For the last couple of years, the question contractors got asked about AI was whether they'd tried it. That question is getting stale. The one replacing it is harder to answer: what did it actually do for the business?
A survey released this spring puts numbers on that shift for commercial construction, and the numbers are more useful than most adoption headlines because they don't just count who is using AI. They count who can point to a result.
What the survey measured
ServiceTitan published its 2026 Commercial Specialty Contractor Industry Report on March 30, 2026. The fieldwork was done by Thrive Analytics, an independent research firm, and it covered more than 1,000 owners, executives, general managers and directors at commercial building firms.
Two things are worth knowing before leaning on any of it.
First, the scope. This is commercial specialty contractor data. It isn't residential trades, it isn't manufacturing, and it isn't Canadian. The workflows are the same ones Okanagan contractors run every day, but the sample skews toward larger commercial firms, not a two-truck shop.
Second, the sponsor. ServiceTitan sells software to contractors, including AI features. That doesn't make the data wrong. The sample size and the research firm are disclosed, and the fieldwork was run by a third party. It does mean the company publishing the numbers has a commercial reason to want AI adoption to look strong, and it's fair to read the results with that in mind. The full report sits behind a contact form; every figure in this article appears in the public press release.
The number doubled, and most firms are still on the wrong side of it
In 2025, 17% of the contractors surveyed said AI was delivering a measurable impact on their business. In 2026, that figure was 38%. More than double, in one year.
That is not the same as saying adoption doubled. The question was about measurable business impact: a result someone could name. A bid that landed. An estimate that came together faster. A cost that got caught before it turned into a change order.
Read the same number the other way, though, and it's less of a victory lap. If 38% can point to a measurable result, 62% can't. Some of those firms haven't started. Others are running AI tools somewhere in the business without any way of saying what changed because of them.
That second group is the one worth paying attention to, because it's the easiest group to end up in. Turning a tool on takes an afternoon. Knowing whether it earned its keep takes a decision about what you were trying to improve, made before you started.
Where the results actually showed up
The impact wasn't spread evenly across the business. The report breaks out where contractors are applying AI, and two areas lead: cost estimation and budgeting at 24%, and bid management at 22%.
Not dispatch. Not field operations. The desk where the bid gets built, before anyone shows up to do the work.
That lines up with how a construction job actually makes or loses money. The framing and the finish work matter, but the margin on a job is mostly decided by the number on the bid. Get the takeoff wrong, miss a scope item, carry stale material pricing, and no amount of good field execution gets that money back.
Why those two workflows
Estimating and bid management have a few things in common that make them a natural first home for AI.
- They're math heavy and repetitive. The same kinds of calculations come up job after job, which is the kind of work software handles well.
- They have an expensive failure mode. A wrong number on a takeoff doesn't just cost time. It costs a job you should have won or a margin you thought you had.
- The result is easy to see. Win rate, turnaround time and estimate accuracy are already things most firms track, even informally.
- They're lower risk than the field. Testing something new on a bid doesn't put a crew, an inspection or a pour schedule at risk.
Compare that with running AI on an active job site, with subs, inspections and weather all in play. The stakes are higher and the proof comes slower. Estimating is where a contractor can find out whether a tool works without betting a project on it.
If you want the longer version of which parts of estimating AI can and can't help with, I covered that in AI for Construction Estimating: Where It Belongs (and Where It Doesn't). The short version is that estimating is several jobs stacked together, and AI only belongs in some of them.
The pressure behind the numbers
The same report explains a lot about why contractors are moving now, and it has less to do with AI headlines than with margins.
Demand is holding. More than three-quarters of the contractors surveyed reported at least nine months of secured work, and 41% said they feel optimistic about market conditions. At the same time, 71% reported rising wages, up from 55% the year before, and material costs are climbing too.
Put those together and the problem is clear. Plenty of work, thinner margin on each job. That pushes owners toward protecting profit on work they've already won rather than only chasing more of it, and the estimate is the first place profit gets protected or lost.
That framing matters. It's a margin problem first and a technology problem second. AI is one way to work on it. It isn't automatic, and it isn't the only lever. A firm with sloppy change-order discipline or a pricing database nobody has updated in a year has cheaper fixes available than a new tool.
Only one contractor in five runs on a single system
The finding in this report that should worry owners most isn't about AI at all. Only 20% of the contractors surveyed operate on a single platform. Everyone else is stitching together separate systems for accounting, estimating, scheduling and project tracking.
That isn't new in this industry. Job costing in one place and scheduling in another has been normal for a long time. What's new is putting AI on top of that arrangement and expecting a clean answer back.
An AI tool pulling from four disconnected systems is doing the same guessing your office staff already does when the numbers don't match between them. It just does it faster. Whatever is wrong with the data gets copied more quickly, not corrected.
Adding AI to disconnected systems doesn't fix the disconnection. It gives you one more system working from the same incomplete picture.
This is the part of AI readiness most firms skip, and it's easy to see why. Buying a new tool feels like progress. Auditing where your data lives and whether your systems agree with each other feels like overhead. But the second job decides whether the first one pays off.
Using it is not the same as proving it
There's a simple test that separates the 38% from everyone else, and it has nothing to do with which tool got purchased.
Using AI means a tool is running somewhere in the business. Proving it worked means you can name what changed because of it, in terms you'd put on a job cost report: a specific number of hours off a takeoff, a quote turned around in a day instead of three, a bid that closed that probably wouldn't have.
Most owners can describe what their software does. Far fewer can describe what it changed. That gap is where a lot of AI spending quietly goes to waste, because a tool nobody can evaluate is a tool that either gets renewed out of habit or abandoned on a hunch.
Start with a number you already track
Taken together, the three findings in this report point in one direction. Measurable impact more than doubled. The gains cluster in estimating and bid management, not in company-wide rollouts. And firms without connected systems are working against themselves before they start.
None of that argues for buying more software. It argues for picking one place in the business where a number is already being tracked, like estimate turnaround, win rate or estimate-to-actual variance, and putting the tool there first. If you can't measure the before, you won't be able to prove the after.
That's also why the contractors seeing results didn't apply AI broadly and hope. They pointed it at the one or two places where a mistake already had a dollar figure attached, which made the result easy to see when it came.
Questions worth asking before you buy the next tool
Most of the AI features contractors will see this year will arrive inside software they already pay for, or through a sales call from a vendor with a demo that looks very good. A few questions help separate a tool that can be measured from one that can only be believed in.
- Which number in my business is this supposed to move, and do I know what that number is today?
- Where does the tool get its data, and is that data current and in one place, or spread across systems that disagree?
- Who checks the output before it reaches a client, and how long does that check take?
- After ninety days, what would I need to see to keep paying for it, and what would make me cancel?
None of those questions require technical knowledge. They're the same questions you'd ask before buying a new piece of equipment for the shop. The difference is that equipment comes with an obvious way to tell whether it's working, and software usually doesn't unless you set one up.
If the answer to the first question is "I'm not sure," that's worth knowing before the purchase order, not after. It doesn't mean the tool is bad. It means there's no way to tell yet, and the vendor's case study isn't a substitute for your own before-and-after.
What this means for an Okanagan contractor
A U.S. commercial survey is not a description of a mid-sized mechanical contractor in Vernon or a framing outfit in Kelowna. The percentages belong to the firms that answered it. I'd be careful about treating 38% as a benchmark for a local business.
The pattern underneath it travels well, though. The estimate is where the money gets decided on any job, in any market. Fragmented systems are close to universal in the trades, regardless of firm size. And the difference between using a tool and being able to say what it did is the same at fifteen employees as it is at five hundred.
If anything, a smaller shop has an advantage here. Fewer systems to untangle, fewer people to train, and an owner who can usually tell you from memory how long a typical estimate takes. That's a usable baseline, and it's more than many larger firms have.
The firms that will be counted in next year's version of this number aren't the ones that bought the most. They're the ones that decided, before spending anything, what result they were looking for and how they'd know when they got it.
That decision is what the AI Discovery and Readiness Assessment is for. It's a paid engagement: a discovery call and a scored review of your data, workflows and systems, ending in a written report on where AI would move a number in your business and where it wouldn't. If you'd rather know which side of the 38% you're on before spending on tools, that's the place to start.