Every week another AI tool promises to transform your business. Most were built by software developers who have never stepped on a job site. The result is expensive technology that doesn't actually connect to how a trades business runs.

After 18 years in the trades and now working with Okanagan companies on AI, I keep seeing the same thing: two or three places where AI genuinely saves time and money, and a dozen where it creates more work than it replaces.

The tools aren't lying about what they can do. They're just built for the wrong job. Software designed to summarize contracts or write marketing copy doesn't know the difference between a rough-in inspection and a final walkthrough. It doesn't know that your framing crew underbids every third job because they eyeball the waste factor instead of calculating it. That knowledge lives in your business, not in the tool, and no amount of "AI-powered" marketing changes that.

I've sat across the table from enough owners to know the instinct that follows a bad AI purchase. Write the whole category off. That's the wrong lesson too. The businesses doing this well aren't smarter or better funded, they just aimed at the two or three places where the work is repetitive, the data already exists, and getting it wrong is cheap to catch. Everything below is either a place I'd point you toward or a place I'd steer you away from, based on what's actually held up over repeated projects.

Estimating and quoting

This is where I see real payback, when it's set up properly. I've watched quote turnaround drop from three days to under two hours. The catch: the AI has to learn from your own historical job data. Generic tools guess from industry averages. You'll win jobs you lose money on.

What "set up properly" actually means

Say a mechanical contractor hands me five years of past quotes, actual costs, and change orders. That's the raw material. The AI isn't inventing pricing logic from nothing, it's finding the patterns already sitting in the paperwork: which assemblies always run over, which suppliers' lead times blow the schedule, which types of jobs the estimator underbids because they're rushed at the end of the day. Once that pattern is captured, a new quote for a similar job pulls from real history instead of a gut feeling typed up at 6pm on a Friday.

That setup takes weeks, not a weekend. Someone has to clean the historical data, decide which past jobs still reflect how the business prices today, and catch bad assumptions before quotes start going out to clients. Skip that step and you've automated your best guess instead of your track record, which is worse than doing nothing.

The businesses that get burned here usually bought a generic estimating tool that promised AI-powered quotes in minutes without asking where the pricing data came from. If the answer is "industry benchmarks" rather than "our last 200 jobs," you're pricing against a market average that has nothing to do with your labor rates, your suppliers, or your crew's actual output on a Tuesday in February.

Client communication

Quote requests, project updates, invoice follow-ups. These eat hours every week and none of them need your expertise, just consistent, professional communication.

AI-drafted responses, reviewed before you hit send, keep clients informed without pulling you off the tools. A project manager can dictate three sentences about why the drywall delivery is running late, and the AI turns it into a properly formatted update that goes out to the client, the site super, and the scheduler at once. That's the real win, not that the writing is better than what the PM would have produced, but that it happens at all instead of getting pushed to the end of the day and forgotten.

The review part isn't optional. AI doesn't know your client relationships. It doesn't know that one client shrugs off a schedule slip while another flags every day of delay to their finance department. Send the wrong tone to the wrong person and you've traded ten minutes of typing for a testy phone call that eats an hour.

Where this goes wrong

The failure mode I see most is businesses that let AI send messages without a human reading them first. It's tempting, because reviewing every message eats into the time savings. But an unreviewed AI response is a liability with your name on it. Set a rule: anything going to a client gets a human glance before it sends, even if that glance takes ten seconds. That one habit is the difference between a tool that saves an hour a day and a tool that costs a client.

Scheduling and dispatch

This one depends on the work. AI scheduling performs well when jobs are predictable in scope and duration. It struggles when every job is different and half of them run long. Know which category your business is in before spending here.

A service company doing furnace inspections, HVAC maintenance calls, or standard panel swaps has a good shot at this working well. Those jobs cluster around a known duration, and an AI dispatcher can match technician skill, location, and availability against a queue of calls faster than a person juggling a whiteboard and three phone lines. I've seen dispatch tools cut the time between a service call coming in and a tech being assigned from twenty minutes to under two.

A general contractor running custom builds has a much harder time. No two framing jobs run identically, weather blows up the schedule, and a supplier running three days late on windows cascades into six other trades shifting around it. AI can flag that cascade faster than a spreadsheet would, but it isn't going to predict that a subcontractor's other job is running long and will eat into your crew's start date. That still takes someone picking up the phone.

The middle ground

Most trades businesses aren't purely one or the other. A plumbing company might run predictable service calls in the morning and unpredictable new-construction rough-ins in the afternoon. In that case, AI scheduling earns its keep on the service side and stays out of the way on the construction side. Trying to force one tool to handle both usually means it does neither one well.

What this actually costs, and how long it takes

Say a fifteen-person electrical contractor decides to tackle estimating first. The software itself might run a few hundred dollars a month, which is the part every vendor leads with. The real cost is the setup: someone, usually the owner or the office manager, spends a few weeks pulling old job files, checking them for accuracy, and correcting the ones where the actual cost never got reconciled against the quote. That's unpaid, unglamorous work, and it's also the entire reason the tool ends up useful instead of decorative.

Most owners underestimate that timeline. They expect results in a week and get frustrated by month two when the quotes still need heavy editing. In my experience it takes a full quarter of real jobs running through the system before the output needs only light review, and closer to two quarters before you'd trust a new estimator to send a quote it generated without a second set of eyes on it. That's not a criticism of the technology. Any new estimator takes about that long to get good too. AI doesn't skip the learning curve, it just moves who's doing the learning.

The upside is that the learning compounds. A new hire who leaves takes their judgment with them. A properly set up estimating tool keeps every job it's seen, which means the fifth year of data makes the sixth year's quotes better, not worse. That's worth factoring in against the sticker price.

Documentation and the paper trail

This is the one nobody asks me about first, and it's often the easiest win. Every job generates paperwork: site photos, daily logs, safety checklists, warranty documentation, permit submissions, change order records. Someone has to produce it, and on most crews that someone is a foreman typing notes after a ten-hour day, which means it either doesn't get done or gets done badly.

AI is good at turning rough inputs into structured documents. A foreman can walk a site with a phone, narrate what he's looking at, and snap photos of anything that needs flagging. That becomes a daily log with photos attached, timestamped, and organized by location, instead of a text message to the office that says "framing done, electrical rough-in starting tuesday, need more 2x6."

Warranty documentation benefits the same way. Instead of an admin manually assembling a binder of manufacturer specs, install photos, and inspection sign-offs three weeks after a job closes, that record gets built as the job happens. It's not exciting work, but it's the kind of thing that saves you when a client calls eighteen months later asking why their deck is cupping and whether that's covered.

Subcontractor coordination

Coordinating subs is mostly information transfer: schedule changes, updated drawings, material substitutions, inspection results. AI can keep that information moving between the people who each need their slice of it, without a PM manually re-sending the same update six different ways. It won't manage the relationship, and it won't catch that a sub is quietly overcommitted across three of your jobs at once. That judgment still belongs to whoever's running the project.

Where AI doesn't fit

Site supervision, quality control, anything requiring physical judgment. AI can't look at a rough-in and tell you whether it's right. Nobody's going to sell you on that honestly yet.

I get asked about this constantly, usually by an owner who's seen a demo of a tool that flags construction defects from photos and wants to know if it can replace a site super. Some of these tools are genuinely useful for catching obvious issues, a missing fire block, a fastener pattern that's clearly off spec. What they can't do is understand context. They don't know that the framing crew deviated from the plan because the truss package arrived wrong and the super made a call on site to keep the schedule moving. A photo-scanning tool flags a deviation. A person decides whether the deviation is actually a problem.

The same goes for a client relationship built over years, or a negotiation where reading the room matters more than the numbers on the page. AI can help you prepare for that conversation. It shouldn't have it for you.

There's a version of this that gets uncomfortable fast: safety. Some vendors are now pitching AI that reviews jump-cut clips from site cameras and flags PPE violations or fall hazards. I don't think that's wrong in principle, a second set of eyes catching a guy without a harness is a good thing. But I'd treat any tool in this category as a spotter, not a supervisor. If the plan is to reduce how often a real person walks the site, that's the wrong use of it. If the plan is to catch what a person already walking the site might miss on a bad day, that's a reasonable one.

Questions worth asking before you spend anything

Most of the trouble I see doesn't come from AI being bad technology. It comes from businesses buying the wrong tool for the wrong reason. A few questions are worth running through before signing anything.

None of these questions require technical knowledge. They're the same due diligence you'd apply to hiring a new estimator or bringing on a new supplier. AI tools deserve the same scrutiny, not more and not less.

The question worth asking isn't whether AI can do something. It's whether AI doing it makes your business better or just more complicated.

Most of what I've described here didn't come from one big rollout. It came from picking a single place where the friction was obvious, fixing that, and letting the next opportunity show itself once the first one was working. That's a slower story than the ads promise, but it's the one that still holds up a year later.

If you're not sure where to start, the free assessment will show you where AI fits in your operation and what to tackle first.