Three quarters of UK architecture practices now use AI on most of their projects. RIBA's AI Report 2026, the institute's third annual survey of its membership on the subject, puts adoption at 74%, up from 59% twelve months earlier. A fifteen-point jump in a single year is not a trend line. It is a tipping point, and it changes what a practice has to be able to answer about its own numbers.
What makes the 2026 data interesting is not the adoption figure itself. It is the distance between the enthusiasm and the evidence. Seventy-five per cent of practices report productivity gains. Far fewer can point to where those gains landed in the accounts. This article looks at what the report actually found, what it means for architects and architectural practices running on thin margins, and how to tell the difference between AI that saves hours and AI that saves money.
Why AI adoption in UK architecture matters right now
RIBA has surveyed its members on AI three years running, which makes the 2026 report the first with enough history to show direction rather than a snapshot. The direction is unambiguous. Practices using AI for most projects rose from 59% to 74% in a year. That is no longer an innovation-team story or something confined to large multidisciplinary offices. It now describes the median UK practice.
The tools have broadened as well. Early adoption clustered around image generation and early concept work. Current use spreads across specification drafting, planning and design and access statements, precedent research, option testing, energy iterations, bid and fee-proposal writing, and a large volume of routine project administration. Much of that work sits around the design rather than inside it, which is precisely why the productivity numbers look strong while the design-quality numbers do not move.
Two findings in the report deserve more attention than they have had. Fifty-nine per cent of practices expect AI to lead to staff reductions, which tells you how seriously principals take the capacity implications. And the report arrives while RIBA is developing an AI Overlay to the Plan of Work, mapping stage by stage where AI can legitimately be used and where human authorship and accountability must remain. That overlay will become the reference point clients, insurers and professional indemnity underwriters reach for when they ask how AI was involved in a piece of work. Practices that cannot answer with records rather than recollection will find that conversation uncomfortable.
Key takeaways
- AI use on most projects rose from 59% to 74% in a single year, according to the RIBA AI Report 2026.
- 75% of UK practices report productivity gains from AI tools.
- 59% expect AI adoption to lead to staff reductions in their practice.
- Productivity gains only become profit where recovered hours are deliberately redirected.
- Stage-level time and fee data is what turns an AI claim into an auditable business case.
What this means for architects and architectural practices
For practice principals, the commercial risk in these numbers is a quiet one. If AI genuinely removes hours from a stage and nothing else changes, the saving does not arrive as margin. It arrives as slack, and slack gets absorbed by extra design iterations, unbilled scope creep or a longer, more comfortable programme. The practices seeing a real return are the ones that noticed the hours coming out and made a decision about where they went.
For associates and project architects, the shift is in the mix of the work rather than its volume. AI compresses documentation, drafting and administrative tasks while leaving coordination, client management, site work and professional judgement broadly untouched. That changes what a chargeable week looks like. A practice reporting utilisation annually, averaged across everyone, will not see the change at all until it shows up in a fee dispute.
For practice managers and finance leads, there is a new cost line and a new evidence burden arriving together. AI licences are visible, recurring and easy to total. The benefit is diffuse, spread thinly across dozens of tasks on dozens of projects. Without stage-level time data, renewal season becomes a matter of opinion rather than analysis. And with 59% of practices expecting headcount consequences, being able to distinguish a genuine capacity gain from an enthusiastic impression is not a nice-to-have.
Fee conversations will change
Informed clients already ask whether AI reduced the effort behind a fee. A defensible answer needs recorded hours by stage, not an estimate. With data, it is a negotiation. Without it, it is a concession.
Utilisation shifts before it falls
The chargeable mix moves long before headline utilisation does. Documentation time drops while review and coordination time rises. Annual averages hide the shift until it is expensive.
Checking becomes real work
Verifying and signing off AI-assisted output is professional labour, and the forthcoming AI Overlay to the Plan of Work will formalise it. Track that review time now so you can price it later.
How UK practices are measuring the return today
Most practices sit somewhere between a well-kept spreadsheet and a stack of tools that do not talk to each other. Both are reasonable, and both hit the same wall. Proving that AI improved profitability means joining time, fee, stage and resource data together, and neither approach joins them without someone doing it by hand every month.
| What you need | Spreadsheets | Point tools | Quantim |
|---|---|---|---|
| Live fee and WIP visibility | Accurate the day it is updated, then it drifts. Reconciliation is manual. | Strong when finance-led, but usually blind to design stage progress. | Fee, stage and WIP position update continuously as time is booked. |
| Stage-level hours baseline | Possible with discipline, but retrospective entry weakens the record. | Timesheet tools handle this well in isolation, then need exporting. | Stage-level timesheets feed reporting directly, with full history. |
| Forward resource forecasting | Workable for a few projects; breaks down across a portfolio. | Good visual planners exist, but rarely reconcile to recorded time. | Forecast and actuals sit in one model, so variance is visible weekly. |
| Before-and-after comparison | Achievable with tagging, but nobody maintains it for long. | Rarely supported; few tools understand comparable project types. | Compare hours per stage across similar projects and periods natively. |
| Effort at month end | Days of consolidation, and one broken formula changes the answer. | Several exports, then a spreadsheet to join them anyway. | Dashboards are already current, so reporting is review, not rebuild. |
How Quantim helps
Quantim is the operational layer that sits underneath the design tools. It is where the time savings become visible as billable capacity and margin, using the resourcing and fee data practices need to prove a return. Design software changes how the work gets done. Quantim answers the question a director has to take to a board meeting: did it improve profitability, and by how much?
- Baseline before you scale. Record stage-level hours across three to five typical project types so you have a comparable starting point. Any later claim about AI savings without a baseline is unprovable.
- Watch the work mix, not just the total. Track which stages and task types are compressing and which are growing. Falling documentation hours alongside rising review and coordination hours is the expected pattern, and it has fee implications.
- Convert the saving deliberately. Feed real numbers back into fee proposals and resource forecasts so recovered hours become either margin or additional capacity, rather than quietly disappearing into longer programmes.
See how your practice can turn AI productivity gains into billable capacity and measurable margin.
Request a demoAdoption was the easy part. The practices that profit from AI will be the ones that can show, in recorded hours, exactly what it changed.
"We rolled AI tools out across the studio and everyone agreed the work felt faster. It was only when we compared recorded hours per stage against the same project types from the year before that we could see where the saving actually sat, and how much of it we had already given away in extra iterations."
Practice Director, 45-person architecture practice, Manchester
Checklist for practice leaders
- Establish a stage-level baseline of recorded hours on three to five typical project types before expanding AI use further.
- List every AI licence and subscription with its true annual cost, including training and onboarding time.
- Agree a written position on where AI may and may not be used, ready to align with the RIBA AI Overlay to the Plan of Work.
- Review fee proposals for stages where AI has genuinely reduced effort, and decide consciously whether to reprice or retain the margin.
- Schedule early-career learning time explicitly in the resource forecast rather than assuming it still happens.
- Report AI return on investment quarterly using recorded hours and fee recovery, not anecdote.
Frequently asked questions
How is AI changing UK architecture practices?
Mainly by absorbing the work that surrounds design rather than the design itself: drafting specifications and planning statements, summarising precedent and consultation material, generating early massing and layout options, running rapid environmental iterations, and producing bid and fee documentation. The RIBA AI Report 2026 found 74% of practices now use AI on most projects, up from 59% a year earlier, and 75% report productivity gains. The practical effect is that the mix of chargeable work shifts towards coordination, review and judgement, which has direct consequences for how fees are structured and how utilisation should be read.
What are the AI productivity gains for architecture firms worth in practice?
They are worth whatever the practice deliberately does with them. Three quarters of UK practices report productivity gains, but a gain only reaches the bottom line if the recovered hours are converted into fee-earning capacity, faster stage turnaround or retained margin on a fixed fee. Where hours are not recorded by stage, saved time is typically absorbed by additional design iterations or an extended programme and never appears in the accounts. Baseline data captured before adoption is the single thing that separates a provable gain from a plausible story.
What are the best AI tools for UK architects in 2026, and how should a practice choose?
The RIBA report deliberately does not rank products, and there is a good reason for that: the right tool depends on where your practice actually loses time. A studio bleeding hours in specification writing needs something different from one losing them in planning documentation or option testing. The more useful selection method is to measure where hours currently go, trial against that specific bottleneck, then compare recorded hours before and after on similar projects. The full survey findings are published by RIBA.
AI in architecture has passed the point where adoption alone says anything useful about a practice. What separates the firms seeing a real return from everyone else is the ability to see hours, fees and utilisation in one place and act on what they show. Explore the features and benefits that help UK practices turn productivity gains into measurable profit.