What Construction CPAs Really Think About AI

Ahead of our open-room roundtable at the second annual Construction Intelligence Summit, we asked the CPA partners in attendance to write down what they are genuinely seeing with AI across their construction clients. Not predictions, not vendor talking points. What is actually in production, what clients are asking about, and what worries them. More than a handful sent answers back, and more weighed in live, in a room that also held the contractors they advise.

That input is worth more than any single case study, and here is why. A contractor sees one company's books. A construction CPA sees dozens. When somebody who reviews dozens of sets of books a year tells you where AI is landing, they are describing a pattern, not an anecdote. And the room was deep on that front: attendees included leaders from firms on Accounting Today's 2026 Top 100, the people setting technology direction for a meaningful slice of the construction accounting market.

They did not see each other's answers. They converged anyway.

AI is already in the building. It’s just in the back office.

Ask where AI is showing up in contractors' businesses right now, and the answers cluster hard. Not jobsite robotics. Not autonomous scheduling. Invoices.

Jason Bainter at CBIZ pointed straight at AP, "A lot of accounting systems currently have AI integration where invoices get routed via AI to the respective PMs or invoice approvers, and then once signed, sent to AP processing for payment." Brian Muncy, also at CBIZ, listed accounts payable processing and approval, drawing takeoffs and estimating, and reviewing and summarizing contracts. Lisa Baalman at Pinion named accounting automation, document processing, job costing, and reporting. Chris Mast at GBQ described "relatively contained tasks," including drafting communications, summarizing contracts and meeting notes, searching project documents, and bid-related workflows.

The pattern is clear enough to state plainly. AI is landing first on repetitive, document-shaped, high-volume work. It is not landing on judgment. As Chris put it, fully automated financial decision-making "is still much more talk than reality for most contractors."

If you are a contractor wondering where to start, that cluster is your shortlist.

The question contractors are not asking

Here is the part that stopped us.

We asked what the most common AI question is that contractors bring to their CPA. Chris's answer: "I'm not getting questions from my clients on AI, which is surprising."

That is worth sitting with. Either contractors aren't there yet, or they don't think of their accountant as the person to ask. Both are a problem, because the accountant is the one who can tell you whether your books can support the thing you are about to buy.

The questions split along two lines. Jason hears governance and usage policy, meaning who is allowed to put what into which tool. Lisa hears the ROI question. Clients "want to know whether AI will actually save administrative time and improve profitability, or if it's just another software feature with a lot of marketing behind it."

Both are fair, and neither one gets answered by a demo. You answer them with pointed questions for the vendor before you sign anything: how they secure your data. Whether the tool actually connects to your ERP. Whether it shows its sources or just hands you a number. We walked through that list in AI for Contractors: What Questions to Ask AI Vendors Before You Buy, and it is a fair test of any vendor in this space, including us. If a salesperson cannot answer those, the ROI question has already answered itself.

False confidence is the actual risk

Ask a CPA what worries them, and you do not get "the robots will take over." You get something much more specific.

Chris named it: "False confidence. AI can produce a polished and plausible answer even when the data are incomplete, the assumptions are wrong, or the question requires construction-specific context and judgment."

Then he took it further, and this is the line we keep coming back to. If accounting, payroll, job-cost, scheduling, purchasing, and field data don't reconcile or don't arrive on time, "AI can simply make an unreliable process faster and more convincing."

Faster and more convincing. Not more accurate. That is the trap, and it is worse than the old problem. A bad spreadsheet looks like a bad spreadsheet. A bad answer from an AI tool looks exactly like a good one.

Lisa landed in the same place from a different direction, describing clients who "expect clean reporting from inconsistent or incomplete job-cost data." Chris's version of the expectation gap is the sharpest we heard: contractors expect AI to find the answer to a margin or cash-flow question on its own, when in reality it mostly accelerates analysis of the information they already captured, correctly or incorrectly.

Now here is the part that complicates all of that

Every CPA who warned us about false confidence is using AI. Daily.

That is not a contradiction. It is the whole lesson. They are using it in the exact places where a wrong answer gets caught by a human before it goes anywhere.

One firm has gone furthest. They are running AI against the WIP schedule to determine what questions to ask the client during analytics. They use it as a first pass on general ledger review to flag anomalies, and for both preliminary and final analytical reviews. Other offices in the same firm use it to build financial dashboards and KPI analysis from five years of a client's historical financials.

That is the most concrete answer in this entire piece. AI is not producing the analysis. It decides what to look at and generates the questions a human then asks.

The others describe the same shape of work. One partner finds it most useful as a first-draft and research assistant, organizing questions, improving explanations, summarizing material, and identifying issues worth investigating, which frees up time to apply professional judgment to a client's specific facts. Another uses it to organize research, summarize information, draft, and accelerate brainstorming, for the same reason: more room for client-specific analysis and recommendations. A third kept it to five words. Summarizing information and better illustrations.

Contract review came up repeatedly, along with pulling historical job performance forward so it can inform the next decision instead of sitting in an archive.

Notice what is missing from all of it. Nobody is letting AI conclude. It drafts, it summarizes, it flags, it organizes, it points. The judgment stays with the person whose name goes on the work.

That is the model contractors should copy. Not because it is cautious, but because it is where the value actually is. The bottleneck in most finance departments was never the analysis itself. It was getting to the right question fast enough to matter.

The one place the room did not agree

It was interesting to see where the group disagreed on how AI should be used.

Jason and Brian lean toward AI as part of the fix. Jason sees cash flow improving as AR and AP get streamlined, and sophisticated controllers and CFOs using AI to hunt anomalies in the general ledger. Brian points to better AR collection, over- and under-budget summaries, and cash flow projections. In that view, AI helps you make sense of data that was always going to be a little messy.

Chris treats messy data as the ceiling rather than the starting point. Disconnected project management, accounting, and field systems limit the inputs AI needs for meaningful insight.

Lisa sits between them. Success when AI captures and classifies transactions, disappointment when expectations outrun the underlying data.

Nobody in that group is wrong. The difference is how far below the ceiling your company currently sits.

So the practical question is not whether to use AI, but where your data stands today. If your project management, accounting, and field systems already talk to each other, the wins Jason and Brian describe are within reach now. If they don't, Chris's point applies, and the first step may be connecting the data rather than adding a new tool on top. Either way, Lisa's experience is a good guide: set expectations by what your data can support today, and raise them as your data gets better.

What has to be true first

We asked what has to be true about a contractor's books before AI is genuinely useful. Here is a checklist, and it doubles as a maturity test:

Timely closes
Disciplined job-cost coding
Clear ownership of master data
Consistent change-order and WIP processes
Enough confidence that the inputs reconcile to the books

Read that list again and notice what it is not. Nothing on it is new, none of it is AI-specific, and every item was a good idea a decade ago. It is nearly the same list this group gave us at our first Summit, back when the question had nothing to do with AI and was simply 10 things your CPA wishes you did. Monthly job status meetings. WIP that ties to the GL. Live cost-to-complete instead of a post-mortem. Slicing the data you already have.

The advice did not change. It just got a lot more expensive to ignore.

Where this leaves you

Five things the room converged on, in no particular order:

Data quality is the gate. AI is only as good as the books underneath it.
False confidence is the real risk. A confident answer built on incomplete data is more dangerous than no answer at all.
AI accelerates; it does not decide. It speeds up analysis of what you already captured. A human still owns the judgment, the client-specific recommendation, and anything touching a filing, a covenant, or a decision.

It lands first in the back office. AP, invoices, cash, and reporting. The repetitive, data-heavy work.
The fix is unglamorous. Timely closes, clean job-cost coding, and systems that reconcile.

That last one is the whole thing. The interesting AI conversation and the boring accounting conversation turned out to be the same conversation, which is exactly what the accountants in the room have been telling contractors for years.

So start in the right order. Ask your CPA whether your job-cost data would survive the question you want AI to answer. Then take the vendor questions into the sales conversation. The tool matters less than the two things on either side of it.

When you are ready to see what this looks like on your own numbers, start a free trial of Nova and ask it something your current reports make you wait for.