AI automation for architecture and engineering consultancies

Which work AI can take over in an architecture or engineering consultancy

Most of the admin inside a design practice is repetitive and rule-driven. Someone raises a question, it has to reach the right discipline, a reply is due on a date, and the answer has to be recorded against the right drawing. Multiply that across fifteen live projects and the office spends a large share of its week on logging, routing and chasing rather than on design. That is the work an AI worker can run.

An AI worker is software that reads incoming emails and forms, writes records into your project system, sends reminders on a schedule and produces reports. It does not decide the design. It does not decide whether a structural answer is adequate. What it removes is the clerical layer around those decisions: the logging, the matching, the chasing and the compiling.

For the office team, the practical change is that coordinators and document controllers spend their time on exceptions. A reply that contradicts an earlier drawing, a consultant who has missed three deadlines, or a client approval that arrived by phone all still need a person. The routine items, which are most of the volume, no longer do.

How big is the shift? The American Institute of Architects reported in its March 2025 study that 8% of firm leaders said they had integrated AI into their practice, while another 20% were in the process and 35% were considering it. The Royal Institute of British Architects found in its 2025 AI report that 59% of practices used AI for at least the occasional project, up from 41% in 2024. Adoption is rising, but most firms are still early. That gap is an opportunity for a firm that organises its admin first.

The whole workflow, stage by stage

The chain below follows a typical project from the first enquiry to archive. For each stage, the first paragraph describes what staff do now. The second describes what an AI worker takes over.

Enquiry and bid

Today a principal or bid coordinator logs the enquiry in a spreadsheet, looks up similar past jobs, and writes a fee proposal from an old template. Follow-up calls and emails are tracked in someone's inbox. An AI worker records each enquiry in the firm's CRM, pulls the fee history of comparable projects, drafts the proposal from the approved template and schedules the follow-up. The principal still sets the fee and decides whether to bid.

Appointment and project set-up

Once an appointment is signed, a project coordinator creates the folder structure in the common data environment, builds the drawing register, enters the consultant contact list and sets the numbering rules. This takes a day or more per project and is error-prone when the numbering differs from the firm's standard. An AI worker builds the folders and the register from the appointment letter and the firm's template, then flags any drawing number that breaks the agreed convention.

Design stages and drawing revisions

Architects and technicians produce drawings. Document controllers check that each sheet carries the right revision letter, that issue sheets match what was sent, and that transmittals list everything. An AI worker compares the drawing register against the files in the project system, lists missing or out-of-sequence sheets, and drafts the transmittal list for review. The person who issues the package still signs it off.

Consultant coordination and RFIs

This is the stage where most time is lost. A contractor or consultant raises a question by email. Someone logs it, decides which discipline should answer, pastes the reference into the RFI log, emails the right engineer, and then chases the reply. Reminders happen when someone remembers. An AI worker logs each RFI from the email or form, links it to the drawing reference, routes it by discipline (architect, structural engineer, mechanical and electrical consultant), records who has the next action, and sends reminders on the dates you set. Platforms built for this work already do parts of it. Buildend, for example, routes RFIs to a reviewer type automatically, and Autodesk Construction Cloud connects RFIs with drawings and models. An AI worker differs in that it can sit across your existing email, spreadsheets and project system rather than requiring everyone to move into one platform.

Client approvals and deliverables

Client approvals often arrive as a reply buried in a long thread. Staff copy the decision into the log, confirm the date and send a summary. Consultant deliverables, such as structural calculations or energy reports, are tracked against the programme in a separate spreadsheet that nobody updates. An AI worker detects approval language in replies, asks the project lead to confirm it, records the approval with its date and the document it refers to, and marks each deliverable late or on track against the programme.

Site visits and contract administration

After construction starts, site observation notes, site instructions and inspection reports pile up. Staff type up the notes, check them against the contract and file them. An AI worker turns the site notes into a structured log, flags instructions that may change cost or time, and drafts the summary for the project lead. Certificates and any decision that changes the contract stay with the registered professional named in the appointment. The AI worker prepares the paperwork; it does not certify.

Fee tracking, timesheets and invoicing

Project accountants match timesheets to phases, check them against the fee budget and prepare invoices. Overrun warnings often come too late. An AI worker compares logged hours with the phase budget each week, drafts invoices from the agreed billing schedule for review, and reminds the project lead when a phase is close to its fee limit. Invoice approval and any write-off stay with a person.

Reporting and close-out

Project leads currently ask for status updates in person or by email on Friday afternoon. An AI worker compiles the weekly status note from the RFI log, the drawing register, approvals and deliverables, and sends it to each project lead. At close-out, it checks that the operation and maintenance information is complete, lists missing items and files the final package.

Staff today and with AI workers

The table below lists the office roles that architecture and engineering consultancies usually employ, with the job titles they commonly use. It shows what each role does now, what the AI worker takes over and what remains with a person.

The highest-value automations, ranked by effort against impact

Not every automation is worth building first. The table ranks eight candidates by how much effort they take to set up against how much office time they recover. Effort assumes your project system and email are already in use. If your drawing register lives in a spreadsheet on one laptop, add a clean-up step before anything else.

Start with rank 1 and rank 2 together. They use the same data, and the status note becomes a by-product of the RFI log. Leave site logs until the rest runs cleanly, because they involve the most judgement.

Compliance and regulatory points

The most relevant standard for information handling in this sector is ISO 19650. It is a family of international standards for managing information across the life of a built asset using building information modelling (BIM). Its six parts cover concepts and principles, delivery, operation, information exchange, security-minded management and health and safety information. ISO 19650 is voluntary, which means it is not law, but clients often require it in the appointment. It also does not prescribe software. Your RFI and drawing workflow must therefore fit the information requirements in each appointment, and an AI worker should write its records into the common data environment the appointment names, not into a private folder.

Data protection is the second point. In the European Union, Article 22 of the GDPR restricts decisions based solely on automated processing when they have legal or similarly significant effects on a person. An RFI logger does not make such decisions. A tool that automatically scores a contractor's performance, withholds payment or flags a named individual for dismissal would. Keep those decisions with a person. The UK GDPR has a parallel provision. Client drawings and personal data in correspondence also need to sit in systems covered by your data processing agreements.

The EU AI Act, Regulation (EU) 2024/1689, is the third point. Its prohibitions and AI literacy duties have applied since 2 February 2025. Sources disagree about when the stricter high-risk obligations begin. One tracker gives 2 August 2026 for most high-risk uses, while another, citing the Digital Omnibus amendment, gives 2 December 2027 for standalone high-risk uses. In my reading, an RFI or status-note worker does not fall into a high-risk category, but a firm should confirm this with counsel before it deploys AI in any hiring, staff-evaluation or credit-related role.

How the transition happens without disruption

The safe route is staged. In the first two weeks, the existing team keeps working exactly as before, and the AI worker runs alongside them. It logs the RFIs they already log, and the coordinator compares the two lists each morning. Errors show up quickly and cheaply.

In the next phase, the AI worker takes over logging and reminders for one or two live projects. Staff handle every item the worker flags as unclear: an RFI with no drawing reference, a reply that may be a decision, a consultant who has replied on a different thread. Track how many items reach a person each week. That number falls as the rules are tuned.

After about a quarter, the roles shrink to review. The project coordinator spends the first hour of each day checking exceptions, approving reminders and sending the status note. Nobody is given notice on the basis of the AI worker's output. Where a role is no longer needed, the decision is taken through your normal HR process and does not depend on the tool.

Common mistakes

The first mistake is automating a messy process. If three people name drawings three different ways, the AI worker will log the mess faster. Fix the drawing numbering convention and the register before you connect anything.

The second is letting the AI worker reply to clients or consultants without review. Early on, every outgoing message should go through a person. Relax that only for routine reminders with the same wording.

The third is ignoring the appointment. Each client may ask for a different information structure, naming rule or approval route. Build those into the rules for that project, not into one global template.

The fourth is measuring the wrong thing. Counting emails sent tells you little. Count RFIs closed on time, the number of drawing errors caught before issue, and how many exceptions reach a person each week.

The fifth is expecting the AI worker to know the design. It can tell you a reply is overdue. It cannot tell you whether the answer is structurally sound. Keep the design judgement with the people who hold the professional responsibility.