Data Governance Before AI: A Practical Checklist for Construction IT

Data Governance Before AI: A Practical Checklist for Construction IT

Key Insights:

  • Only 26% of U.S. contractors in a 2025 Dodge survey rate their data quality as high. Governance decides whether AI outputs deserve trust.

  • Every core data domain, from vendors to cost codes, needs a named owner, a written definition, and a map from retired codes to current ones.

  • AI tools inherit your permission model, and owner contracts may limit what project data AI can read. Review both before any rollout.

  • Documents such as RFIs, submittals, and daily logs need the same lineage, version control, and audit trails as ERP records.

A written AI use policy, a short set of governance metrics, and one logged pilot show whether you are ready to scale.

Construction companies have spent years digitizing bids, schedules, payroll, and field reports. Those records now live in ERP systems, the software that runs accounting, project controls, and workforce data in one place. AI promises to turn that history into forecasts, alerts, and automated tasks, but only when data governance comes first.

The returns depend on the data underneath. In a 2025 Dodge Construction Network survey, only 26% of contractors rate their current data quality as high.

Governance closes that gap. This checklist covers ownership, access, lineage, and AI use policy, plus the contract terms and metrics that complete the picture.

Why Does Data Governance Have to Come before AI in Construction?

AI models find patterns in the data you give them and act on those patterns at speed. If a vendor exists under three spellings, or a cost code means different things on two jobs, the model treats each version as fact.

The Dodge research shows how widespread the problem is. Beyond the 26% who rated their data quality as high, 58% acknowledged it was only moderate. Over half of contractors also expressed concerns about data accuracy (57%) and security (54%).

Global findings point the same way. In the RICS AI in Construction 2025 report, 30% of respondents cited data quality and availability as a barrier. Another 25% cited a lack of standards and guidance, which RICS says may leave companies unsure how to assess or govern AI use responsibly.

Governance answers that uncertainty. It is the set of controls that keeps records correct after a cleanup ends. It assigns ownership, sets access rules, records where each value came from, and defines what AI may do with it.

The stakes rise with AI agents, software that carries out tasks such as coding invoices or drafting forecasts. In a manual process, a bad record usually misleads one report and the person reading it. An agent can read that same record across every job it touches and act before anyone reviews the result.

That reach explains why governance starts with ownership. EveryAI deployment eventually asks who owns a given record and who may change it.

Construction Data Governance Starts with Ownership

Ownership comes first because every other control depends on it. Access rules, lineage requirements, and AI policies all need someone with the authority to approve them.

Start with master data, the shared reference records that every module draws from. In construction, that includes vendors and subcontractors, cost codes and job structures, employees and union classifications, equipment, and legal entities. A duplicate vendor or a misused cost code at this layer spreads into every report, payment, and forecast that relies on it.

Each domain needs two named roles. The business owner, often a controller or operations lead, sets the rules for that data. The data steward applies those rules daily, reviews new records, and merges duplicates.

Write the definitions down. Terms such as committed cost, percent complete, and approved change order can mean slightly different things to project and finance teams. A shared data dictionary gives every user, and every AI tool, one meaning for each field.

History needs the same attention. Cost code structures change over the years, and AI models compare current jobs against closed ones. Map retired codes to current ones before a model reads your job history, or it will treat mismatched costs as equivalent.

The standardization gap runs wide. Dodge Construction Network research on project controls, published in 2026, found that only 11% of contractors have the data integration, optimized processes, and real-time visibility in their project controls needed to achieve reliable project outcomes. IT closes part of that gap inside the system. Required fields,validation rules, and approval workflows turn written policy into the default path for every user.

Ownership settles who decides. The next control settles who can see each record and what your contracts allow.

Access Controls, Contract Rights, and Data Lineage

AI tools inherit your permission model. If a project manager can view labor rates on every job, an AI assistant acting for that manager can read and summarize them too. That makes access design an AI design decision.

Review roles before any rollout. Map each role to the jobs, entities, and record types it needs, then remove permissions that piled up through years of promotions and staff changes. Payroll, union rates, and owner contract terms deserve the closest review.

Your contracts set a second boundary. Guidance from Seyfarth Shaw advises that contracts should clearly define who owns inputs and outputs, and whether vendors may use project data for training or other purposes. Michelman & Robinson counsel adds that standard form agreements, including widely used documents from the American Institute of Architects (AIA) and ConsensusDocs, were not drafted with AI in mind. Check owner agreements, joint venture terms, and subcontracts before AI reads a project’s files.

Lineage answers a third question: where did this value come from? Every record an AI model reads should carry its origin, timestamp, author, and approval status.When a forecast looks wrong, your team can trace the input behind it and correct the source.

Apply the same controls to documents. Project teams already use AI to draft or summarize RFIs, submittals, and meeting minutes, which calls for clarity around human review, version control, and document retention. Audit trails should show who changed each document, when, and why, including changes an AI agent makes.

With these controls in place, your data has a sound foundation. The last control governs the AI tools that use it.

What Should an AI Use Policy for Construction Data Include?

Start with an inventory. RICS found that 7% of respondents were unsure about their organization’s AI adoption status, which may stem from AI capabilities sitting inside tools without clear labeling. List every AI feature already running in your software before you write rules for new ones.

The policy itself names approved tools and the data each may read, with payroll and personal employee records off-limits by default. It also sets human review before outputs reach contracts, pay applications, or financial statements. Agent permissions need dollar limits, especially forAI agents that create job cost transactions or change orders.

External rules give the policy its benchmarks. Under the EU AI Act, AI tools for employment and management of workers count as high-risk, and obligations including activity logging and human oversight apply from December 2027. ISO/IEC 42001 sets requirements for an AI management system. The National Institute of Standards and Technology (NIST) AI Risk Management Framework offers voluntary guidance built around four functions: govern, map, measure, and manage.

  • Before the first workflow goes live, confirm that your governance holds up against these measures:

  • Ownership coverage: Every master data domain has a named owner and a named steward.

  • Duplicate rate: Stewards track duplicate vendor and employee records against an agreed target.

  • Access review: IT has reviewed every role with access to payroll, union rates, or contract terms this quarter.

  • Lineage completeness: Records feeding AI carry origin, timestamp, and approval status.

  • Contract clearance: Legal has checked active projects for data-use restrictions.

  • Pilot results: One logged workflow matches steward review within an agreed tolerance.

FAQs

These answers address the questions construction IT leaders raise as they plan governance work ahead of an AI rollout.

What is data governance in a construction ERP?

Data governance is the set of rules, roles, and system controls that keep ERP data accurate, secure, and traceable. It covers who owns each data domain, who can access it, how records get approved, and how AI tools may use them.

Who should own data governance, IT or the business?

Both share it. Business owners such as controllers and operations leads set the rules for their data domains. Stewards apply those rules daily, IT enforces them through permissions and validation, and legal confirms what contracts allow.

Do construction contracts affect how AI can use project data?

They can. Owner agreements, joint venture terms, and subcontracts may limit how project data is used, shared, or retained. Review data-use and confidentiality clauses before an AI tool reads a project’s files, and carry the same expectations down to subcontractors.

How does a single database support AI governance?

A single database gives each record one version, one owner, and one permission model. AI tools read the same governed data your finance and project teams use. That removes the reconciliation gaps that appear when AI pulls from disconnected systems.

Can you start using AI before governance is complete?

Yes, if you choose the first workflows carefully. Start where the data is well governed and the output stays internal, such as summaries of approved cost reports. Keep payroll, personal employee data, and anything bound for an owner out of scope until your governance checks pass.

Governed Data Gives AI a Foundation You Can Trust

Each control in this checklist answers a question an AI tool asks of your data. Who owns it, who can see it, where did it come from, and what may AI do with it?

CMiC answers those questions on a single database that unifies financials, project management, and workforce records. NEXUS AI features operate within the CMiC security architecture, using authentication controls, role-based access, and audit logging. CMiC also holds ISO/IEC 42001 certification for the AI management system covering AL, the analytics and reporting assistant within NEXUS. CMiC serves one-quarter of the Engineering News-Record (ENR) Top 400 Contractors, and more than $100 billion in construction revenue runs through CMiC annually.

Sources:

  1. New Research Reveals Strong Contractor Optimism About AI’s Transformative Impact on Construction Industry

  2. New Research Reveals How Contractors Can Optimize Project Control Capabilities to Drive Reliable Budgets, Schedules and Profit Margins

  3. RICS Artificial Intelligence in Construction Report 2025

  4. AI & Digital Tools on Construction Projects: Contract Risks to Address Before Peak Season

  5. Who’s Liable When AI Builds?

  6. AI Act: Regulatory Framework

  7. ISO/IEC 42001:2023 Information Technology, Artificial Intelligence, Management System

  8. AI Risk Management Framework