Construction RFI and change order management with AI triage

What is a construction RFI?

A request for information (RFI) is a formal question asking someone to clarify information needed to carry out the work. On a construction project, that question commonly concerns a drawing detail, a specification, an existing site condition or a proposed substitution. The objective is to give the project team a clear, traceable answer before work proceeds on an uncertain assumption.

A subcontractor may first raise a question with the general contractor. The contractor checks whether it can be answered from the contract documents and, when needed, refers it to the architect, engineer or other design professional. The response returns through the agreed project channel. Actual communication and authority depend on the contract.

Good RFIs are specific. The reviewer should know the project and location, the drawing or specification in question, what has been observed, what decision is requested, and the date by which the answer is needed. A proposed solution can help, provided it is clearly distinguished from an approved instruction.

The free RFI template includes these fields so questions enter the process in a consistent format.

What is a construction change order?

A construction change order is a written, authorized change to the original agreement's scope, contract price, contract time or a combination of these. An RFI may reveal a design discrepancy that requires a change; the RFI response identifies the issue, while the change order records the parties' agreement on the resulting adjustment.

Change orders also arise from owner-directed revisions, unforeseen conditions, substitutions and value-engineering proposals. Before approval, the project team needs to understand what changed, why it changed, the relevant quantities and rates, and any effect on milestones or completion.

The terminology matters. A change request generally proposes an alteration for evaluation. A change directive may instruct work to proceed before the full commercial adjustment is agreed, where the contract permits. A change order records an authorized amendment. The exact terms and approval powers differ between contract forms and jurisdictions, so the signed contract takes precedence over generic definitions.

A complete change order provides the related RFI or instruction reference, a description of revised work, an itemized cost breakdown, previous and revised contract sums, the time adjustment and named approvals. Supporting quotations and drawing revisions should remain attached to the record.

Our change order templates provide Word and Excel formats for capturing these details and maintaining an organized log.

The RFI and change order lifecycle

A manageable process follows the question from first report to final decision. Each hand-off needs a named owner, a due date and a reliable way to see what happens next.

1. Raise the questionA subcontractor, supervisor or project engineer identifies a missing detail, discrepancy or condition and submits the RFI.
2. Log and acknowledgeThe contract administrator or project team assigns a number, records the source and confirms receipt.
3. Review and assignThe contractor checks for duplicates and completeness, then sends the RFI to the responsible design or discipline reviewer.
4. Respond and clarifyThe authorized reviewer answers, references the relevant documents and requests clarification if the question is incomplete.
5. Assess commercial impactThe project manager and quantity surveyor consider whether the answer changes materials, quantities, sequencing, price or time.
6. Prepare a change requestWhere required, the commercial team documents revised scope, supporting costs and possible schedule effects.
7. Approve the changeThe people authorized by the contract review and sign the change order or follow the applicable directive process.
8. Update project controlsThe team updates the budget, forecasts, programme, procurement instructions and affected drawing register.
9. Close and fileThe contract administrator links the final RFI response and approved change record, then closes the issue with a traceable history.

Not every RFI results in a change order. Some simply confirm the existing design. Even then, retaining the clarification with the related project record helps later teams see which instruction was followed and why.

Where RFI management and change order management break down

RFI management becomes difficult when incoming questions are scattered across email threads, site messages, meeting notes and document systems. A reviewer may be assigned in one channel while the latest attachment lives in another. A new query can also duplicate an issue that is already being answered.

The problem grows with simultaneous trades and multiple subcontractors. Reducing it requires consistent intake fields, clear routing, visible deadlines, and a record of each substantive decision. Software helps when it supports those controls rather than adding another disconnected inbox.

How AI triage works for construction RFIs and change orders

AI reads and sorts incoming information; project professionals make the decisions. The useful division of work is to let AI interpret messy field submissions and let ordinary workflow rules enforce due dates, approval limits and assigned responsibilities. For a broader view of this approach, see our AI in construction guide.

1. Classify by urgency and trade

The incoming RFI, its stated question and relevant attachments are passed to an AI classification step. The output identifies a likely discipline — structural, MEP, civil or finishes — and indicators of urgency. The model can flag language about blocked work, hazardous conditions or an approaching construction activity. The classification is a prompt for review, not an official safety determination.

Where a requirement can be tested directly, a rule performs the check. For example, if the response-by date is within 48 hours, the workflow applies the project's urgent flag without making another AI call. A date calculation should not depend on a model's interpretation.

2. Summarize the question and likely impact

AI produces a concise review brief: the question, project area, reported condition, relevant drawing or specification references and a possible schedule or cost consequence. The original documents accompany the summary so the reviewer can verify the interpretation. Missing references remain marked as missing; the model should not invent drawing numbers.

This is especially useful when a field engineer sends a long email chain, annotated photo and spoken note about the same issue. The reviewer sees a consistent summary while retaining access to the source evidence.

3. Route to the right reviewer and approval chain

The workflow uses the trade, project and defined responsibility matrix to assign the RFI to the appropriate consultant or internal reviewer. Potential commercial impacts are flagged to the project manager or quantity surveyor. If the authorized response calls for a change, a related change request can be opened with the RFI references already attached.

Change order reviews then follow the project's value thresholds and required approval roles. Approvers can review sequentially or in parallel, depending on the contract and internal policy. The workflow sends reminders and escalates overdue items without silently approving a technical or commercial decision.

4. Preserve the audit trail

Each stage records the date, assigned person, material reviewed, AI suggestion, actual decision and subsequent action. Versioned or controlled filing protects an approved answer from casual edits. The record should show where an AI-generated summary ended and where an authorized instruction began.

Automated RFI generation from field questions

Automated RFI generation can prepare a structured draft from a supervisor's voice note, an email or a photo accompanied by a short explanation. AI identifies the question, the location and any document references present in the supplied materials, then populates the standard RFI fields. It can flag missing information for the project engineer to complete.

The project engineer reviews and issues the RFI. AI does not create contractual authority merely by drafting a form. The same principle applies to change orders: the contract and named decision-makers determine what constitutes an authorized change.

Free RFI and change order templates

Download a consistent starting point for project questions, variations and commercial tracking. The package includes three editable files; access is provided after you submit the short form below.

Standardized inputs make subsequent AI classification and workflow routing more reliable. Adapt the templates to your contract, regional requirements and internal approval rules.

How to introduce AI into an RFI process

Start with one repeatable part of the process, rather than automating every possible variation at once. A pilot that handles a few common trades and gives reviewers clear ownership will reveal the right exceptions before the workflow expands.

1. Standardize the RFI form. Establish required fields, numbering and a clear place for supporting documents.
2. Define trade and urgency categories. Agree on the discipline list and what truly makes a question urgent.
3. Map approval responsibility. Record which role reviews each trade and who can authorize financial or schedule changes.
4. Make AI classification repeatable. If you use an AI workflow automation platform, such as Zenphi, add classification as a configurable AI step. Put the agreed trade categories, urgency criteria, required output fields and exception handling into its prompt or system instructions. Every submission then uses the same instructions, helping make the output more predictable and consistent. Configure a confidence threshold so uncertain classifications escalate to a named reviewer, and include a human-in-the-loop approval step before anything consequential is issued. If you use a standalone AI tool, such as Claude or Gemini, save the same prompt as a reusable asset rather than rewriting it for each RFI. In Gemini: Open Gemini, choose Gems → New Gem, name it (for example, “Construction RFI triage”), paste your classification instructions, optionally add your category or policy reference under Knowledge, test with anonymized historical RFIs, and select Save. Review each output before routing it manually. Google’s Gem setup guide explains the steps.
5. Preserve human sign-off. Require named project and commercial reviewers to issue responses and authorize variations.
6. Review performance monthly. Track open RFIs, overdue responses, incomplete submissions, escalations and the proportion linked to approved changes.

Project teams should review false urgency flags and misrouted questions as part of regular process maintenance. That feedback improves the classification prompts while leaving contractual authority with the people assigned to exercise it.

Running RFI and change order triage with Zenphi

Zenphi is an AI workflow automation platform for designing and deploying controlled operational processes, including workflows that span project teams, consultants and commercial reviewers. Teams build and manage workflows visually, choosing where AI is useful and where deterministic rules are more appropriate.

Consider an RFI submitted through a project form. One AI step classifies the trade and urgency and prepares a concise summary. A conventional rule checks whether the response-by date is within two working days. The workflow assigns the request to the designated consultant. If the reviewed response requires a change, a related commercial request is prepared and sent through the agreed value-based approval chain. The records, decisions and approved documents are retained with controlled editing permissions.

Zenphi supports sequential and parallel approvals, reminders, escalation and an auditable history of workflow activity. Routine checks — required fields, deadline calculations and monetary thresholds — run without invoking an AI model. Teams can select Gemini, OpenAI, Claude or their own models for AI steps, with model-use controls available in the Zenphi Admin console. This helps limit AI consumption to tasks such as interpretation, extraction and classification.

To design a process, teams can build from scratch or use ZAIA to generate an editable workflow structure from a plain-language brief. The team then configures the approval paths and data mappings, tests the workflow and deploys it. Deployment can be as quick as seven days depending on complexity and required approvals.

Zenphi pricing is based on workflows, without per-user, per-tenant or flow-run fees on paid plans. That structure can be useful when submissions come from numerous subcontractors, supervisors and reviewers.

Construction RFI and change order FAQs