AI for Architecture Practice Management
A sober look at what AI does well in running a practice today, where it stops and how to try it safely.
AI can now take real admin off a practice: drafting timesheets, pulling past fees, flagging overruns and filing email. It cannot judge a fee or manage a client. This guide sets out both, the data questions to ask a vendor and a simple way to pilot it.
In practice management, AI is most useful when it prepares the draft or the flag and a person makes the decision.
Where AI genuinely helps a practice today
AI is useful in architecture practice management where the work is repetitive, the information already exists in the practice's records and a person checks the result. Drafting timesheets, pulling past fees, flagging overruns and filing email all fit that description. Setting fees and managing clients do not.
Most of the admin in a practice is moving information from where it arises to where it is needed: from a day of drawing work into a timesheet, from last year's projects into a new fee proposal, from an email into a register. That is the kind of task current AI handles well, because the source material is there and the output is easy for a person to check.
The practical test for any AI feature is simple. Does it save time on a task people already do, and can someone check the result in a few seconds? If both answers are yes, it is worth trying. If the output needs as long to check as it would to do, it is not yet worth it.
Five practical uses
These five uses are available in practice management software now. Each one drafts, suggests or flags; none of them should act without a person confirming.
Suggest timesheet entries from activity
AI can propose a day's timesheet from the documents, drawings, tasks and meetings a person worked on, coded to project and stage. The person confirms or edits. This tends to improve accuracy as well as speed, because the suggestion is based on what happened rather than a Friday afternoon reconstruction. See how to track time against RIBA stages.
Draft fee proposals from past projects
Given a new enquiry, AI can find similar past projects and show the hours each stage actually took, as a starting point for the new fee. The director still decides the fee. Our guide to calculating architects' fees covers the method.
Flag fee burn and overdue invoices
Rules have always been able to flag a stage over budget. AI adds the ability to explain the flag in plain language and draft the chaser or the client note, ready for a person to review and send.
Answer questions across project records
Questions such as "when did the client approve the façade change?" or "which RFIs are still open on Elm Court?" can be answered from the project's own records, with links back to the source so the answer can be checked.
File email into registers
AI can read project email and propose where each item belongs: a client instruction, a planning condition, an RFI, a decision. Someone confirms, and the register stays current without manual copying.
What AI cannot do
AI cannot take responsibility for a judgement. Fees, client relationships, staffing decisions and professional advice depend on context and accountability that sit with people in the practice.
- 01
Prepare
Timesheet drafts, email filing, data look-ups.
- AI role
- Does the first pass
- Person
- Confirms in seconds
- 02
Flag
Fee burn, overdue invoices, missing timesheets.
- AI role
- Spots and explains the pattern
- Person
- Decides whether it matters
- 03
Draft
Fee proposal starting points, client notes, chasers.
- AI role
- Assembles evidence and wording
- Person
- Edits, approves and sends
- 04
Decide
Fee levels, client conversations, hiring, advice.
- AI role
- Supplies information only
- Person
- Owns the decision
AI is useful across the first three columns. The fourth stays with the practice.
A model can show that similar Stage 3 work took 450 hours on three past projects. It cannot know that this client is likely to bring repeat work, that the planning authority is slow or that the practice wants the job enough to accept a thinner margin. It can draft a polite chaser, but it should not decide when to call a client whose invoice is overdue for the second time. These remain decisions for directors and project leads.
AI output can also be confidently wrong. Answers drawn from project records should link to their source, and any figure that will go to a client should be checked by a person first.
Data and privacy questions to ask a vendor
Practice data includes client information, fees and staff records. Before switching on any AI feature, get clear written answers to a short list of questions.
Ask every vendor
- Hosting: Where is our data stored and processed, including by any AI model provider? Is it in the UK?
- Training: Is our data used to train the vendor's or anyone else's AI models? Can we see that in the contract?
- Sub-processors: Which third parties process our data, and where are they?
- Access: Can AI answers include information a user is not otherwise permitted to see?
- Sources: Do answers link back to the records they came from?
- Control: Can we switch individual AI features off, and what happens to our data if we leave?
- Agreement: Is a UK GDPR data processing agreement standard?
The Information Commissioner's Office publishes guidance on AI and data protection, which is a useful reference for the practice's own checks. If client contracts restrict how project information may be processed, check those terms too.
How to pilot AI in a practice
Start with one task, a small group and a measure agreed before you begin. A four-week pilot is long enough to see whether it saves time and short enough to stop if it does not.
- 1
Choose
Pick one task, such as timesheet suggestions. Record how long it takes today.
- 2
Check
Confirm data answers in writing. Brief a team of three to six people.
- 3
Run
Use it on live work. Note every correction and every miss.
- 4
Decide
Compare time and accuracy with the baseline. Extend, adjust or stop.
Without a before figure, a pilot ends in opinions. Record the time a task takes before you start.
- Pick a low-risk task first.Timesheet suggestions or email filing, where every output is checked anyway.
- Include a sceptic.Someone who will report honestly when the tool gets it wrong.
- Count corrections.The share of suggestions accepted without edits is a good simple measure.
- Write down the rules.What may be used with AI, and what must always be checked before it leaves the practice.
Questions practices ask
Short answers to the AI questions UK practices ask most often.
How can architects use AI for practice management?
The most practical uses today are suggesting timesheet entries from project activity, drafting fee proposals from past projects, flagging fee burn and overdue invoices, answering questions across project records and filing email into registers. In each case a person checks and approves the result.
Can AI fill in timesheets for architects?
AI can suggest timesheet entries from the documents, tasks and meetings a person worked on, coded to project and stage. The person should still confirm each day's entries, because they are evidence for invoices and fee reviews.
Can AI set our fees?
No. AI can show what similar stages took on past projects and assemble a draft, but the fee depends on risk, client and commercial judgement that the practice owns.
Is it safe to put client data into AI tools?
It depends on the tool and the information. Check where data is hosted and processed, whether it is used to train models, which subprocessors are involved, and the applicable contract terms. Before using client information, check the practice's client commitments and agree an approved process with the people responsible for data protection.
How should a practice start with AI?
Run a four-week pilot on one low-risk task with a small team, measured against how long the task took before. Extend it only if it saves time without adding errors.
Where Avoice fits
Avoice uses AI inside the same project records the practice already works in, so suggestions and answers come from the practice's own data and are checked by the people who own it.
In Avoice, timesheets suggest hours from document and project activity for each person to confirm. The Email Agent captures project email into registers, the Project Record keeps documents and decisions, and the AI side panel answers questions across the project. Deals in Clients and CRM are priced by stage, team and hours, with the director setting the fee. For current information about hosting, subprocessors, data processing and model training, ask Avoice for its applicable terms and documentation before using the features with practice or client data.
To see these features on a live project, book a demo.