Two conversations in the same week, at two different general contractors. An estimating lead told us she's running a couple of AI programs already and is trying not to inundate her team with more. A regional estimating principal put it harder: they have to slow down their tech intake or it's going to make everyone crazy.
Neither was a brush-off. Both were right, and we'd rather write the honest version of this than pretend the objection doesn't exist.
The answer isn't a number. It's a test, and most precon teams have never applied it.
Tool count isn't what's breaking teams. Unowned tools are.
Every piece of software you buy needs someone whose job includes making it work. Not a champion who likes it. An owner who is accountable for whether it's used on the next pursuit. Buy three tools and assign one owner and you don't have three tools, you have one tool and two logins nobody opens.
That's why "how many" is the wrong question. A team of six can run four tools if each one replaces a step somebody used to do by hand. The same team drowns on two if both tools add a step and remove none.
Before you buy anything, answer one question about it: does this replace a step, or add one?
Replaces a step. Somebody was doing this work manually and now they aren't. The hours come off a real person's week. Adoption is close to automatic, because the alternative is worse.
Adds a step. The tool produces something useful that nobody was producing before, so now there's a new output to review, file and act on. These can be worth it. They also need budget for the review time, and they're where AI fatigue comes from.
Most precon AI falls into the second category and gets sold as the first. Ask which one you're buying.
A senior PM acting as owner's rep at a Toronto developer gave us the cleanest threshold we've heard, from the interviews behind The Scope Gap Playbook: "If we could catch three scope gaps or three missed items on every scope of work, then this thing pays for itself."
Three items. That's the bar. It's low because the downside is asymmetric. Change orders run 8% to 14% of project cost on commercial work, and past 25% on projects with weak scope, according to Navigant figures republished by the AIA. A missed line doesn't cost you the price of the line. It costs you the change order plus the argument.
Set that threshold against each tool on your list. Anything that can't clear a defensible version of it is a pilot, not a purchase.
For what it's worth on our side of the table, ProWest Constructors put a number on it. Michael DeMarie, their President, reported roughly 50% savings in time from award to all subcontracts written. On an average $25 million project, that's award to every subcontract issued in 60 days.
There's a version of this objection that sounds like discipline and works out as drift. It goes: our estimating platform is adding AI, so we'll wait and integrate with whoever they choose.
We hear it often, and honestly the logic isn't wrong. Nobody wants to buy a tool their platform then makes redundant.
Two problems with it though. Platform roadmaps slip, and "next year" has a way of becoming the year after. And the AI your platform bundles will be built for the average customer, because that's the only thing a bundled feature can be built for. If your bottleneck is scope qualification on complex document sets, an average-case feature won't touch it.
The middle path is cheap. Run one tool on one pursuit, on a workflow you can measure, while you wait. You learn what the category can do on your documents. If your platform eventually ships something better, you switch, having lost a pilot fee instead of eighteen months.
Pick by workflow, not by category.
Look at your last three change orders and your last three lost bids. Whatever shows up twice, that's the workflow to point a tool at. If it's missed scope, look at scope tools. If it's turnaround speed on document review, look there. If it's quantities, buy a takeoff tool, and there are several good ones.
Then cap it. One new tool per quarter, one named owner, one workflow it has to improve, and a date when you decide to keep it or kill it. That cadence is slow enough that people learn the tool and fast enough that you're not the last GC in your market still reading 2,000-page spec books by hand.
If scope is the workflow that keeps showing up, Scope Agent is the one to test, and you should test it on a project you already finished. See how GCs run it, or book a walkthrough with your own documents.
As many as you have owners for. One accountable owner per tool, and one workflow each tool has to measurably improve. Teams fail on tool count only after they've failed on ownership.
You can, but run one pilot while you wait. Bundled features get built for the average customer, and platform timelines move.
Set the threshold before you start. Catching three missed scope items per scope of work is a defensible bar that operators use.
Run it on a project you've already closed out, where you know where the gaps turned up. You get a scored answer instead of a demo impression.
Sometimes, and that's the question to ask directly. Tools that replace a manual step reduce load. Tools that surface new findings add review time, which can still be worth it if the findings carry cost.
Book a 30-minute demo and see Provision running on the kind of project you bid every week.
Book a demoMore Articles
