The honest answer for most mechanical, electrical and plumbing contractors is on-pace, and on-pace is a worse position than it sounds.
Here is the benchmark. ServiceTitan surveyed more than a thousand commercial construction leaders and found 38% now report measurable business impact from AI, up from 17% the year before. Cost estimation and budgeting accounted for 24% of use, bid management 22%. A separate Bluebeam survey of a thousand AEC professionals put current AI use at 27%, with 94% of those users planning to expand in 2026.
Read the direction rather than the level. A number that doubles in twelve months is not describing a technology curve. It is describing a market crossing from the firms that experiment to the firms that expect. Somewhere in the next two bid cycles, the general contractors and owners you work for stop treating AI-assisted preconstruction as a differentiator worth mentioning and start treating it as table stakes they no longer ask about.
Where the three groups sit
Early means you have at least one workflow in production with a number attached to it. Not a pilot. Not a license somebody bought. A specific process where you know what it cost before and what it costs now. The tell is that you can say which two jobs you bid last quarter that you would have passed on a year ago.
On-pace is the crowded middle, and it looks more comfortable than it is. Someone in precon runs AI takeoff on the jobs where it is convenient. A PM uses a chat tool for RFI first drafts. There is no standard, no baseline, and no owner. The work is happening, but it is not accumulating.
Late is not the firm that has done nothing. It is the firm whose documents are in no condition to be read. Subcontracts as scanned PDFs. Historical cost data across three systems that do not talk. Change order records that live in a superintendent's email. That firm can buy any tool on the market tomorrow and get thin results, because the constraint was never the software.
What compounds and what does not
This is the part that makes the timing question real rather than rhetorical, and it is easy to miss.
Software does not compound. If you buy the same estimating platform your competitor bought, you get the same capability on roughly the same timeline, and any advantage lasts until they sign their own contract. Vendors sell to everyone. Capability is rentable.
What compounds is the asset the tools run on. Your historical cost data, cleaned and structured, gets more valuable every job it absorbs. Your subcontract library, once machine-readable, keeps paying out on every claim window it catches. Your coordination models, once they reliably drive fabrication, expand what you can prefab on the next job and the one after. And under Procore's newer company-specific instruction, the standards you teach a system become an operating asset that a competitor cannot buy off a price sheet.
None of that can be bought late at a premium. It has to be accumulated, and accumulation takes calendar time you cannot compress by spending more. A firm that starts structuring its cost history this quarter is eighteen months ahead of one that starts next year, and the gap widens rather than closes, because the leader's tools now work better on cleaner inputs.
The urgency, in other words, is not that you will miss out on a tool. It is that the input the tools need takes longer to build than the tools take to buy.
The labor math is the deadline
There is a second clock running. Estimating desks sit open for two and three months in this market. Preconstruction and VDC roles are competitive to fill. Wages are climbing, with 71% of contractors in the ServiceTitan survey reporting rising wages against 55% the year prior.
You cannot hire your way to more bid capacity at a price that works. That is not a technology argument, it is a labor market condition, and it is what turns adoption from a strategic preference into an operating necessity. The contractors pulling ahead are not the ones with the best software. They are the ones who freed two days a month per estimator and turned it into bids they previously declined.
What to do in the next ninety days
Pick one workflow. Symbol takeoff, spec review, or subcontract notice extraction are the three with the clearest denominators for an MEP shop. Write down the current number before you touch anything, because a baseline you construct after the fact is a story rather than a measurement.
Then do the unglamorous half. Get the subcontracts out of scanned PDFs. Consolidate the cost history. Name one person who owns the verification standard and has authority to enforce it. That work is not exciting, it does not demo well, and it is the entire difference between a tool that compounds and a subscription that renews.
Being late is survivable. Plenty of firms have arrived second at a technology and done fine. What is harder to recover from is arriving with nothing for the technology to read, while a competitor three years into structuring their data bids the same job with better information and a faster turnaround.
The party is not the point. The invitation list for the next round of work is.