Denver Approved 37% of Building Permits on the First Try. An AI Is Fixing the Other 63% Before They Hit the Queue.

In March 2026, the City and County of Denver signed a five-year, $4.6 million contract with CivCheck, an AI-powered plan review platform built by Clariti. CivCheck's job is not to approve permits. It is not to replace the plan reviewers who evaluate structural loads, fire separation distances, and egress paths. Its job is to read every incoming application and flag the missing documents, incomplete fields, and obvious code violations before a human reviewer ever opens the file.

Denver needed it badly: only 37% of permit applications in the city were approved on the first submission. The other 63% bounced back, sometimes multiple times, clogging the review queue with rework that consumed staff hours better spent on substantive engineering review. As Julia Richman, vice president of government relations at Clariti, put it: "Most plan review delays start upstream, when submissions enter the queue incomplete or inconsistent."

Denver is not alone. And the pattern it reveals is not about AI replacing government workers. It is about a structural defect in how American cities process building permits, one that costs residential builders and homeowners billions in delay that nobody bothers to quantify because everyone assumes the problem is the reviewer.

22.9 Days Is the Published Number. Reality Is Worse.

PermitPlace's 2026 State of Building Permits is the most comprehensive public analysis of permit review timelines ever published: 741 cities across 44 states, drawn from a proprietary database built over 20 years of commercial permit expediting. National average for initial plan review: 22.9 days. Median: 14. Chicago, the slowest major metro, publishes a 92-day initial review. San Francisco: 60 days, Portland: 51.

Those are the published numbers, which PermitPlace is careful to note are "best understood as starting points for project planning, not guarantees." Their operational experience across 2,000-plus jurisdictions puts the gap between published timelines and actual commercial project timelines at 2x to 5x. A new residential permit in Oakland averages 80 days. A major residential permit in Aspen, Colorado, takes 16 to 18 weeks for the first review round alone, with each subsequent correction round adding another 8 to 10 weeks.

But the headline number masks the mechanism. Most of that time is not a reviewer hunched over blueprints, meticulously checking your joist spacing against IRC Table R502.3.1(1). Most of it is your application sitting in a queue, waiting for someone to discover that page 7 of your structural calculations is missing, that your site plan doesn't show the required setback dimension, or that you submitted fire-resistant wall assembly details for a jurisdiction that requires a different rating than the one you specified.

What a Single Rejection Cycle Actually Costs

Nobody publishes this figure. So I calculated it.

A typical single-family residential project in a mid-cost metro carries a construction loan between $500,000 and $700,000 at current rates of 8% to 10%. That is $110 to $190 per day in interest alone while the project sits idle. A permit rejection-and-resubmit cycle adds two to six weeks depending on the jurisdiction, the complexity of the corrections, and how quickly the architect or engineer can revise and resubmit. At the conservative end, call it 15 business days.

Interest on a $600,000 loan at 9% for 15 idle days: $2,219. Your general contractor's crew was scheduled to start framing the week after you expected your permit. They moved to another job. Rescheduling costs $500 to $2,000 depending on your market and how tight your GC's pipeline is. Your architect charges $500 to $1,500 to revise and resubmit. Add permit resubmission fees in jurisdictions that charge them.

Conservative total for a single rejection cycle on a single residential project: $3,200 to $5,700. Not catastrophic on any single project, but not trivial either, and 63% of Denver applicants were eating this at least once, and many ate it twice.

Denver issues roughly 4,000 residential permits per year. At a 63% first-submission rejection rate, with an average cost of $4,500 per rejection cycle, the aggregate friction cost to Denver's residential construction economy approaches $11.3 million annually. That cost does not appear on any city budget line. It appears on your closing statement as a slightly higher home price, baked into the builder's margin because everyone treats permit delays as weather.

Spell-Check for Zoning

In Surrey, British Columbia, Jerome Thibaudeau, the business transformation manager of the city's planning and development department, found that 80% of residential permit applications contained "significant zoning deficiencies," requiring an average of 1.6 resubmissions per application. That is not 80% of applications with engineering concerns or structural problems. That is 80% of applications with zoning deficiencies that an automated system could check in minutes.

Thibaudeau described Archistar's AI PreCheck platform, which Surrey launched for public use on March 10, 2026, as "spell-check for zoning." The analogy is precise, and it is revealing. Nobody argues that spell-check replaced editors. Nobody claims that catching a misspelled word is the same as evaluating whether a sentence is true. But a manuscript full of typos wastes an editor's time before they can engage with the substance of the writing, just as a permit application missing its setback dimensions wastes a reviewer's time before they can engage with the substance of the design.

Archistar's platform uses computer vision and machine learning to read submitted architectural drawings, maps every element against a city's digitized zoning and building codes, and generates a standardized compliance report in minutes. More than 30 cities globally are implementing or piloting the system, including Vancouver, Austin, Houston, and Seattle. Archistar claims up to a 55% reduction in permit review cycles and first-submission quality boosted to 90% for applications that pass through the pre-check.

Honolulu's experience reinforces the pattern. Dawn Takeuchi Apuna, director of the city's Department of Planning and Permitting, launched the CLARITI system in February 2026, cutting residential permit processing times by 70 percent. She is now integrating CivCheck on the front end, which she describes as working "not unlike TurboTax" for permit applications. Her assessment of the underlying problem is blunt: "I think that for more than 50 percent of the building permit plans that come in, the quality is not good."

What the AI Cannot Do

Let me be direct about the limits, because the vendor claims are running ahead of the evidence.

AI pre-check tools work on standardized, rule-based checks: setbacks, lot coverage, document completeness, dimensional requirements that can be digitized and compared against a submitted drawing. This is the low-hanging fruit of permit review, and it is genuinely the part that generates the most rejection cycles. But it is not the part that generates the longest delays on complex projects.

Discretionary design review, where a planning commission evaluates whether a proposed home fits the character of an existing neighborhood, cannot be automated. Complex engineering judgment in unusual structural situations, seismic zones, or flood-prone areas still requires a human who has seen what happens when the textbook answer is technically code-compliant but structurally dangerous. The reviewer who has processed 10,000 plans and catches the "technically legal but I wouldn't build it that way" detail is not replaceable by a computer vision model trained on drawings. Political and community opposition to a proposed project is not a code compliance question.

Archistar's "55% reduction in review cycles" and "90% first-submission quality" claims are self-reported. No independent audit has verified these numbers across a statistically meaningful sample of jurisdictions. CivCheck's target of moving Denver from 37% to 80% first-try approval is a goal, not a result. The only verified performance data in this space comes from NREL's SolarAPP+ program, which processed 18,906 residential solar permits in 2023, completed projects 14.5 days faster than traditional permitting, and saved 15,400 hours of staff time. SolarAPP+ works because residential solar permits are structurally simple: standard equipment configurations against known roof types. Extrapolating those results to full residential building permits, which involve dozens of interconnected code sections and multiple review disciplines, requires caution.

59 Positions

Denver's Community Planning and Development Department cut 59 budgeted positions for 2026, bringing its total staff to 251. The CivCheck contract and the staffing reduction were announced in the same season. There is no public statement linking them, and they may reflect entirely independent budget decisions. But the optics are the risk.

If AI pre-check tools handle the volume of routine checks that currently consume reviewer time, cities face a choice: redeploy those hours toward the complex reviews that create the longest delays, or cut the hours entirely. Denver set a 180-day shot clock for permit decisions with a promise to refund developers up to $10,000 in application fees if the deadline is missed. That creates a financial incentive to process applications faster. Whether it creates an incentive to process them better is a separate question that nobody has answered yet.

Altamonte Springs, Florida, ran an early pilot with a company called AutoReview.AI in 2023. City manager Frank Martz described the results in HousingWire: "What used to take us a week took us three to four minutes." He also noted that the AI found errors in the city's own code that human reviewers had not caught. AutoReview.AI subsequently folded. The technology worked. The company did not survive. That is a different kind of risk for cities betting on AI vendors for critical infrastructure.

What This Means If You Are Filing a Permit

If you are a homeowner planning an addition, a small builder filing for new construction, or a contractor submitting a renovation permit, the practical takeaway is narrow but real.

Check whether your jurisdiction offers an AI pre-check tool. Austin, Denver, Honolulu, Surrey, and Los Angeles have them live or in beta. Seattle and several other cities will follow in 2026. Where available, run your plans through the pre-check before formal submission. These tools are designed to catch the exact errors that trigger rejection: missing documents, setback violations, incomplete calculations, dimensional discrepancies between the site plan and the floor plan. A clean first submission is worth $3,000 to $6,000 in avoided delay costs on a typical residential project.

Where no AI tool exists, the self-help version is the same discipline: call the building department before you submit, ask what the most common rejection reasons are, get a checklist, and have your architect or engineer review the application as if they were the city reviewer. The technology is just formalizing what good permit expediters have done for decades.

In the United Kingdom, Google DeepMind and the government have built a prototype now testing in three English councils, aiming to halve household planning application processing from eight weeks to four. The U.S. Department of Energy announced a formal initiative targeting "rapid permitting" as part of its AI in buildings program. The direction is clear, even if the pace is uneven.

None of this will fix the fundamental problem, which is that American permitting is fragmented across roughly 30,000 local jurisdictions, each with its own codes, its own amendments, its own interpretation culture, and its own staffing constraints. AI pre-check tools can eliminate the most expensive form of waste in that system. They cannot consolidate it.

Limitations of This Analysis

PermitPlace's data reflects published timelines from 741 cities, not actual processing times. Their own experience puts real-world delays at 2x to 5x the published figures. My cost-per-rejection-cycle calculation uses Denver-specific assumptions for loan size, interest rate, and crew rescheduling costs; it will overestimate costs in lower-cost markets and underestimate them in places like San Francisco or New York. CivCheck and Archistar deployment results are vendor-reported. No independent audit exists. SolarAPP+ data from NREL is credible, but solar permits are structurally simpler than full building permits and the comparison has limits. Surrey's 80% deficiency rate is specific to one Canadian jurisdiction's R3 residential zone and may not generalize to U.S. jurisdictions with different code structures. I was unable to find data on whether AI pre-check tools reduce the type of errors that cause the longest delays, as opposed to the most frequent but quickest-to-fix errors.