Requests for Information (RFIs) are one of the most common sources of project information in construction. They document design clarifications, coordination issues, missing information, constructability questions, and proposed solutions.

But RFIs can also contain something more valuable: early signals of project risk.

The problem is that these signals are often buried inside thousands of project records and are rarely transferred systematically into the project risk register.

In this workflow, we use Claude Opus 5 to transform RFIs into structured risk-register entries in just a few minutes.

The result is a practical workflow that can help project teams move from:

RFI → Risk Identification → Risk Classification → Risk Scoring → Risk Register

And, importantly, create a structured knowledge base that can later be connected to contractual requirements, project correspondence, and claims analysis.

Watch the accompanying video to see the complete RFI-to-Risk-Register workflow in Claude:

An RFI is usually treated as a communication and documentation mechanism.

For example, an RFI might identify:

Each of these can represent an emerging risk. The challenge is scale.

On a large construction project, hundreds or thousands of RFIs can be generated. Manually reviewing every RFI, determining whether it represents a risk, categorizing that risk, assigning likelihood and impact, and updating the risk register is time-consuming.

This creates a gap between project communication and project risk management. AI can help close that gap.

The workflow demonstrated in the accompanying video uses a predefined Claude Skill called RFI to Risk Register.

A Claude Skill can be thought of as a reusable set of instructions that defines how Claude should process a particular type of task.

You can read more about Claude Skills in this article: Claude Skills for Construction: What They Are and How to Use Them?

Instead of repeatedly explaining:

“Read this RFI, identify risks, categorize them, score them, and put them into this structure.”

the workflow can encode these instructions once and reuse them.

The resulting workflow consists of six main stages:

This turns an otherwise open-ended AI interaction into a repeatable process.

The first step is to load the predefined Skill into Claude.

Inside Claude, go to:

Customize → Skills → Yours → Add a new skill → Upload skill

The Skill can be uploaded as a skill, .md, or zip file.

Once imported, Claude provides a tag that can be used to activate the workflow inside a conversation. The important concept here is that the Skill contains the methodology.

It defines what Claude should extract, how risks should be classified, what scoring methodology should be used, and what the final output should contain.

This makes the workflow much more consistent than relying on an individual prompt each time.

Once the Skill is enabled, a simple instruction can initiate the workflow:

“Initiate risk register workflow.”

Claude then follows the predefined process and asks how the RFIs should be supplied. For a simple demonstration, an RFI can be uploaded directly. However, the workflow can also be extended to work with project folders.

Using Claude’s Co-work capabilities, the AI can be directed toward a project folder and work with files stored there.

This creates an interesting opportunity for larger implementations:

Project Folder → RFI Collection → AI Processing → Risk Register

Instead of manually uploading every RFI, the workflow can potentially operate against a broader project information environment.

The AI then analyzes the RFI and extracts the information required to determine whether there is a potential project risk.

A typical RFI contains information such as: Project information, RFI number, Subject, Classification, Priority, Status, Description of the clarification requested, Proposed resolution, Potential impact, and more.

The important part is that the AI is not simply summarizing the RFI. It is looking for risk signals.

For example, a clarification concerning a wall section and a slab penetration may indicate a coordination issue. A missing specification may indicate a potential design or contractual risk.

A proposed solution that changes an element may indicate potential cost, schedule, or scope implications.

The AI therefore moves from:

“What does this RFI say?”

to:

“What could this RFI mean for the project?”

Once potential risks have been identified, the Skill instructs Claude to classify and score them. The workflow can use categories and subcategories to make the resulting register easier to analyze.

For example:

The risk can then be scored using a simple likelihood × impact methodology. This creates a consistent basis for prioritization. Instead of having a collection of RFIs with different levels of importance, the project team can start seeing a structured risk landscape.

The output from the workflow is a structured risk register rather than a simple text response. In the video demonstration, Claude generates a spreadsheet containing fields such as: Risk ID, Date identified, Source RFI, RFI subject, Status, Risk title, Risk description, Category, Subcategory, Evidence, Likelihood, Impact, Risk score, Risk rating, Cost exposure, and more.

This is where the workflow becomes particularly useful.

The AI is not simply producing a list of potential risks. It is creating a structured dataset that can become part of the project’s broader risk-management system.

The workflow can go beyond processing a single RFI.

The generated workbook can contain multiple components, including an RFI log, risk register, dashboard, and risk-scoring methodology.

This creates the foundation for something much more powerful. Instead of processing RFIs individually, future RFIs can be appended to the existing register.

Over time, the project could develop a structured relationship between:

RFIs → Risks → Project Activities → Contract Clauses → Notices → Claims

This effectively turns project correspondence into a searchable risk knowledge base.

For example, if multiple RFIs relate to a particular design package, the project team could identify a recurring risk pattern rather than treating each RFI as an isolated event.

One of the most interesting applications is the connection between risk management and claims management. An RFI can represent the first documented indication that something is changing or going wrong.

If that RFI is automatically linked to:

The project team has a much stronger information trail. This does not mean AI determines whether a contractor is legally entitled to a claim. That remains a contractual and professional assessment.

Instead, AI can help ensure that potentially important information is captured, classified, and connected early. That distinction is important.

AI should support the project team’s judgment, not replace contractual or legal analysis.

Traditional workflows often look like this:

RFI → Response → Close RFI

The proposed workflow adds another layer:

RFI → AI Risk Detection → Risk Register → Monitoring → Action

That small change can have a significant impact. An RFI that might previously disappear into a project-management system can become an active risk-management record.

This also creates the possibility of identifying risks earlier. Instead of waiting until a delay, variation, or dispute has materialized, project teams can monitor the signals appearing in project correspondence.

There are several things to keep in mind before deploying a workflow like this on a live project.

AI should not invent the project’s risk-management methodology. Define the categories, scoring system, terminology, responsibilities, and output structure first.

Then encode those requirements into the Skill.

Every identified risk should remain traceable to the original RFI and supporting documentation. The AI-generated risk should never become detached from its source.

AI can identify potential risks, but the project team should validate them.

A human should confirm whether a detected issue actually constitutes a project risk and whether the assigned likelihood, impact, owner, and actions are appropriate.

The greatest value comes when the risk register is connected to schedule, cost, procurement, quality, and contract information.

The RFI itself is only the starting point.

The workflow can accelerate identification and organization of information, but it should not independently make contractual, financial, or legal determinations.

RFIs contain far more information than their traditional role suggests.

They can provide early indicators of design problems, coordination issues, scope changes, schedule threats, cost exposure, and contractual risks.

With a structured AI workflow, these signals can be extracted automatically and transformed into a functioning risk register.

The result is a shift from:

“We have thousands of RFIs.”

to:

“We can continuously extract project risk signals from our RFIs.”

And that is potentially a much more valuable use of AI in construction.

/