the guardrails, and gives you a 90-day plan to roll it out without breaking anything.

1. What does "AI" actually mean for a contractor?

For a contractor, AI usually means one of four things: voice agents that answer phone calls, chat assistants like ChatGPT or Claude that write and summarize, automations that act on triggers such as a missed call or an unsigned estimate, and image tools that measure roofs or read photos. Most products combine several.

Under the hood, most of these run on large language models: conversational AI that predicts good responses from patterns in text. That's why they're good at language and bad at facts they haven't been given.

2. Where does AI work well in a trades business today?

AI works well where the job is fast, repetitive communication with clear rules: answering routine calls, booking appointments, texting back missed callers, sending reminders, following up open estimates, asking for reviews, drafting emails and job posts, and categorizing transactions. Those are the tasks that slip when the office is busy.

3. Where does AI fall short?

AI falls short anywhere it needs to see the job, judge safety, or carry responsibility. It can't diagnose a furnace by phone, price a remodel with rot behind the drywall, decide whether a gas smell means evacuate, or calm a customer whose ceiling just collapsed. It also states wrong things confidently.

NIST's generative AI risk profile calls those confident mistakes "confabulation." In a trades business, that looks like an AI quoting a price you never set, promising a same-day slot you don't have, or giving a safety instruction that's wrong. The fix is design, not hope: give the AI only the facts it may use, tell it what to never do, route edge cases to people, and review transcripts. Will AI replace receptionists? covers what people still do better.

4. Step one: find your biggest money leak

Before buying any AI tool, find where you're losing the most money: unanswered calls, slow follow-up on estimates, unpaid invoices, or office hours spent on paperwork. Put a dollar figure on each, then fix the biggest one first. AI is a tool for a specific leak, not a strategy.

5. Step two: put AI on the phones

For most shops the first AI tool should answer the calls you miss. An AI receptionist picks up overflow and after-hours calls, answers common questions from facts you give it, books into your calendar, and transfers or pages for emergencies. Pair it with a missed-call text back for anyone who still slips through.

Prices vary by model. On vendor pages we checked in October 2026, Jobber's AI Receptionist was a $29-a-month add-on on most plans, Smith.ai's AI Receptionist started at $150 a month for 75 calls, and Rosie started at $49 a month for 250 minutes. Housecall Pro's CSR AI and ServiceTitan's AI Virtual Agent don't publish prices. See the best AI receptionists for contractors comparison, is an AI receptionist worth it? and the full never miss a call guide.

6. Step three: automate follow-up

Next, automate follow-up on every lead and open estimate. A sequence of texts and emails at set intervals, stopped as soon as the customer replies or signs, closes more work than a busy owner remembering to call. AI helps by writing natural messages and handling simple replies.

1. 1

   Same day

   Thank-you text with the estimate link and a direct number for questions.

2. 2

   Day 2

   Short check-in: "Any questions on the estimate? Happy to walk through options."

3. 3

   Day 5

   Address a common worry: financing, warranty or scheduling availability.

4. 4

   Day 10

   Final friendly nudge; offer to hold a slot or revise the scope.

5. 5

   Stop on reply

   Any reply hands the conversation to a person.

Templates for these messages are in estimate follow-up templates, and how to follow up on an estimate covers timing. Add automatic review requests after every completed job with the Google review request templates.

7. Step four: use a chat assistant for office writing

A general chat assistant such as ChatGPT, Claude or Gemini saves real time on writing: job posts, customer emails, review replies, scope-of-work drafts, policies, and summaries of long documents. Free tiers exist, and paid plans cost about the price of a lunch a month; Claude Pro listed at $20 a month billed monthly in October 2026.

Don't paste customer personal data or financial details into a consumer chat tool unless you've checked its data settings. ChatGPT prompts for contractors has dozens of prompts that work in any assistant.

8. Step five: let AI help with the books, with review

AI can suggest categories for bank and card transactions, match receipts, flag duplicates, and remind customers about unpaid invoices. That keeps books closer to tax-ready all year. It still needs a person to review, and it doesn't replace your bookkeeper's judgment or your CPA's advice.

Clean data in makes AI useful out. Set up your chart of accounts, projects and classes properly first with the QuickBooks setup guide, and keep field software and accounting in sync with a two-way sync so the AI isn't working from half the picture. See the bookkeeping checklist for contractors.

Is AI useful out in the field?

In the field, AI is a helper, not a decision-maker. Techs use it to look up specs and manuals faster, dictate job notes that get cleaned into a readable summary, translate for customers who speak another language, and draft the repair-or-replace explanation they then present. The tech still diagnoses, tests and decides.

• Voice-to-notes. A tech talks for 30 seconds after the job; AI turns it into clean notes on the work order. Customers and the office both get a clearer record.
• Options write-ups. AI drafts good-better-best descriptions from the tech's findings. See good-better-best pricing.
• Manual and spec lookup. Faster than scrolling a PDF, but check the answer against the manufacturer's document before acting on it.
• Photo documentation. Before-and-after photos attached to the job protect you on warranty claims and disputes, AI or not.
• Training. New techs can ask a chat assistant to explain a concept, then confirm with a senior tech.

What to avoid: letting AI pick a part, a wire size or a pipe size without the tech checking code and the manufacturer's spec. Calculators like the voltage drop calculator and pipe size calculator are rough estimates too; code and the engineer of record decide.

9. Guardrails and rules you must follow

Using AI with customers brings a few legal and practical rules. Disclose call recording where required, get consent before automated texts and register your texting numbers, and get prior consent before AI-voiced outbound calls. Above all, decide what the AI may say and when it must hand off to a person.

• Call recording. Some states require everyone on the call to consent to recording. Put a recording disclosure in the greeting.
• Automated texts. Get consent before texting, honor STOP replies, and register business texting numbers with carriers (A2P 10DLC).
• AI voices on outbound calls. The FCC's February 2024 ruling treats AI-generated voices as artificial voices under the TCPA, which require prior express consent.
• Facts only from you. Feed the AI your services, areas, hours, fees and policies; tell it to say "a tech will confirm" for anything else.
• Escalation rules. Emergencies, complaints, legal questions and anything unusual go to a person.
• Weekly transcript review. Read a sample of conversations every week and fix the instructions.

10. How do you evaluate an AI vendor?

Evaluate AI vendors on what happens to the customer, not on demos. Ask whether it books directly into your system, how it handles emergencies, what it costs at your real volume, whether you can test on live calls month to month, and who owns the recordings and data.

11. AI use cases by trade

The office jobs AI handles are similar across trades, but each trade has a best first use and a specific thing to watch.

For a wider look, see AI for home service businesses, and the trade AI receptionist pages for roofing, general contractors and landscaping.

12. A 90-day AI rollout plan

1. 1

   Days 1–14: measure

   Pull call logs, count open estimates, check DSO and office hours on admin. Put a dollar value on each leak.

2. 2

   Days 15–30: fix the biggest leak

   Turn on missed-call text back and an AI receptionist for overflow and after hours. Write the facts sheet and escalation rules.

3. 3

   Days 31–45: review hard

   Read transcripts daily for two weeks, then weekly. Fix wrong answers in the instructions.

4. 4

   Days 46–60: add follow-up

   Automate estimate follow-up and review requests. Track close rate and review count.

5. 5

   Days 61–90: office writing and books

   Adopt a chat assistant for writing; tighten bookkeeping categorization. Compare every number to day one.

AI readiness checklist

• Measured missed calls, response time, open estimates and DSO.
• Written facts sheet: services, areas, hours, fees, policies.
• Written escalation rules for emergencies and complaints.
• Recording disclosure in the greeting where required.
• Texting consent and carrier registration handled.
• AI books into your real calendar, not a message inbox.
• Weekly transcript review on the calendar.
• Before-and-after numbers tracked for every tool.

Common AI mistakes

• Buying AI before finding the leak. Tools without a target become another subscription.
• Letting it improvise. Without a facts sheet, AI fills gaps with confident guesses.
• No human path. Customers who can't reach a person in an emergency don't come back.
• Never reading transcripts. Small errors repeat hundreds of times.
• Too many tools. Separate AI apps for calls, texts and follow-up that don't share data drop leads between them. See CRM.