How to Reduce RFIs in Construction: Where They Come From and What Actually Works

An RFI is a question the drawings should have answered. Every one that gets raised is a small admission that somewhere in the set, the information a builder needed was missing, ambiguous, or contradicted by another sheet.

Almost everything written about reducing RFIs is actually about processing them faster: better logs, tighter routing, response deadlines, a mobile app. That work is worth doing, but it is a different problem. Turning a 14-day response into a 4-day response is throughput. Reducing RFIs means the question never gets asked, because the set already answered it.

This guide is about the second one.

Two problems that share a name

Say plainly which one you are solving, because the interventions barely overlap.

Fewer RFIs is a document quality problem. It is won before issue, by the design team, and the levers are coordination, completeness and cross-checking.

Faster RFIs is a workflow problem. It is won during construction, mostly by the contract administrator, and the levers are routing, ownership, escalation and deadlines.

Teams overwhelmingly buy tools for the second and then report disappointment that the first did not improve. A faster pipe moves the same water. If your RFI count is the number that hurts, the intervention has to happen upstream of issue, which means it competes for time with getting the set out — and that is the real reason it rarely happens, not ignorance about its value.

What an RFI actually costs

Worth being precise here, because the case for spending design-phase time rests on it.

At the Navigant averages, a single project's RFI processing runs to roughly $860,000 before a single answer changes anything on site.

The 1999 finding is the one worth sitting with. The expensive part of an RFI is not writing it and not deciding the answer. It is gathering and cross-referencing information — someone opening four sheets, a schedule and a spec section to reconstruct what the documents collectively say. That is the same work a reviewer would have done before issue, except now it is being done twice, by more expensive people, with a clock running and a crew waiting.

Both studies are old. Treat them as the documented scale of the problem rather than today's price list.

Where RFIs come from

Here the honest answer is less satisfying than the one you will find elsewhere.

A figure circulates widely in vendor content: that the Construction Industry Institute found 30–50% of RFIs are caused by errors, omissions or ambiguities in the construction documents. It appears with no publication, no year and no author, repeated between blogs that cite each other. We could not trace it to a CII publication, so we are not going to repeat it as though we had.

What is reasonably well established:

• RFIs cluster in the interfaces, not in the middle of any one discipline's scope. Architectural against structural, structural against MEP, drawings against specifications, drawings against the schedules that describe them.
• The categories that recur across studies and logs are design clarification, coordination conflicts between disciplines, dimensional and level discrepancies between sheets, specification-versus-drawing conflicts, unforeseen site conditions and material substitutions.
• Only some of these are preventable by better documents. Site conditions and substitutions are not. A guide implying you can drive RFIs to zero by reviewing harder is selling something.

There is a better source of causal data than any published study, and you already own it: your last three projects' RFI logs. Categorise a few hundred of them by root cause and you will have a distribution specific to your building types, your consultants and your standards — which is the only distribution that should shape where you spend review time. Almost nobody does this, which is why the industry keeps citing a 1999 paper and an untraceable percentage.

The strategies, ranked by leverage

Ordered by how far upstream they sit. The earlier the intervention, the cheaper the fix and the more questions it removes.

1. Fix the cross-sheet layer before issue

The single highest-yield check, and the one most often skipped.

A callout pointing at a detail nobody drew is invisible on the sheet it sits on. It reads as completely correct in isolation. It is only wrong in company — and finding it means holding two sheets in your head at once, several hundred times.

The arithmetic is why this fails as a manual task. Cross-sheet checking scales with the square of the sheet count: a 40-sheet set has 780 possible sheet pairs, a 400-sheet set has just under 80,000. No reviewer performs that comparison, so in practice it gets sampled — the sheets someone remembers changing, checked against the sheets someone remembers them relating to. Sampling is a rational response to an impossible task, and it is precisely why the errors that survive to site are disproportionately cross-sheet errors, and why they arrive as RFIs rather than redlines.

What to check: callouts resolving to sheets and details that exist, section and elevation markers pointing at views actually produced, door and window marks agreeing with their schedules, room names and numbers agreeing between plans and finish schedules, levels and datums agreeing across plans and sections, no duplicate sheet numbers, sheet index matching what was issued. The drawing QA/QC checklist covers the full set.

2. Reconcile grids, levels and datums across disciplines

Architectural grid C should not be structural grid D. Slab levels should agree between architectural and structural. Ceiling heights should agree between architectural sections and MEP coordination drawings.

These are cheap to check and expensive to miss, because a grid mismatch does not generate one RFI. It generates a stream of them, from every trade that hits the discrepancy independently, over months — and each one is answered in isolation without anyone noticing they share a root cause.

3. Resolve drawing-versus-specification conflicts before they become contractual

When drawings and specs disagree, someone has to decide which governs, and the answer depends on your contract. FAR 52.236-21 gives specifications precedence on federal work. ConsensusDocs also prioritises specifications. AIA A201 deliberately declines to establish a hierarchy, treating the documents as complementary and leaving interpretation to the architect.

That matters for RFI volume in a specific way: under a contract with no order of precedence, every drawing-versus-spec conflict is guaranteed to become an RFI. There is no rule the contractor can apply to resolve it themselves. So the conflicts have to be found before issue, or they will all come back as questions with a clock attached.

Practically: check that the products, finishes, fire ratings and performance criteria named on sheets match what the corresponding spec sections require, and that spec sections referenced on drawings actually exist in the project manual.

4. Buy a constructability review from someone who builds

A contractor reading the set from a builder's perspective finds a category of gap that no designer's review surfaces: sequencing that does not work, access that does not exist, tolerances that cannot be held, details that are drawable but not buildable.

This is the highest-value human review available, and it is wasted on mechanical checks. If your constructability reviewer spends their day flagging empty tags and unresolved callouts, you have bought expensive proofreading. Clear that layer first so their attention goes where only a builder's judgement works.

5. Run model coordination for what it is actually good at

Clash detection resolves hard geometric conflicts before they reach site, and it does that genuinely well.

Be clear about its limits. Clash detection compares geometry in a model. It says nothing about whether the documents published from that model agree with each other. A model can be entirely clash-free and still produce a set with callouts pointing nowhere, schedules disagreeing with plans, and a spec calling for a product the drawings never mention. Those produce RFIs at the same rate as a duct through a beam, and no clash report will show them.

6. Batch the pre-construction RFI round

Getting the trades to read the set and raise their questions before mobilisation converts a stream of clock-bearing site queries into one scheduled block that can be answered together. It is a widely recommended practice; the specific reduction percentages attached to it in vendor content do not have visible sourcing, so judge it on the mechanism rather than the number. The mechanism is sound: the same questions, asked when the answer costs a drawing revision instead of a work stoppage.

7. Close the loop from your own RFI log

Every RFI is a defect report on your documents and your standards, arriving free of charge.

Tag each one with a root cause. Once a quarter, look at what recurs. If the same interface generates questions on every project, that is a template problem, a standard-detail problem or a consultant-scope problem — and it is fixable once rather than answered forty times. This is the only strategy on this list that gets cheaper the longer you run it, and the only one that compounds across projects.

8. Write in a way that produces one interpretation

Replace ambiguous language with criteria. Where performance language is unavoidable, pair it with enough detail that two contractors would price and build it the same way. If two competent people can read a note differently, the difference will arrive as an RFI, and it will arrive at the worst moment.

9. Make the RFI process itself unattractive for things that are not RFIs

A meaningful share of any log is not a genuine documentation gap: it is normal communication that got formalised because the formal channel was the only one with an audit trail, or because someone wanted the question on record. Give routine coordination questions a real path with real responsiveness, and reserve the RFI for what needs a contractual answer. This does not improve your documents at all. It does make the RFI count mean something, which you need before you can measure anything else on this list.

The expensive part of an RFI is reconstructing what the documents collectively say. That work happens either once, before issue, or repeatedly, at site rates, with a crew waiting.

What is oversold

Faster response times as a reduction strategy. Answering in 3 days instead of 10 is a genuine improvement in schedule impact. It does not reduce RFI count by a single question, and treating the two as one metric hides whether your documents are getting better.

Clash-free as a proxy for coordinated. Covered above. Different failure mode, different check, and the clash report's clean bill of health is actively misleading about the documents.

Reviewing harder. Sets are not under-reviewed because reviewers are careless. They are under-reviewed because the cross-sheet layer is combinatorially larger than the time available, and it gets sampled. More diligence applied to an impossible task yields a slightly larger sample.

Any tool claiming to catch design intent. Whether a wall belongs where it is drawn is not a question a set can answer about itself. Software can tell you the callout does not resolve. It cannot tell you the room is the wrong size, and anything claiming otherwise is guessing with confidence.

How to tell whether it worked

Raw RFI count is not comparable between projects. Normalise, then track over time.

The third one is worth the effort to collect, even though it takes judgement rather than a report filter. It splits your log into two piles with completely different fixes: questions where the answer was already in the set but nobody could find it — a clarity, structure and cross-referencing problem — and questions where the answer genuinely was not there — a completeness problem. The interventions for those are not the same, and averaging them together tells you nothing.

Also expect a counterintuitive first result. Teams that introduce a pre-construction RFI round often see their total count rise in the first period, because questions that used to appear as site queries now appear as batched ones. Judge on schedule impact and rework, not on the raw count, for at least the first project.

A realistic sequence

1. Categorise your last project's RFI log by root cause. Two hours of work, and it replaces every industry average in this article with your own distribution.
2. Clear the mechanical and cross-sheet layer on the current set before any human reviews it. Callouts, references, schedules, sheet index, grids, levels. This is the part that scales badly by hand and reaches site most often.
3. Spend human review on judgement: constructability, design intent, the interfaces your own log says are your weak points.
4. Batch the trades' questions before mobilisation.
5. Feed what still came back into your templates and standards, not just into the answer.

Steps 1 and 5 are the ones teams skip, and they are the only two that make the next project cheaper than this one.

Frequently asked questions

How many RFIs is normal on a construction project?

The Navigant Construction Forum's 2013 study of roughly one million RFIs across 1,300 projects found an average of 796 RFIs per project. That average spans wildly different project sizes, so it is close to useless as a target. Normalise against contract value or sheet count and compare against your own history rather than an industry figure.

What is the average cost of an RFI in construction?

The most-cited figure is $1,080 to review and respond to a single RFI, from the Navigant Construction Forum's 2013 study, which also found a median response time of 9.7 days. Earlier research by Mohamed, Tilley and Tucker put the mean cycle time as high as 17 person-hours, most of it spent gathering and cross-referencing information. Both are old enough that they indicate scale rather than current price.

What causes most RFIs in construction?

RFIs cluster at interfaces: between disciplines, between drawings and specifications, and between drawings and the schedules that describe them. Unforeseen site conditions and material substitutions make up a share that no amount of document review prevents. The reliable way to know your own distribution is to categorise your own RFI log by root cause, because it varies by building type, consultant team and standards.

Does BIM reduce RFIs?

Model coordination reduces the RFIs caused by physical clashes, which it is genuinely good at. It does not address inconsistencies in the documents published from the model. A clash-free model can still produce a drawing set with unresolved callouts, schedules that disagree with plans and specifications that name products the drawings never mention.

What is the difference between reducing RFIs and managing RFIs better?

Reducing RFIs means the question is never asked, because the documents answered it. Managing them better means the question is answered faster once asked. The first is won before issue by the design team; the second during construction by the contract administrator. They are usually reported as one initiative, which makes it impossible to tell which one improved.

Can drawing review be automated to prevent RFIs?

The deterministic checks can be: unresolved callouts, duplicate sheet numbers, empty tags, unstated scales, grid and level labels disagreeing between disciplines, and drawing content conflicting with the specification. These are exactly the checks that scale badly by hand and reach site most often. Design intent cannot be automated and should not be presented as if it can.

About Bimiq

Bimiq runs the pre-issue review on a drawing set you upload as a PDF. It reads each sheet, checks it on its own, checks it against every other sheet in the set, and checks the set against your firm's standard, the specification or the building code. Callouts that point nowhere, duplicate sheet numbers, empty tags, unstated scales, grid labels that disagree between disciplines and drawing-versus-spec conflicts all surface with the region of the sheet they came from highlighted, so you can judge each one against the drawing rather than take a report on trust.

It holds the whole set in memory and does the comparisons that scale with the square of the sheet count — the ones that get sampled by hand and arrive later as questions. It does not need a model, a plugin or an install.

See how Drawing QA/QC works, see how RFI Intelligence drafts grounded answers to the ones that still come in, or start a free trial.