On every construction project of meaningful scale, the Request for Information — the RFI — is the primary channel through which subcontractors communicate technical queries to the design team. When a discrepancy is spotted on site, when a drawing conflicts with a specification, when a contractor needs clarification before proceeding: an RFI is raised.
The process sounds routine. In practice, it is one of the most consequential administrative burdens a project manager carries.
On a mid-size commercial project, between 200 and 500 RFIs are common over the course of the build. On large infrastructure projects — hospitals, data centres, rail schemes — the number can reach into the thousands. Each one requires the PM to cross-reference the specific question against potentially hundreds of pages of documents: architectural specifications, structural drawings, M&E schematics, and the contract itself. Manual processing takes between 30 and 90 minutes per RFI depending on complexity.
The consequences of delay compound quickly. An RFI is frequently on the critical path. The subcontractor who raised it cannot progress that element of work until a response is received. Downstream trades are held up. Programme floats are consumed. Cost overruns follow.
In the UK context, where JCT and NEC contracts impose strict notification and instruction requirements around scope variations, an unanswered or poorly answered RFI can create contractual exposure that outlasts the project by months or years.
What the PM Is Actually Doing
The PM processing an RFI manually performs a specific sequence of tasks:
They identify the relevant specification division from the RFI subject — a query about raised access flooring relates to Division 09 (Finishes); a structural query about concrete works relates to Division 03 (Concrete). They locate the relevant section in the specification document. They cross-reference the specific question against the drawings — checking plan views, sections, and details. They check whether any drawing addenda have been issued that supersede the original drawings. They assess whether the question raises a contractual issue — a variation that requires a formal instruction, or a discrepancy that creates a potential claim.
They then draft a response: technically accurate, contractually careful, addressed to the right party, and issued within the contractually required timeframe.
Under JCT Standard Building Contract 2016 and NEC4, an RFI response that misinterprets a specification clause, contradicts a drawing, or creates a scope variation without proper instruction may constitute a breach. The PM who drafts that response carries professional exposure alongside the corporate one.
And the RFI typically arrives with a short deadline — 48 or 72 hours. The PM processing it under deadline has less time to cross-reference, less time to consider the contractual implications, and more temptation to draft a quick answer that is approximately right.
Approximately right is not good enough in construction.
The Administrative Burden Is the Real Problem
Research from McKinsey and the Chartered Institute of Building consistently estimates that construction project managers spend up to 75% of their working hours on administrative tasks: document management, correspondence, progress reporting, meeting preparation, and — centrally — information management through RFIs, submittals reviews, and instruction logging.
This is time spent not on site, not managing people directly, and not exercising the professional judgement that justifies their day rate.
RFI processing is one of the most time-intensive and consequential administrative tasks. It requires sustained concentration on technical documents, precise cross-referencing, careful drafting, and a sound understanding of contractual implications. It is exactly the kind of work that benefits from intelligent assistance — not because it is simple, but because it is repetitive and document-heavy in ways that AI systems are well-suited to support.
The framing is important: administrative offloading, not professional replacement. AI handles the documentary legwork — the cross-referencing, the clause-location, the first-pass drafting. The PM retains full ownership of decisions. AI generates the first-pass draft; the PM applies professional verification. The PM's name goes on the response; the PM's accountability is not reduced by one iota.
The Procore + Claude 3 Workflow
The AI-Assisted RFI Resolution Protocol uses Procore as the system of record and Claude 3 as the document comprehension and first-pass drafting engine.
Procore stores specifications, drawings, addenda, and RFI records. Its role in this workflow is document export — pulling the relevant PDFs from the Common Data Environment so they can be passed to Claude.
Claude 3, with its 200,000-token context window, reads the specification, cross-references the drawings, identifies relevant clauses, and generates a structured draft response. Its role is document-reading assistant — doing the legwork that the PM would otherwise do manually.
The four-step workflow:
- Step 1 — Data ingestion: The relevant specification division and drawing addenda are identified and exported from Procore as PDFs. Approximately 3–5 minutes.
- Step 2 — Context assembly: The two PDFs and the RFI text are uploaded to a new Claude conversation. The RFI text is pasted first — priming Claude to answer the specific question before synthesising from the broader document context. Approximately 2–3 minutes.
- Step 3 — Prompt execution: The structured three-part prompt is executed. Claude produces: (1) specification clause identification with citations, (2) contractual conflict assessment, and (3) a first-pass draft response. Approximately 5–15 minutes.
- Step 4 — PM verification and issuance: The PM reviews every citation, corrects any inaccuracies, assesses the contractual position, sets the variation flag, and issues under their own sign-off in Procore. Approximately 15–30 minutes.
Total time per RFI: approximately 25–50 minutes versus 30–90 minutes manually. The 60% time saving comes primarily from eliminating the manual document search and cross-reference step — the part of the task that is tedious but not professionally demanding.
Why the PM Verification Step Cannot Be Skipped
Claude 3 can misread a specification clause. It can miss a drawing addendum that supersedes the section it cited. It can generate a response that is approximately right but subtly wrong on a contractual point. These are not hypothetical failure modes — they are the documented limitations of large language models that every professional user must account for.
The PM who issues a response based on unverified Claude output is exposed in exactly the same way as the PM who issues a response based on rushing. The AI does not reduce the PM's professional accountability. It reduces the time spent on the documentary legwork so that more time is available for the professional review.
The 8-point PM RFI Checklist in the guide exists precisely to ensure that the verification step is completed systematically — not skipped under deadline pressure.
Related Resources
The complete AI-Assisted RFI Resolution Protocol — including the Procore export workflow, the three-part Claude 3 prompt structure, the 8-point PM RFI Checklist, and the variation flag framework — is documented in The RFI Resolution Engine. That guide includes a worked example using a raised access flooring RFI, the JCT/NEC contractual context, and the professional accountability framework.
If you are managing design coordination alongside RFIs and dealing with the clash detection backlog that generates coordination meeting outputs, The Clash Resolution Memos workflow covers the parallel AI-assisted workflow for drafting Design Coordination Memos from Navisworks clash reports — using Claude 3 and the same administrative offloading principle.
Stop Guessing, Start Systemising
The RFI bottleneck is not a people problem — it is a structural problem. The documentary legwork of RFI processing is exactly the kind of tedious, repetitive work that AI handles well. The PM's professional knowledge is precisely what is needed for the verification step.
The RFI Resolution Engine has the complete four-step protocol, the structured three-part Claude 3 prompt, the JCT/NEC contractual context, and the 8-point PM RFI Checklist. Sixty percent reduction in RFI response time — with full professional accountability at every step.
Grab it here.
Practical AI for Professionals