A national general contractor’s three weeks into a hospital renovation. A superintendent notices the electrical drawings show a run of conduit passing through what the structural drawings call a shear wall. The subcontractor stops. The foreman writes an RFI. It goes to the CM, who sends it to the architect, who sends it to the structural engineer, who’s on vacation. Eleven days later, an answer comes back. It needs a small redesign. In those eleven days, two other trades pull back from that area, the schedule slips a week, and a change order gets written. The line item on the general conditions report reads: RFI response, structural clarification. It looks small. It isn’t.
This isn’t a story about one project. It’s the story of construction. And once you look at it as a data problem, the numbers are hard to argue with.
The data
The best public data we have on RFIs comes from the Navigant Construction Forum’s study Impact & Control of RFIs on Construction Projects, which analyzed 1,362 projects and over one million RFIs. The findings, summarized by the American Society of Concrete Contractors and cited across the industry, are consistent, verified, and grim.
The average project generates about 800 RFIs. Each one costs roughly $1,080 to review and respond to (about eight hours of combined administrative and technical time). On a typical project, that’s around 6,368 hours and $860,000 spent on a workflow that produces no building, no design, no material. It only clarifies what should’ve been clear.
Volume scales with cost. Projects between $5M and $50M generate about 17.2 RFIs per $1M of construction cost. Larger projects benefit from better front-end coordination, dropping to roughly 1.1 per $1M above $1B. Across the middle of the market, where most work happens, the industry-average benchmark cited across construction education and PM literature is 9.9 RFIs per $1M.
The response side is worse. Median response time hovers around 10 days. Roughly one in five RFIs never gets a response at all. The average RFI has a potential schedule impact of seven days. And the Navigant analysis found that about 22% of RFIs could’ve been avoided with better front-end coordination, and another 13% were unnecessary altogether: information the drawings already contained if anyone had looked in the right place.
Do the math on a $100M project. Roughly 990 RFIs. About a million dollars in direct review-and-respond cost, before any of the second-order damage: schedule slip, trade re-sequencing, standby time, change orders, disputes. Around a third of that was avoidable. The industry’s priced this in and moved on for four decades.
The hidden pre-bid tax
The Navigant numbers only count RFIs that get logged after award. There’s a second tax, invisible in the data, that runs earlier and larger.
During the bid window, estimators and subcontractors read the same conflicted drawings and specs. When they find a gap. A scope that isn’t clearly assigned, a spec that contradicts the drawing, a detail that could go two ways. They don’t always write a formal RFI. They price it. They call the CM. They chase down the architect’s office for a clarification. They eat the hours as part of the cost of bidding, because they need the number to be right and they don’t want to lose the job over an assumption.
Every subcontractor bidding a package is doing this in parallel. Five mechanical subs on a hospital job are separately figuring out the same conflict between the plumbing plan and the structural set, on their own time, before a single contract’s been signed. The general contractor doesn’t see it directly. The owner doesn’t pay for it as a line item. But it shows up in higher bid prices, thinner subcontractor margins, and the exhausted precon teams who’ve absorbed the cost of the industry’s coordination gaps for free.
If anything, that makes the case for reading drawings and specs together, before the bid goes out, stronger. The formal RFI tax is one number. The pre-bid version is bigger, quieter, and paid by the people least able to absorb it.
Why the tax exists
RFIs aren’t a failure of diligence. They’re a structural feature of how construction produces information.
A modern building is described by two artifacts that grew up separately. The drawings say what goes where. The specifications say what it’s made of and how it performs. They’re authored by different disciplines, on different schedules, in different software, and they meet each other for the first time on the day someone tries to build from them. Add the schedule, which lives in a third system, and the contractor, who’s reading all three under a bid deadline, and the surprise isn’t that 800 questions surface per project. The surprise is that only 800 do.
Before the mid-19th century, none of this existed. The person who designed the building was on-site every day and there was no one to ask. As design and construction separated into distinct disciplines, and as buildings got more complex, the RFI became the connective tissue between them. It’s what happens when the artifacts don’t agree and the schedule doesn’t wait.
Which means the tax has two roots. First, drawings and specs are internally inconsistent, because they’re authored in parallel and reconciled by hand. Second, spatial coordination between trades happens on paper, in 2D, in the estimator’s and superintendent’s heads, until it happens for real in the field. Every gap the paper doesn’t catch, the field will.
The prevention layer
The industry’s response to the RFI tax, for twenty years, has been better tracking. Cloud RFI logs. Assigned deadlines. Response SLAs. That’s real progress on the response side. It’s done nothing to the volume side. The RFI still gets written. The question still exists. The paper still disagreed with itself.
To reduce volume, something has to read the drawings, the specs, and the schedule together, and flag the disagreements before the drawings ever reach the field. That’s not what a keyword search does. It isn’t what a classic BIM clash detector does either, because a clash detector needs a full 3D model, which most projects don’t have until far too late, and it only catches geometric collisions, not the deeper class of inconsistency: a spec that calls for a material the drawing doesn’t show, a scope gap between two trades, a permit trigger no one’s flagged, a note on sheet A-4.02 that contradicts a callout on sheet M-6.
A spatial foundation model can. It reads plans and specs the way an experienced estimator does, connecting a callout on one sheet to a paragraph in section 08 71 00, tracking scope from the bid docs through to the takeoff, and noticing when the story doesn’t hold together. That’s the layer that gets between the design and the field. Boon Agent, our AI teammate for preconstruction, runs this workflow inside the tools estimators already use, on a live bid folder in Procore or Autodesk Construction Cloud: it summarizes the package, flags spec-versus-drawing conflicts, pulls inclusions, exclusions and permit triggers, and generates the RFI when a genuine question does remain. The RFIs that reach the field are the ones that actually need a human answer. And the questions that used to get quietly resolved on a subcontractor’s own time, before the bid, get answered once, for everyone, instead of five times in parallel.
Both sides of the handoff
The reason we can attack RFI volume is that the same spatial model also powers the takeoff. On the construction side, it produces the quantities and the estimate directly from the drawings. On the design side, the same reasoning that measures a wall correctly can also notice that the wall in question is described one way in the architectural set and another way in the structural set.
That’s the position we occupy that other AI-in-construction companies don’t. Point solutions on either side, a takeoff tool that doesn’t read specs, or a document-search tool that doesn’t understand the drawings, can’t close the loop. The prevention only works when the same model sees both artifacts and reasons across them. Design-side review and construction-side estimating from one underlying model, with the same integrations into the tools teams already use.
The investor frame
The point, in one paragraph. RFI waste is one of the largest dollar-denominated inefficiency lines in a multi-trillion-dollar US construction industry: at 9.9 RFIs per $1M of construction and $1,080 per RFI, the direct cost alone runs into the tens of billions of dollars a year, before any of the schedule and change-order damage that follows it. And that’s only the formal count. Every project also carries a shadow tax paid by estimators and subcontractors resolving the same conflicts during the bid, on their own time, so the number wins the job. A third of the whole thing’s avoidable. It isn’t a productivity story or a workflow story. It’s a foundational-model story: the disagreements that produce RFIs are visible in the drawings and specs before the drawings ever reach the field, and reading them at that layer is the only thing that reduces volume. That’s the layer Boon operates on, and we’re the only company doing it from both the design side and the construction side.
Closing
We don’t think RFIs disappear. Some of them are real questions that a real human needs to answer, and the process of asking them is how the industry catches its own mistakes. What we think is that the volume, the cost, and the schedule tail attached to them can come down by a large multiple, if the drawings and specs get read carefully, together, before the field ever sees them.
The industry’s quietly paid the RFI tax for forty years. Nothing about it’s inevitable. The disagreements are on the page. We built a model that reads the page.
If you’re running preconstruction and want to see what your last bid folder looks like through this lens, getboon.ai/boon-agent will show you.
Acknowledgements
Thanks to Jon Miao for the industry review of this piece.
