Most of the AI conversation in architecture right now is about design tools. Generative visualisation, parametric modelling, concept iteration through image models. These are interesting developments and they get a lot of coverage. But for a studio of five to fifteen people, a significant portion of every working week has nothing to do with design at all. It is spent on proposals, document processing, client reporting, consultant coordination, and the various forms of admin that keep a practice running. That operational work is where AI is starting to make a measurable difference, and smaller studios are picking it up faster than the larger firms.
Where the conversation has been, and where it is going
The RIBA AI Report 2025 found that 89% of architects are already using AI for report writing, and 58% for bid creation. Those numbers are higher than most people in the profession would guess. But the same report showed that only 28% are using AI for project cost management, 29% for contract management, and 29% for fee calculation. Architects have adopted AI for the writing-heavy parts of practice management but have not really started applying it to the financial and operational side in any consistent way.
This is partly because the design-focused AI conversation has been so dominant that the operational applications have received far less attention. For a small studio where the same three or four people handle both design and business development, the time spent on operational tasks is time taken directly from billable work. Any tool that reduces that overhead has an immediate and visible impact on the practice’s capacity.
What operational AI actually looks like in a small practice
There is no single AI platform that handles everything a small studio needs on the operational side. What is happening instead is that practices are applying specific tools to specific bottlenecks, and the pattern of adoption tends to follow wherever the most repetitive work sits.
The specific bottleneck varies by practice, but the common thread is work that follows a predictable pattern and takes significant time without producing anything billable. Processing tender documents and planning submissions to extract key requirements, deadlines, and constraints is one that comes up often. So is drafting proposal sections by pulling language and structure from previous successful bids and adapting it for a new brief. Studios are also using AI to generate project status updates from timesheet data and programme records rather than writing them from scratch each week, or to pull action items out of consultant correspondence that would otherwise take time to read through in full. These are all tasks that take hours without producing anything the practice can bill for, and they are the ones where AI is making the most obvious difference.
A study by Monograph found that desk-based workers in architecture and construction practices using AI tools saved an average of 4.11 hours per week on administrative tasks. For a small studio where every team member’s time shows up in the fee tracker, four hours a week is not a marginal improvement. It is close to half a day of billable capacity recovered for each person using the tools.
Document processing and brief extraction
Of all the operational use cases, document processing is probably the most immediately practical for architecture practices. Studios handle a constant flow of planning documents, client briefs, tender packs, consultant reports, and regulatory submissions. Extracting the structured information from those documents, the site constraints, planning conditions, key dates, compliance requirements, and client priorities, is manual work that someone in the practice does repeatedly on every project.
AI document intelligence tools can now read through unstructured documents and pull out the specific information that matters for a given workflow. A tender pack that would take someone an afternoon to read, annotate, and summarise can be processed in minutes, with the key requirements presented in a structured format that the project team can review and act on. The technology is not flawless and still needs someone checking the output before it goes anywhere, but it handles most of the extraction work and flags what needs attention.
For smaller firms, the choice between off-the-shelf document processing platforms and something more tailored to their workflow is a real one. Enterprise tools built for large contractors do not always map well onto how a 10-person design studio actually handles documents. The document types, the extraction needs, and the workflow that follows are all different from what those enterprise tools were designed around. This is part of the reason that more studios are looking at custom AI solutions designed for smaller firms, where the tool is built around the practice’s actual document types and extraction needs rather than a generic configuration that assumes a different scale of operation.
Proposals and bid preparation
The RIBA figure of 58% using AI for bid creation reflects how quickly this particular use case has been picked up, and it makes sense when you consider how much time proposal writing consumes in a small practice. A competition entry or a framework bid involves pulling together capability statements, project references, methodology descriptions, team CVs, and responses to specific evaluation criteria. Much of this content exists somewhere in the practice’s previous submissions, but finding the right sections, adapting them, and assembling everything into a coherent document takes hours of someone’s time on every bid.
AI tools designed for the AEC sector allow practices to draw on their library of previous submissions and generate first drafts of proposal sections that are already adapted to the language and requirements of the new brief. Joist AI is one example of a platform built specifically for this, and there are others emerging. These tools do not replace the judgement of the partner deciding which projects to pursue or the project director shaping the design approach. What they do is reduce the time spent on the writing and assembly that sits between having a good idea for a bid and getting it into a document the client will read.
Why smaller practices are adopting faster
The headline numbers on AI adoption in architecture still look low. The Bluebeam 2026 AEC Technology Outlook survey found that only 27% of AEC firms are currently using AI in any form, and a joint AIA and Deltek study from March 2025 put the figure for fully integrated AI adoption at just 6%. But these numbers include firms of all sizes, and the adoption pattern is not evenly distributed.
Smaller studios are moving faster, and the reasons are mostly structural rather than technological. A five-person practice does not need a committee to approve a new tool. If the principal decides on Tuesday to try an AI document processing tool, it can be running by Thursday, and the team will know within a fortnight whether it is saving them meaningful time. Larger firms tend to have IT procurement processes, security reviews, and pilot programmes that stretch over months before anyone uses the thing in production.
There is also a practical incentive that matters more at smaller scale. If the admin overhead on proposals takes 10 hours a week, a larger firm can spread that across several people without it dominating anyone’s schedule. A five-person studio does not have that option, and the people absorbing those hours are usually the same people who should be spending their time on design work and client relationships. The efficiency gain from AI is proportionally more valuable when there are fewer people available to absorb the operational load.
The data foundation question
AI tools for operational tasks work best when they have access to the practice’s existing data in a reasonably structured form. A studio that keeps its timesheets in one system, its fee tracker in a spreadsheet, its project documents across three different folder structures, and its previous proposals in a shared drive with inconsistent naming conventions will get less from an AI tool than one where the basics are organised and findable.
This does not mean a practice needs to overhaul everything before it can start using AI. It means that the first AI tool a studio adopts will often highlight where the underlying data could be better organised, and the process of tidying things up tends to be worthwhile on its own terms regardless of what happens with the AI. Most studios are closer to having a usable data foundation than they assume. The issue is usually not that the information does not exist, but that it has never been structured with retrieval in mind, because until recently there was no particular reason to structure it that way.
Where to start
For a small studio thinking about this, the most practical approach is to pick the single most repetitive operational task in the practice, the one that takes the most hours each week and follows a recognisable pattern, and investigate whether an AI tool already exists for it. For most practices that will be either proposal preparation or document processing, because those are the two areas where the repetition is highest and the existing tools are most developed. Running a focused trial on one specific workflow, rather than trying to adopt AI across the practice all at once, tends to produce clear enough results within a few weeks to inform where to go next.
