Why Architectural Practice Profitability Keeps Shrinking.

The fee scale never recovered, but the workload did

RIBA's fee percentages were written for a different Stage 2 and Stage 3 than the ones practices deliver today. That gap is where the first slice of margin disappears, quietly, before a single design decision has gone wrong.

Planning submissions now routinely carry a Design and Access Statement, a sustainability statement, a pre-application pack, transport and heritage notes where relevant, and a set of visuals that reads more like a marketing brochure than a planning drawing. None of that appeared from nowhere. Local planning authorities have steadily raised the bar on what a "complete" application looks like, and each addition has become baseline expectation rather than an optional extra. A 2010 planning pack and a 2026 planning pack are not the same document by volume or polish, even if the exact multiple is hard to pin down precisely: the direction of travel is unmistakable, and it points one way, upward.

The visuals point is worth sitting with on its own. Photorealistic imagery used to be reserved for the moment a scheme was sold, whether to a client, a funder or an investment committee. Now it is expected at feasibility and planning stage, well before fee certainty exists and often before the scheme itself is settled. That is a structural shift in when visualisation cost lands in a project, not a stylistic preference.

Small practices absorb this because the alternative is losing the commission. Nobody wants to be the practice that pushed back on a DAS request and lost the job to a competitor who just said yes. So the scope grows, quietly, fee proposal after fee proposal, while the percentage-based fee model stays exactly where it was a decade ago.

Where the hours actually go: the stages nobody bills properly

Ask any practice principal where the unpaid hours hide and the answer is rarely one dramatic event. It is a stack of small, recurring leaks, each individually defensible, collectively ruinous.

Concept design is the first and biggest. Clients want to see options before they commit, which is reasonable, but "let's see a few directions" quietly becomes four or five fully worked schemes, each redrawn, each rendered, each presented, all inside a fee that priced for one direction refined twice. Iteration at Stage 2 is treated as part of the base service almost everywhere, even though it is, functionally, additional design work.

The same pattern repeats after submission. A planning committee defers a decision and asks for amended massing. A case officer's comments require a redrawn elevation or an adjusted site strategy. None of that is framed as a variation in most appointments; it is simply absorbed as "getting the scheme through," which is true, but it is also unbilled labour on top of the original design fee.

Then there is the format problem. The same scheme gets redrawn for a client board meeting, reformatted for a planning committee pack, and rebuilt again for a marketing brochure or a competition entry, often in three separate pieces of software with three separate layout conventions. Each rebuild costs real hours that never show up as a line item anywhere.

Coordination overhead compounds this. As projects pick up structural, M&E and landscape consultants, the handoffs multiply, but most fee structures still assume the simple two-way conversation that suited a smaller, simpler job. And Stage 3 often reopens because Stage 2 oversold what the scheme could actually deliver: a beautifully rendered concept that technical coordination cannot support becomes a correction exercise, not a refinement one.

The tooling tax: five subscriptions to do one job

Ask a small practice to list its software subscriptions and the number is usually higher than anyone expects: CAD, a rendering package, an image generation tool, a presentation or layout tool, project management, sometimes a separate file-sharing platform on top. Each is billed monthly, whether or not it earns its keep that month.

The direct subscription cost is only half the problem. The bigger cost is the friction between tools. A single client presentation might mean exporting geometry from CAD, round-tripping it through a rendering package, exporting stills, then rebuilding the whole board again in a layout tool because the render software's own presentation output looks amateur next to a proper competition board. Every one of those handoffs is a file format conversion, a broken layer, a missing texture, a "why has this line weight changed" moment. None of it appears on an invoice. All of it eats fee-earning time.

The maths on subscription fatigue is unforgiving. Five tools paid for at low utilisation cost more per useful hour than one platform used constantly, because the fixed monthly cost never scales down when a project goes quiet. A practice running Linea for drawing and massing, Vector for cleanup and detailing, Retouch for image work and Layout for the final board, inside one project workspace, is not just saving a subscription line. It is removing the re-export tax that quietly consumes an afternoon every time a scheme moves between tools.

Rework is the silent margin killer

Every planning committee deferral triggers a redraw cycle. Every negotiator comment does the same. None of it was priced into the original fee, and none of it comes with a clean mechanism for billing it as a variation, because most appointments never defined what counts as one.

Client indecision at concept stage follows the same shape. Three iterations of a single elevation, four versions of a material palette, a request to "just try it in brick instead" the week before a client meeting: individually small, collectively a second unpaid design phase running quietly alongside the first.

The economics of a single client meeting change entirely depending on how fast a practice can respond inside the room. Producing six alternative material studies in advance, on spec, hoping one lands, is expensive and often wrong. Generating variations quickly and adjusting live in the meeting, based on what the client actually says rather than what was guessed in advance, changes the outcome of that meeting and the hours behind it.

This is where speed to revise stops being a nice-to-have and becomes a margin decision. A practice that can re-light a render or swap a material on an elevation in the room, using Retouch's one-click re-light presets for a different time of day, or a material swap against a reference the client just mentioned, avoids booking a second meeting that will never be billed. Anthropic's current Claude models carry a 1M-token context window on several tiers, at standard per-token rates with no long-context surcharge, and Claude Sonnet 5's introductory pricing of $2/$10 per million tokens has been confirmed as the standard going rate rather than a temporary offer. That matters less as a spec sheet and more as a signal: the tools underneath fast revision work are becoming cheaper to run at scale, not more expensive, which makes speed a viable thing to build a workflow around rather than an occasional luxury.

What actually protects margin in 2026

The practices holding their margin are not the ones working longer hours. They are the ones who have stopped treating scope as infinite and started pricing it as a list of deliverables.

That starts with scoping fees against deliverable count rather than RIBA stage percentage alone. If a scheme needs four concept options, three planning visuals and a sustainability statement, those are three separate, nameable things, not an assumed feature of "Stage 2 design work." Pricing them as line items, even inside a single overall fee, makes the conversation about extra rounds an easy one rather than an awkward one.

Contingency for planning negotiation rounds needs to be priced at the proposal stage, not discovered three months into a committee cycle. A single line in the fee proposal, "includes up to two rounds of committee or negotiator revision, further rounds charged at an hourly rate," changes the entire economics of a deferred application.

Consolidating the toolchain does real, measurable work here too. Fewer handoffs between drawing, rendering and presentation means fewer hours lost to format wrangling and more hours that can actually be billed. A shared project workspace, where a Linea plan, a Vector-cropped detail drawing and a Layout board sit inside the same project rather than being re-exported and rebuilt at every handoff, removes a cost that never shows up on an invoice but shows up every single time in the timesheet.

Pricing tools themselves matters too, and the discipline around it is instructive. Runway's current app pricing runs 12 credits per second for Gen-4 and Gen-4.5, with a cheaper 5 credits per second Turbo tier, and its Standard plan has recently moved to $12 per user per month on annual billing. Ideogram's plans sit at $15 a month for Plus and $20 for Pro, both billed annually, with API image generation reported around $0.03 an image for faster modes. Kling is widely reported to run a free tier alongside paid plans starting around $6.99 to $10 a month, scaling up to premium tiers in the $128 to $180 range, though the exact matrix varies by region and reseller and should be checked before it's quoted to a client. The lesson from all of it is the same: tool cost is manageable when it is tracked against actual use, and unmanageable when it is five separate subscriptions running quietly in the background regardless of whether anyone opened them that month.

Finally, revision speed itself is a service worth pricing, not just a workflow improvement. A practice that turns client feedback around inside the same meeting, or same day, is offering something clients will pay a premium for: responsiveness, not just deliverables.

Margin does not disappear in one bad decision. It leaks out through scope that was never priced, iteration that was never billed, and handoffs that were never necessary.

The takeaway

Profitability is not shrinking because architects are working less efficiently. It is shrinking because the fee model has stayed still while the deliverable list kept growing, and because the small, recurring costs, unbilled Stage 2 rounds, unpriced planning feedback, five subscriptions doing the work of one workflow, compound quietly across every live project. None of these problems has a single dramatic fix. They respond to the same treatment: name the deliverable, price the revision round, and remove the handoffs between drawing, rendering and presentation that cost hours nobody ever invoices for.