Most firm leaders assume a custom AI model shaves hours off a schematic phase and the savings go straight to profit. The opposite shows up on the fee proposal. When a studio trains a model on its own drawings, the hours line shrinks and the thinking line grows, and the proposal has to be rebuilt to match.
Nobody warns a principal about this part. The model draws faster. The decisions it forces get slower. And the way fees have been calculated for decades, pegged to a percentage of construction cost or a lump sum tied to a deliverables schedule, no longer tracks what the firm is selling.
The Fee Formula Breaks Before the Model Does
The traditional architecture fee is a story about labor. A percentage of the construction budget, or a stipulated sum built up from staff hours by phase, assumes that producing options and iterating drawings is the costly, time-consuming part of the work. Price the hours, mark up for overhead and profit, deliver the drawings.
A trained-in-house model flips that around. Early massing studies and concept options that used to take a team two weeks come out of the pipeline in a day, rendered in something close to the studio's own visual language. The deliverable count stays the same, but the labor behind each one collapses.
If the fee proposal still prices twenty concept options at the labor rate of twenty hand-built ones, the firm is quoting work it isn't doing. Price them at the new rate and the number looks suspiciously low, and clients start asking what they're paying for. For a sense of how other practices are framing this to clients, e-architect coverage of why studios are training AI on their own design work walks through what the early-stage pipeline actually looks like when the model has absorbed the house style.
There is a further complication the proposal has to swallow. Economic analysis from the Royal Institute of British Architects has flagged that AI integration will likely erode the cost base of traditional services and push firms toward performance- or outcome-based fees. That shift sounds abstract until a proposal lands on a developer's desk with the schematic phase cut in half.
Why 'Just Charge Less' Is the Wrong Instinct
The obvious response, drop the fee to reflect the compressed hours, fails for three reasons that compound each other.
Rebuild the Proposal Around Decisions, Not Deliverables
A fee proposal that survives a trained-in-house model stops pricing drawings as if they were the product. The drawings are the artifact. What the firm is being paid for is the decisions behind them: which options to pursue, which to kill, which to show the client, which to let the model keep exploring overnight.
Practically, that reshapes a proposal in a few concrete ways. Separate the model-generated exploration from the principal review and curation, and price them differently. Describe the early-stage phase as a decision engagement, not a drawing count. Add a line for model stewardship, the ongoing cost of keeping the training corpus current with the firm's newest built work, because a model that stops learning in 2026 starts drawing like 2026 in five years.
There is also a billing-side problem the industry hasn't settled: how to log the model's processing time next to human time in a way an owner will accept on an invoice. Firms running mixed human-and-AI hours are discovering that their time-tracking systems weren't built for it, and that pricing conversations get awkward fast when the invoice says four hours of principal review and ninety seconds of inference.
What the Principal Actually Does Now
The shift inside the studio is the one the proposal has to reflect. The principal used to spend early-stage weeks nudging a team toward three or four defensible options. Now the model produces the options, and the principal spends that time choosing, rejecting, and articulating why.
The first-option bias gets dangerous. When every rendering looks like the firm's work, the temptation to accept the model's opening move is strong, and the discipline of saying no twenty times before saying yes is the thing the client is really hiring.
A proposal that says this out loud, that spells out how the schematic phase buys the owner the firm's judgment over a wider option space than any team could draw by hand, lands differently than one that quietly bills the same hours for less work. One invites a procurement fight. The other sells what the studio is now built to do.
The Firms Getting This Right Are Rewriting the Proposal Template First
Training the model is the headline. Rewriting the fee proposal is the work that decides whether the investment pays back. Firms treating the two as separate projects are the ones watching margin leak out through a line item nobody updated. Firms treating them as the same project are selling judgment at a defensible price while the drawings get faster underneath.
