Essay 02 · May 2026

The Coordination Tax, Decomposed

Every analytical piece on American residential construction eventually arrives at the same shape: custom homes routinely come in fifteen to thirty percent over their original budget and roughly twenty percent past schedule, and the numbers have been quietly worsening for two decades. McKinsey’s “Reinventing Construction” report, in 2017, put schedule overruns at twenty percent across asset classes and budget overruns as high as eighty percent on the largest projects. NAHB’s longitudinal surveys show the median time to complete a custom single-family home rising from roughly seven months in the early 2000s to roughly twelve months in recent years, even as the per-square-foot output of the median crew has been roughly flat. Productivity has gone backwards in absolute terms, in one of the largest sectors of the American economy.

The interesting question is not whether this overrun exists. It is where the dollars and the months actually live.

This essay is an attempt to decompose that overrun — to separate the cost of building a complex, site-specific, regulated structure from the cost of coordinating the twenty-some independent firms that build it. The first is unavoidable. The second is what we call the coordination tax. The thesis of this piece is that the coordination tax is roughly half of the observed overrun on a typical custom build, that it is largely software-addressable, and that the economic gravity it represents is the reason an orchestration layer in this industry is now buildable and inevitable.

We will work through it bucket by bucket.

What counts as the coordination tax

The coordination tax is the set of costs that exist only because separate firms have to exchange information and establish trust to work together. It is not the cost of skilled labor on site. It is not the cost of materials. It is not the cost of regulatory compliance itself. It is the cost of the handoffs — between architect and engineer, between designer and builder, between general contractor and subcontractor, between plan reviewer and applicant, between owner and everyone — and the cost of the queue time, information loss, and rework those handoffs create.

A useful test: if a single firm with all relevant disciplines under one roof would not incur this cost on the same project, it is coordination tax. If it would still incur the cost — because the work itself is hard, the site is genuinely uncertain, or the owner genuinely changed their mind — it is not.

By this test, the coordination tax on a typical custom residential build decomposes into roughly five buckets.

Bucket one — pre-construction delay and financing carry

The phase between “we have decided to build” and “the slab is being poured” is the most expensive invisible stage of a custom home project. On most projects it runs six to nine months. Some of that is unavoidable work — schematic design, structural engineering, geotechnical survey, plan review, lender underwriting. The rest is queue time. The architect waits for owner decisions. The engineer waits for the architect. The lender waits for the appraiser. The appraiser waits for plans. The builder waits to bid until plans are stamped. Each handoff sits in a queue measured in days or weeks, accumulating.

The cost of this queue time is mostly invisible to the owner because it is paid in financing carry rather than in invoices. On an $800,000 project with a construction loan at current rates, every extra month of pre-construction delay is roughly $4,000–$6,000 in pure interest carry, plus property taxes on the lot, plus rental costs if the family is renting during the build, plus the opportunity cost of capital deployed in pre-construction soft costs. Three avoidable months — a conservative estimate on most projects — is twenty to thirty thousand dollars of friction cost before a single foundation has been dug.

Bucket size: 1–3% of project cost. Coordination-driven share: most of it. Software-recoverable share: most of the coordination-driven share, because the rate-limiting step is information latency rather than physical work.

Bucket two — design–build handoff losses

The architect and the builder are two different firms with two different optimization functions. The architect is optimizing for design intent, code compliance, and the client’s stated aesthetic. The builder is optimizing for buildability, sub availability, current materials pricing, and schedule. These optimization functions overlap but do not coincide. When the architect hands the builder a stamped set of drawings, the builder discovers — sometimes immediately, sometimes weeks into procurement — that what is on the page assumes a sub-tier of supplier that does not exist in this market, or a material that is now backordered four months, or a detail that adds twelve percent to the framing labor estimate.

What happens next is value engineering after the fact. Plans go back to the architect. Some details are changed. Some changes require re-permitting. Some require re-bidding the affected scope. Each loop adds weeks and adds fees, and in the worst cases compromises the design intent the owner paid the architect to produce in the first place.

Bucket size: 3–5% of project cost. Coordination-driven share: most of it. Software-recoverable share: partial. Real-time builder pricing data, jurisdiction-specific code checks, and supplier-availability feeds could front-load most of these decisions into the design phase, when changes cost a tenth of what they cost mid-construction. The portion that depends on craft judgment between human professionals is not going away.

Bucket three — change orders driven by late-arriving information

The largest single component of the coordination tax on most residential projects is change orders. Industry data, including Construction Industry Institute research and the benchmark reports from major project-management platforms, consistently puts change orders at five to ten percent of contract value on a typical custom home, with the worst projects running fifteen percent or higher. The conventional framing of change orders is that they are owner-driven — the family decided they wanted a different countertop, a bigger window, a different room layout — and therefore are not really overruns but additions.

This framing is wrong about most change orders. The majority of so-called owner changes are responses to information that arrived too late to be incorporated at the design phase. The owner did not know the kitchen island they wanted would require a structural beam they could not afford until the framer was on site. The owner did not realize what the chosen window package would actually feel like until the rough openings were cut. The owner did not see the cost of the upgraded HVAC system until the bid came back, by which point the duct chases had already been framed for the original system.

These are not preference changes. They are information failures. The information existed somewhere in the supply chain — at the structural engineer, at the window supplier, at the mechanical sub — but it did not reach the owner in time to be considered without disrupting work in progress. The change order is the cost of importing that information late. Change orders are typically priced at a thirty to fifty percent premium over the same scope quoted in the original bid, because the change disrupts crew sequencing and procurement that has already been set in motion.

Bucket size: 5–10% of project cost. Coordination-driven share: most of it — in our analysis, perhaps three-quarters. Software-recoverable share: substantial. Better visualization, real-time cost feedback during design, and supply-chain integration at the schematic stage could move a significant fraction of these decisions to before construction starts, when they cost a fraction of what they cost mid-project.

Bucket four — subcontractor sequencing and on-site rework

A typical custom home involves eight to fifteen subcontractors who arrive on site in a sequence orchestrated by the general contractor’s site superintendent. The orchestration tool, in most cases, is a combination of phone calls, text messages, a shared spreadsheet, and the superintendent’s memory. When the orchestration breaks down — sub A finishes early, sub B is not ready, sub C arrives and finds the previous work unfinished — the costs are paid in mobilization fees, in rework, and in cascading schedule slippage that pushes financing carry out further.

A common pattern: the electrical sub roughs in to a framing plan that the framer adjusted on the fly without telling anyone, requiring partial re-rough. The plumber arrives the next day and has to re-route a stack around the electrical change. The drywall sub arrives a week later, on the original schedule, and has to reschedule because the framing inspection has not yet been passed. Each of these errors costs hundreds to thousands of dollars individually and adds up to a meaningful percentage of project value over the course of a build.

Bucket size: 2–4% of project cost. Coordination-driven share: most of it. Software-recoverable share: partial. Better scheduling, real-time site status, and automated dependency tracking can prevent a significant fraction of these failures, but on-site physical work will always require humans who can adapt to discovered conditions. Roughly a third of this bucket is plausibly recoverable.

Bucket five — permitting and code friction

Every American jurisdiction has its own building code, its own plan review process, its own list of preferred details and habitual rejections, its own queue times, and its own staff personalities. An architect or builder operating in three counties knows three sets of jurisdictional preferences in real, lived detail. They do not know the preferences of the fourth county. The fourth county will, on review, send back plans with a list of revisions — some legitimate, some preferential, some idiosyncratic — and the loop begins. On well-resourced projects, plan review iteration adds two to six weeks. On under-resourced projects, it can add six months.

The financing carry on a six-week permit delay is real money. The redesign fees are real money. The lost sub bookings are real money. None of this work produces a single nail driven; it is pure friction tax.

Bucket size: 1–3% of project cost. Coordination-driven share: nearly all of it. Software-recoverable share: most of it. Jurisdictional code in a structured database, automatic pre-submission review against jurisdiction-specific preferences, and direct integration with permit office systems could compress most of this friction. The portion that depends on genuine judgment calls by code officials will remain.

Adding it up

The five buckets sum, on a typical custom residential build, to somewhere between twelve and twenty-five percent of project cost. The fifteen-to-thirty-percent figure that headlines the analytical literature includes some inherent costs — true scope changes, true site discovery, true labor and materials inflation — but the coordination component is the bulk of it. On an eight-hundred-thousand-dollar custom home, the coordination tax is somewhere between $96,000 and $200,000 — call it $150,000 as a central estimate.

The portion that is plausibly recoverable by an AI-native orchestration layer — by which we mean software that can sit in the translation layer of a bespoke project, hold context, route information between specialists at the speed of an API call rather than the speed of a phone call, and surface decisions to the owner with full cost and schedule context — is roughly half of the total. Seven to fifteen percent of project cost is the addressable surface area, or sixty to one hundred twenty thousand dollars on a typical custom build.

We are not the first people to notice this. The industry has tried to capture it through every prior wave of construction-tech innovation — design-build firms, integrated project delivery, BIM, lean construction, modular and prefab, single-platform project management software. Each made progress on a slice of the problem. None captured the orchestration value at scale, because none removed the underlying constraint: that coordination across many small firms required human translation work that did not scale linearly with the projects.

That constraint is gone now. The technology that can do the translation work cheaply — language models that read drawings, parse contracts, monitor schedules, generate code submittals, summarize sub bids, and produce decision-ready briefings for owners — is operational and improving on a quarterly basis. The economic gravity of the coordination tax is large enough to support an entirely new layer of the industry. The layer that captures it will not be a builder, an architect, or a manufacturer. It will be an orchestrator that takes the friction tax as its revenue source and leaves the craft margins of the specialists intact.

Why the incumbents cannot capture this

A natural objection from inside the industry is that, if the coordination tax is this large and this recoverable, the existing players will simply capture it themselves. Builders will integrate software. Architects will adopt better tools. General contractors will run tighter sequences. The orchestration layer will be absorbed by the incumbents rather than displacing them.

There is a structural reason this is unlikely. The coordination tax is, in part, the existing players’ margin. A general contractor’s fifteen-to-twenty-percent markup is partly compensation for the coordination work the GC is doing. An architect’s eight-to-twelve-percent fee partly compensates for the iterative back-and-forth with engineers, owners, and builders. The lender’s points partly compensate for manual underwriting work. Each of these firms is rationally paid to do the coordination work, and each of them would not rationally be paid the same fee for a project where the coordination work was done by software.

This is the same dynamic that played out in travel, retail, and finance. The incumbents in each of those industries were rationally compensated for friction work that a new layer of technology eventually made cheap. They could not, individually, capture the savings, because the savings would have come from eliminating work they were paid to do. Only a new layer — Expedia, Amazon, Stripe — could capture the friction tax as new revenue, because its revenue source was the friction itself.

In construction, the orchestrator is the new layer. Its revenue comes from removing the friction tax. The specialists — builders, architects, engineers, manufacturers, lenders — continue to be paid for the craft work they actually do, on better-defined projects, with fewer dead ends, and with more qualified inbound demand. Their margins on the work they do may even improve. What they lose is the coordination work they did not want to be doing in the first place, and the coordination tax is paid to someone whose entire business model is built around capturing it.

What this means

The first essay in this series argued that the design-to-build industry is structurally ready to be re-aggregated by an AI-native orchestration layer. This essay is the economic case beneath that argument. The coordination tax is large, structural, and recoverable. The economic gravity is sufficient to support a new layer of the industry. The technology to capture it is operational. The discovery mechanism through which customers will find that layer — conversational AI, asked in plain English to help with a complex purchase — is moving rapidly into place.

The orchestrator that captures the coordination tax does not need to be a tenth of one percent more capable than the existing industry at building. It needs only to remove a layer of friction the existing industry was never designed to remove. The math, on every project, is doing the work.

This essay is published by Ilios, LLC. Ilios is an early-stage company building AI-native orchestration infrastructure for end-to-end design-to-build of high-efficiency structures — homes, multi-family, offices, and any other structure type. We work across the full project arc: lot sourcing, financing, architectural design, builder and prefab selection, permitting, and construction coordination, routing each project to the best-suited suppliers in its market. Ilios is a Delaware LLC registered to do business in Maryland. Builders, manufacturers, and designers interested in being surfaced through an AI-native discovery layer, and consumers or businesses beginning a project, can reach us at contact.