The Deloitte Flash for Construction

THE ISSUE:

As construction and real estate organizations recognize the efficiency AI can unlock—faster reviews, sharper forecasting, less manual reconciliation—adoption is accelerating well ahead of the structures needed to govern it. Tools are being deployed project by project, often without a clear answer to two basic questions: who is responsible when the output is wrong, and what needs to be in place before that question is asked in a deposition, a board meeting, or an audit?

That gap is not theoretical. Many procurement decisions are being made based on AI-assisted analysis with no audit trail. Safety and quality conclusions are being drawn from models no one on the project team can explain. Contracts are being executed that predate the tools now running against them. Clear ownership, defined oversight roles, and enterprise-wide accountability are lagging well behind the pace of adoption, a pattern showing up broadly across industries, not just construction.

Construction carries more of this exposure than most industries. The sector has historically been slow to digitize, project data is scattered across systems that rarely talk to each other, and a contractual environment built on standard-of-care obligations and payment certifications creates risk other sectors do not carry. The same Generative AI tools accelerating efficiency are also enabling a new class of fraud including fictitious invoices, falsified certifications, and fabricated worker records that are increasingly difficult for reviewers to catch. For capital program managers, that combination of speed, opacity, and structural exposure is a measurable and real-time risk, not a future one.

INSIGHTS:

Not every organization is equally exposed to that risk. Adoption and governance are not competing priorities; the organizations moving fastest right now are the ones that built governance in from the start. Across construction and real estate, AI use has already moved past pilots: many teams are applying it to permitting and drawing review, construction cost and schedule monitoring, capital planning scenario analysis, and site safety and visual quality inspection. The stakes of getting governance wrong are real: there is financial exposure from misinformed decisions and legal exposure from the compliance gaps described above. Reputational exposure alone drives roughly 26% of S&P 500 market capitalization,³ value that AI failures such as biased outputs, fraudulent use, governance breakdowns can erode quickly. What separates those organizations is not caution; it is a consistent set of principles, scaled to the stakes of each use case, applied wherever AI touches a consequential decision.

Deloitte's Trustworthy AI™ framework² spans seven dimensions in total. The five most relevant to construction risk are below, each with an example of how it shows up on a program:

• Transparent — Decisions need to be auditable and open to inspection before they reach a decision-maker. A flagged drawing conflict or a cost-overrun projection should come with a clear rationale that an owner, auditor, or legal counsel can follow.
• Accountable — Ownership of each AI-assisted determination should be assigned before the tool goes live, so responsibility is already established by the time a permitting review or dispute puts it to the test.
• Secure — AI systems and the project data behind them including bid information, cost models, and unreleased drawings need protection against unauthorized access, manipulation, and misuse, whether from external attackers or unmanaged use of public AI tools outside the organization's own controls.
• Fair — Models perform equitably across the full range of project types, delivery methods, and geographies they are deployed on, rather than reflecting only the narrow slice of projects they were trained on.
• Reliable — Programs should start with lower-stakes functions such as document summarization or capital-planning scenario modeling, validate results before scaling to safety-critical or contractual decisions, and monitor deployed tools on an ongoing basis to catch drift as conditions, data, and usage change.

Those five dimensions describe what a trustworthy AI system looks like. Building the organizational muscle to sustain them comes down to a few practices:

• Tone from the top — executive-level ownership of AI governance, not a delegated IT policy
• Data integrity — AI outputs are only as reliable as the data behind them, a particular challenge in construction, where project data is often fragmented and unsourced across systems and phases
• Written policy — decision rights, escalation paths, and accountability defined before a tool goes live, not after a dispute
• Adoption strategy — a deliberate sequence for rolling out AI across use cases, prioritized by risk and readiness, so governance keeps pace with expansion instead of trailing behind ad hoc deployments

Together, these give owners and program managers the confidence to expand AI use deliberately, rather than reactively pulling back after the first costly mistake.

HOW DELOITTE CAN HELP:

Deloitte brings deep experience across governance, risk, and AI-enabled program controls to help construction owners and capital program managers stand up an operating model built on executive ownership, clean data, and policy in writing before a tool goes live. We assist clients in translating governance principles into practical, auditable workflows without slowing adoption. Deloitte can help:

• Assess AI exposure and readiness by inventorying the tools already in use, evaluating data quality, and prioritizing where governance gaps carry the most risk
• Test and validate high-risk use cases before or after deployment so reliability is demonstrated rather than assumed
• Build monitoring and observability frameworks, including purpose-built agents that watch other AI systems, to catch drift as tools operate over time
• Update risk taxonomies and control libraries so existing risk management materials reflect AI-specific exposures rather than frameworks built for other risks
• Design governance structures, decision rights, and a phased adoption strategy that scales oversight with program risk
• Manage cost and technical efficiency, including usage and tokenization economics, as adoption scales

Getting governance right early costs far less than retrofitting it after a dispute, audit, or public incident. To talk through what this looks like for your program, contact one of our leaders below.

This Flash is adapted from Deloitte’s Infrastructure & Real Estate presentation “Building with Confidence: AI Done Right,” to the Construction Users Roundtable (CURT) at the Leadership in the Age of Artificial Intelligence summit, June 2026.

¹ “From Principles to Practice: A Benchmark Study in AI Governance” American Arbitration Association, 2026.

² "Fraud in the Cyber Era: 2026 Fraud Trends & Insights" Trustpair, 2026.

³ “Reputation new risk hedge: $13.8 trillion in shareholder value tied to corporate trust” Echo Research, 2025.

Previous I&CP Editions