Sixty-one percent said AI. Far fewer said which part.

The adoption curve in construction has stopped being the interesting question. The number moved a long way in twelve months and it is going to keep moving.

What that figure does not tell you is where the investment is pointed, and in preconstruction the where matters more than the whether.

The crowded end

Takeoff has attracted most of the attention, for understandable reasons. It is visual, it is measurable, and the before-and-after demonstrates well. Vendors in that category report compressing manual takeoff from most of a day to well under an hour.

Those are vendor figures rather than independent ones, and worth treating as such. But the direction is real and the category is genuinely competitive.

The end nobody photographs

Takeoff was never where most bids were lost. A bid is lost to a mandatory attendance clause in the front end, to a specification that contradicts its own schedule, to a vendor quote whose stated total disagrees with its own itemisation, and to nine of those quotes landing in the last hour before close.

Quantities were rarely the problem. The documents around the quantities were.

None of that is a measurement problem. It is a reading and coordination problem, which is unglamorous, hard to demonstrate in thirty seconds, and where the actual money sits.

Which is why the gate matters more, not less

The moment AI moves from measuring quantities to interpreting obligations, the failure mode changes. A takeoff that is five percent light is a number you can sanity-check. A misread indemnity clause is a liability you accepted without noticing.

So the question to ask any preconstruction tool is not whether it uses AI. It is what happens at the boundary: what can it do without asking, what must it get approval for, and can it prove which of those two applied on any given day.

If the answer is a dialog box the agent could route around, that is not a boundary. It is a courtesy.