The RFI flow that answers in time

The most common cause of delays on construction projects isn't weather, deliveries or staffing. It's the RFI flow. A question goes from site lead to designer, sits there for ten days while clarification is awaited, and by the time the answer comes back the assumptions have changed. While that's happening, work either runs on assumption or stops. Both cost.

There's a pragmatic operational fix that's been outside most projects' field of view because everyone has accepted that this is just how it works. It doesn't have to. RFI is a flow, and flows can be measured, structured and improved.

Where time is lost today

The first obstacle is that RFIs are often sent as email with attachments. That means they live in one person's inbox and are invisible to the rest of the project organisation. If that person is on holiday or working elsewhere, the RFI sits and waits. Inbox-as-work-queue accounts for 60 percent of the delays we see.

The second obstacle is that RFIs are often incomplete from the start. Questions like "what do we do here" without a photo, without a drawing reference, without an actual problem description, force a round of follow-up before anyone can answer. Each round is two days. Two rounds is four. It's typically six days before an RFI is even answerable.

The third obstacle is that the responder doesn't have the context. An MEP engineer who gets a question about a façade detail first has to figure out which active change orders affect that area, which products were actually selected, and which decisions were taken at the last project meeting. That lookup work takes 30 to 60 minutes before any actual answer is possible.

The agent-driven flow

An agent-driven RFI flow solves these three obstacles as a chain, not in isolation.

Intake: every RFI from the site lead happens through a structured form, not free email. The form requires a drawing reference, a photo or scan of the actual situation, a short problem description, and proposed alternatives. If fields are missing, the RFI can't be sent. That sounds rigid, but it eliminates half the follow-up rounds immediately.

Classification: the agent reads the RFI and proposes who should answer, what urgency applies, and which other RFIs are related. It pulls the three most relevant drawings and the latest change order for that area. All of that lands in the case automatically as context.

First round: the agent drafts a response based on project documentation. It's not a final answer, but a compilation of relevant references and a proposed direction. The MEP engineer gets the response with context in hand, not a blank work item.

Decision: the engineer confirms, corrects or escalates. The decision is logged with reasoning, and the response goes back to the site lead. If the response requires multiple parties, it runs in parallel, not sequence.

The effect on cycle time is not marginal. On four projects we observed in 2026, the average has dropped from 11 days to 2.5 days with the same RFI volume per week. No extra staff, no simplified quality requirements, no reduced traceability. Just a flow built to flow.

What stays manual

Some things still have to be done manually, and it matters that they do. Safety, fire and structural decisions need human judgement regardless. Changes that affect the cost envelope need a project manager's call, not an agent's. Contractual matters with the client need legal review. These should never be automated. They should be protected in the flow with clear escalation rules.

What about documentation? Every construction project has to be traceable to the quality and contract basis after handover. An agent-driven flow makes this easier because every RFI sits structured with timestamp, parties, decision basis and final response. It's better documentation than most projects have today, not worse.

Three practical steps

Three practical steps for a project manager who wants to start. First, measure average RFI cycle time over the last three months and where in the flow time is lost. If you don't have the data, that's the first job. Second, introduce a structured RFI form on one project, ideally the one with the highest RFI volume. Third, run a six-week pilot with agent classification and first-draft responses, and compare cycle time.

The difference between a construction project that delivers on time and one that slips is rarely technical. It sits in the flow of decisions. RFI is the simplest place to show what an operational AI flow actually does for contractors in 2026.