Lorixen.
EN

Client Meeting Follow-Ups for a Small Architecture Practice, Drafted by AI

Short answer: After a client meeting, an architecture practice can use AI to turn a recording or rough notes into three things: a short follow-up email, a log of decisions the client made, and a list of actions with owners and dates. A person reviews all three before anything is sent, because in design work a wrongly recorded decision can become a dispute about scope, cost or programme. Get consent to record, keep the decision log separate from the chatter, and never let the tool invent a commitment nobody voiced.

Why follow-ups matter more in design practice

A small practice runs many conversations that change a project: a client prefers the kitchen at the rear, the planner suggests a lower ridge, the structural engineer asks for a deeper beam. Those choices drift across emails, site notes and memory. Months later, someone says "we never agreed that", and nobody can show the moment.

A prompt, accurate follow-up sent within a day creates a shared record while memories are fresh. The difficulty is time. Principals are often back at a drawing board within minutes, and the notes wait until Friday, or forever. AI can produce a first draft in the time it takes to make a coffee.

Step by step

  1. Agree the ground rules. Tell attendees the meeting will be recorded and transcribed for notes, and why. In a face to face meeting say it aloud and note it in the agenda. For remote calls most platforms show a recording banner; still mention it. If anyone objects, take manual notes instead.
  2. Capture cleanly. A speakerphone in a meeting room works poorly. Use a decent conference microphone, or ask each person to speak one at a time when something important is decided. Say the important lines in full: "Decision: we will proceed with option B for the rear extension."
  3. Transcribe. Use a transcription service or the built-in feature of your meeting platform. Keep the transcript in the project folder, not on a personal device.
  4. Run a structured prompt. Ask the AI for specific outputs rather than a summary: decisions made, actions with named owner and due date, open questions, and items that were discussed but not decided. Tell it to quote the transcript for each decision and to write "not stated" where an owner or date is missing.
  5. Review against the transcript. A person reads each decision and checks the quote. Any item with no supporting quote is deleted or marked for clarification.
  6. Send the follow-up. The email lists decisions, actions and the next meeting date, and invites corrections within a set number of working days.
  7. File and update. Add confirmed decisions to the project's decision log, and actions to the task list the team already uses.

What the output should look like

A good follow-up has a fixed shape, so clients learn to scan it.

  • Decisions confirmed. One line each, numbered, with the date. Example wording: "Agreed to omit the rooflight over the stair."
  • Actions. Owner, task, due date. Practice actions and client actions are shown separately.
  • Information needed. Items such as survey data, product choices or consultant input.
  • Not yet decided. Topics raised that need another conversation.
  • Cost or programme impact. Only if someone stated it; otherwise left out, never estimated by the tool.

Common failure points and fixes

Problem Cause Fix
A decision appears that nobody made The model turned a suggestion into an agreement Require a quoted line for every decision and remove unsupported ones
Wrong person named for an action Similar first names or poor audio Keep an attendee list and tell the tool to use only those names
Technical terms garbled Transcription of trade vocabulary Provide a short glossary of project terms with the transcript
Client sees internal remarks Draft included side comments Ask for a client facing version and an internal version separately
Numbers are wrong Dimensions or fees misheard Never let figures through unchecked; confirm against drawings or the appointment letter

Boundaries that protect the practice

  • No scope or fee commitments from the tool. If a conversation touched additional services, a person writes that paragraph, ideally with reference to the appointment terms.
  • No legal or planning conclusions. The tool can record that the client "will speak to the neighbour", not that a boundary is agreed.
  • Keep confidential material under control. Client names, addresses and financial details are personal or commercial data. In Ireland the Data Protection Commission publishes guidance on processing personal data; as of November 2026 you should read how your chosen AI vendor stores recordings and whether it reuses them, and switch off model improvement where the plan lets you.
  • Consultants are third parties. If an engineer joins the call, ask for their consent to recording too.

A small illustrative example

This scenario is invented to show the shape. In a ninety minute call about a house renovation, the client moves between four topics. The raw transcript runs long. The structured prompt returns six decisions, nine actions and three open questions. On review, the principal removes one decision where the client had only said "that might work", and corrects a due date. The email goes out the same afternoon, and the client replies with a single clarification instead of a long phone call a week later.

Setting up the habit

The tool matters less than the routine. Pick one template for the prompt, one folder structure for transcripts and logs, and one person per project responsible for sending. Keep the prompt text in a shared document so the whole practice produces comparable records. After a month, check a handful of follow-ups against transcripts to see whether errors are creeping in.

FAQ

Do we have to record meetings to do this?

No. Typed rough notes pasted into the same structured prompt work, and sometimes are better because they carry your own emphasis. Recording just removes the reliance on memory.

Can the AI draft the email in our practice's tone?

Yes, if you give it two or three past follow-ups as examples. A person should still read the draft for tone and accuracy before sending.

Should the client see the full transcript?

Not by default. Share the follow-up and decision log. A transcript contains tangents and informal remarks that rarely help the client.

How long should we keep recordings?

Decide a retention period in your privacy policy and stick to it. Keeping confirmed decisions long term and deleting raw audio sooner is a common balance, but take advice for your own obligations.

If your practice wants a repeatable meeting follow-up workflow built into the tools you already use, you can describe your setup on our enquiry page and we will suggest a sensible starting point.

Sources

WhatsAppBook a call