How can construction companies use AI safely?

Construction companies can use AI more safely by approving specific tools, classifying information, limiting what employees may enter, reviewing vendor data practices, requiring human verification, and keeping confidential information out of unapproved public AI services.

The safest first projects use low-risk information and produce drafts that a qualified person reviews.

Start with the information, not the AI tool

Before an employee enters content into an AI service, ask what the information contains.

Examples of sensitive construction information may include:

• Contracts and legal correspondence
• Unpublished bids and estimates
• Customer and subcontractor information
• Employee and payroll records
• Banking and payment details
• Credentials and security settings
• Building plans and access information
• Safety incidents
• Proprietary methods
• Confidential project photos

Some information may be public, some may be internal, and some may be restricted by contract, law, policy, or client expectation. Employees need clear categories and examples.

Approve tools and account types

Do not assume that every AI service handles data the same way. Review:

• Whether prompts and files are retained
• Whether submitted information may be used to improve models
• Available administrative controls
• Identity and multi-factor authentication
• User and activity management
• Data location and deletion options
• Contractual terms
• Integrations with company systems

Consumer and business versions of the same service may have different controls. Review the current vendor documentation and agreement before approval.

Keep project systems in mind

AI features are appearing in platforms used for construction management, design, estimating, service, finance, and collaboration. A construction company may use Microsoft 365 alongside Procore, Bluebeam or Adobe, Autodesk products including AutoCAD and Revit, SketchUp, Trimble tools, Vista Viewpoint, Sage HCM, hh2, HCSS, ServiceTitan, and BuildingConnected.

Before enabling an AI feature or integration, determine:

• Which project or business data it can access
• Which users can invoke it
• Where generated output is stored
• Whether the feature creates new copies of data
• How permissions are inherited
• Whether output can be exported or shared externally
• Who reviews accuracy

An AI feature inside an approved business platform may offer stronger controls than an unapproved public service, but it still requires review.

Begin with lower-risk uses

Possible starting points include:

• Rewriting a public job posting
• Summarizing a public regulation for internal discussion
• Drafting a generic safety-meeting agenda without incident details
• Improving the clarity of a nonconfidential email
• Generating questions for a vendor meeting
• Creating a checklist from an approved public procedure

Avoid live contracts, customer records, employee information, security configurations, and confidential bid data until the tool and use case have been formally approved.

Require human review

AI output can be incomplete, inaccurate, or overly confident. A qualified employee should review any output used for:

• Contracts
• Bids and estimates
• Engineering or design decisions
• Safety
• Financial reporting
• Payroll
• Compliance
• Customer commitments

The reviewer remains responsible for the final work. AI should not be presented as the source of professional approval.

Control integrations

An AI tool connected to email, SharePoint, Procore, Autodesk, Vista Viewpoint, ServiceTitan, or another system may access more information than an employee realizes.

Require technical and business review before connecting AI tools to company applications. Use the minimum permissions necessary and remove unused integrations.

Provide a reporting route

Employees should know what to do if they accidentally enter sensitive information into an unapproved tool. Early reporting gives the company a better chance to assess retention, sharing, contractual obligations, and next steps.

Punishing good-faith reporting can cause employees to hide mistakes. The policy should encourage quick escalation.

Frequently asked questions

Can employees paste a contract into a public AI tool?

Not unless the company has approved the tool and use case and confirmed that the contract and applicable obligations permit it. Confidential contracts should not be entered into unapproved public services.

Is AI output confidential?

Confidentiality depends on the tool, account, settings, contract, and information involved. Do not assume an AI conversation has the same protection as an approved internal business system.

Can AI review a construction estimate?

AI may assist with defined tasks, but estimates require qualified human review. The company should evaluate data sensitivity, accuracy risk, tool permissions, and contractual requirements first.

Learn More

SisAdmin helps businesses review the technical side of AI adoption, including accounts, access, Microsoft 365, devices, integrations, security, and acceptable-use controls. Legal, contractual, safety, and professional decisions should also involve the appropriate advisers and subject-matter experts.