AI Tools for the Trades: What Works Now and What Needs Your Data First

Every couple of weeks another AI pitch shows up, and a good share of them are a gated report telling you that half your competitors are already using it. You hand over your email and find a list of products with the publisher's own near the top, ranked on a survey the software seller ran.

That list is the wrong shape anyway. For a plumbing, HVAC or electrical contractor, the useful question isn't which AI tool is best. It's what each tool needs from you before it can do anything, because that's what decides whether it's working in your shop next month or quietly cancelled after the trial.

Here's the sorting rule I'd use.

Sort AI tools by what they need from you, not by what they promise

There are three tiers.

Tier one needs nothing but language. Tier two needs your customers, your jobs and your calendar to live in one place. Tier three needs your job history in a form a computer can read. Every tool being sold to the trades right now sits in one of those three, and the price tags don't line up with the tiers at all. Plenty of tier three products are cheap to buy and impossible to use.

Most disappointing AI purchases are a tier three tool bought by a shop with tier one readiness.

Tier one: language in, language out

This tier works today, at any size, with no setup beyond a login. A general assistant like ChatGPT, Claude or Copilot will take a technician's rambling voice note and turn it into a work order description a customer can read. It will draft the awkward email to the GC about the site access you're not getting. It will read a twelve page spec section and tell you what it actually obligates you to supply, which is a decent first pass before your estimator reads it properly. It will rewrite the same safety toolbox talk five different ways so the crew doesn't tune it out.

The cost is close to nothing and the payback is measured in evenings. If you do one thing after reading this, it's this tier.

Two honest limits. First, the output is a first draft, never an answer. It will state a code requirement with total confidence and be wrong, so anything technical gets checked by the person who would have written it. Second, watch what you paste. Employee names, customer addresses, rate sheets and anything you'd call private goes into a business account with the training toggle off, not a free consumer chatbot. We've written separately about which tools are safe to put company data into, and for a BC company the privacy rules that apply to your employee and customer information don't stop applying because the software is new.

Tier two: tools that need your customers and your calendar

AI phone answering, booking assistants, automated follow up, review requests, dispatch suggestions. This is where most of the trades marketing is aimed, because missed calls are the easiest revenue leak to describe. The pitch is honest enough in outline. A homeowner with a leak calls three numbers and books the first one that answers.

Whether it works for you turns on one thing: is there a single place where your customers, your jobs and your schedule all live? If you run one field service platform and the AI is built into it, the tool can see the truck that's free at two o'clock. If your customer list is in the accounting system, your schedule is on a whiteboard and your job history is in a filing cabinet, the AI answering your phone is a polite stranger who will book two crews onto the same afternoon and sound completely certain about it.

If you're in that second situation, the fix isn't a smarter assistant. It's getting the schedule and the customer list into one system first, which is ordinary systems work and not an AI project at all.

When you do test one, point it at after hours calls only, listen to every recording, and count how many a human would have booked. That's the comparison, not the vendor's number.

Tier three: tools that need your job history

Estimating from your own past costs. Forecasting which jobs are drifting. Flagging equipment before it fails. Telling you which crew and which job type actually make money. This is the tier with the real money in it, and it's the tier that stalls at most trade contractors, for a reason that has nothing to do with artificial intelligence.

An AI takeoff tool can count fixtures and measure pipe runs off a clean drawing set, and it's genuinely good at that now. What it can't do is tell you your crew's real install rate for a rooftop unit in a 1970s strata building, because nobody ever recorded it. The hours were on a paper sheet filled in on Friday from memory, the cost codes were guessed where the foreman left them blank, and the material that went on the truck was never tied back to the job. There's no history to learn from. There's a pile of invoices.

Same story with predictive maintenance. The diagnostic pitch usually assumes equipment telemetry you don't have on the fleet you service, and without it the tool is a scheduling reminder with a better name.

This isn't an argument against tier three. It's an argument about sequence. Field time capture that gives payroll and job costing the same hours, a cost code list short enough that crews use it, quantities coming off the job weekly: that's what turns your last three years into training data. We've written about each of those separately. They're each a week or two of unglamorous work, and worth doing whether or not any AI ever touches them.

What to leave alone for now

Three things I'd keep a human in front of.

Anything that speaks to a customer in your name about price or scope without someone reading it first. A confidently wrong quoted price is a commitment you may end up honouring.

Anything that writes into your accounting system on its own. Suggested coding is useful. Automatic posting means your month end is now a forensic exercise.

Anything sold on a claim you can't test in two weeks. If the value only shows up after a year of data collection, you're being sold the data collection, so buy that on its own merits.

How to test any of them in two weeks

Pick one task that already has a number attached. Hours spent on quotes last month. Calls that went to voicemail last week. Days between job completion and invoice.

Measure it the boring way for one week, by hand, before the tool arrives. Run the tool on the same task for one week. Compare, then decide, and write one page recording what you measured and what you found. The page matters because in six months somebody will ask why you dropped it, and "it didn't feel like it helped" is not an answer you can act on.

Where this usually lands

For a BC trade contractor running somewhere between five and sixty people, the honest sequence is usually tier one this month for everyone in the office, tier two if your customer and schedule data is already in one system, and tier three after the capture side is fixed. That middle piece, getting the daily record honest and in one place, is the work that unlocks everything above it, and it's the part no vendor is selling because it isn't a product.

If you're staring at a shortlist of AI tools and can't tell which tier they're in, that's a reasonable thing to sort out on a call. We run a free 30 minute discovery call: describe what's slowing you down, and if the answer is that you don't need software yet, we'll say so.