Can AI Replace a Construction Controller?

Vendors keep promising that artificial intelligence will “run finance” for contractors. The pitch is seductive: automate the WIP, close the books overnight, and let the model flag every margin leak before the surety asks about it. Construction Dive’s recent coverage of AI-connected ERP projects shows why the industry is listening — contractors that stitch field activity to financials can see labor variances, unbilled change orders, and WIP drift while the job is still open, not after closeout.

That is real progress. It is not a Controller.

A construction Controller’s job is not to type invoices faster. It is to own the estimates, the controls, and the story the lender and the bonding agent are reading. AI can clear the runway. It cannot sit in the chair.

What AI is already good at in construction finance

When the data model is construction-specific, AI and automation earn their keep on volume work:

- Invoice capture, three-way matching against purchase orders and subcontract commitments

- Bank reconciliation and routine GL categorization

- Pulling current job-cost activity so the close is not a three-week scavenger hunt

- Exception flags when cost run rates or billing lag look off-pattern

Industry write-ups make the same point from different angles. Wiss’s May 2026 overview of AI-powered construction accounting frames the win as transaction throughput and fresher job-cost data between closes — not as a substitute for percentage-of-completion judgment. Construction Dive’s sponsored reporting on AI and profit leaks similarly stresses early exception visibility: connect project and financial data, surface variances while corrective action is still possible, and leave the call on what to do with humans.

Treat that layer as an assistant. Deloitte’s Bradley Niedzielski, writing in the Journal of Accountancy (January 2026), put the control posture plainly: treat AI as a responsive assistant that processes and highlights data, and keep a human in the loop for validation. That framing fits construction finance better than any “autopilot Controller” slogan.

Where U.S. GAAP still demands a person

Most long-term construction contracts recognize revenue over time under ASC 606. FASB’s criteria in ASC 606-10-25-27 ask whether control transfers as the work progresses — for example, because the contractor is creating or enhancing an asset the customer controls (work on the owner’s land), or because the asset has no alternative use and the contractor has an enforceable right to payment for performance completed to date.

Once over-time recognition applies, progress measurement — often a cost-to-cost input method — turns on estimated cost at completion. That estimate is not a scrape from yesterday’s AP feed. It is a judgment about remaining scope, labor productivity, subcontractor performance, change-order probability, and whether certain costs (including uninstalled materials in some cases) actually depict progress toward satisfying the performance obligation.

A model can draft a WIP schedule from posted costs and last month’s EAC. A Controller still has to:

- Challenge optimistic cost-to-complete inputs from the field

- Decide how change orders and claims enter the transaction price

- Separate contract assets and contract liabilities correctly by contract

- Explain gain/fade and over-/under-billing to the banker and the surety

- Own the internal-control story when AI touched the numbers

Wiss is explicit on this split: AI helps WIP accuracy by keeping underlying data current and structured; the professional judgment on estimated cost-to-complete, gain/fade, and billings-versus-costs positions remains human work. That is the right bright line.

Controls did not get an AI exemption

As finance teams wire generative and agentic tools into close and estimating workflows, governance expectations are moving upstream. COSO’s February 2026 guidance on internal control over generative AI (discussed widely in 2026 risk and controls commentary, including EY’s coverage) emphasizes contemporaneous evidence of how AI-influenced judgments were reviewed and challenged — not just a tidy final number. The Journal of Accountancy piece above lands in the same place: identify where AI sits in the stack, match review protocols to the process, and keep human oversight on exceptions.

For a contractor, that means documenting:

- Which AI tools touch job cost, WIP, or revenue entries

- Who reviews model output before it posts

- How prompt/version/input evidence is retained when material estimates change

- What happens when the model and the project manager disagree on EAC

Blind trust in a polished WIP export is an internal-control failure wearing a modern interface.

What a strong construction Controller does with AI

The firms that will win this transition are not the ones that fire the Controller and “let the ERP think.” They are the ones that redeploy Controller time:

1. Cleaner inputs — AI clears invoice matching and reconciliation so month-end starts with current costs, not a backlog.

2. Sharper estimates — Controllers spend the hours that used to go to data wrangling on EAC challenge meetings and gain/fade analysis.

3. Earlier intervention — Exception dashboards (the Construction Dive theme) become a weekly operating rhythm, not a surprise in the bonding package.

4. Lender-ready narrative — Someone still has to explain why Job 47’s percentage complete moved, why retainage timing hit cash, and why the underbill is temporary rather than structural.

Fractional CFO and Controller support for real estate and construction is built around that judgment layer. Automation is a tool in the kit. Accountability stays with the person who signs off on the package.

Bottom line

Can AI replace a construction Controller? No.

Can AI make a construction Controller dramatically more effective? Yes — if you buy construction-aware workflows, keep ASC 606 estimates under human ownership, and design controls that assume the model will be wrong in plausible-looking ways.

If your close still burns most of its calendar gathering job costs so nobody has time to challenge the estimates, AI belongs in the stack. If someone is pitching you a Controller-shaped vacancy filled by a chatbot, keep your bonding agent’s phone number handy — and keep a Controller who knows what the WIP is actually saying.

Sources

Construction Dive (sponsored), “AI helps contractors insulate against profit leaks” (discussing connected WIP/budget visibility and AI exception surfacing while jobs are in progress): https://www.constructiondive.com/spons/ai-helps-contractors-insulate-against-profit-leaks/827876/

Wiss, “AI-Powered Accounting for Construction Companies” (May 19, 2026) — transaction automation versus WIP / cost-to-complete professional judgment: https://wiss.com/ai-powered-accounting-for-construction-companies/

FASB Accounting Standards Codification ASC 606-10-25-27 (criteria for recognizing revenue over time when control transfers as performance occurs), including the create-or-enhance and no-alternative-use / enforceable-right-to-payment pathways commonly relevant to construction contracts. See also PwC Viewpoint, “6.3 Performance obligations satisfied over time”: https://viewpoint.pwc.com/dt/us/en/pwc/accounting_guides/revenue_from_contrac/revenue_from_contrac_US/chapter_6_recognizin_US/63performance_obliga_US.html

Journal of Accountancy, “Shaping AI governance and controls” (January 1, 2026), featuring Bradley Niedzielski (Deloitte) on human-in-the-loop oversight for AI in finance: https://www.journalofaccountancy.com/issues/2026/jan/shaping-ai-governance-and-controls/

EY, “COSO 2026 and the shift in AI governance” (on COSO’s February 2026 Achieving Effective Internal Control Over Generative AI and upstream evidence of AI-influenced judgments): https://www.ey.com/en_us/cro-risk/coso-2026-and-the-shift-in-ai-governance

Ultramar provides fractional CFO and controller support for real estate developers, contractors, and property operators. This post is for general information and is not accounting, tax, or legal advice.