Wei Jie is a project coordinator at a 40-person M&E contractor. Every Tuesday, he takes notes at the site progress meeting. This week, the team agreed to move the project forward. Someone suggested Friday for a review. Finance promised to "check the budget".
By the next meeting, nobody is sure what was actually decided. Was Friday agreed? Who was meant to call the supplier?
AI can turn rough notes like Wei Jie's into tidy minutes and an action list quickly. The real skill is making sure it doesn't tidy away the uncertainty. Here's how.
I. Start with notes that are safe to use
Use meeting notes your company has approved for the AI tool. Practise with a few made-up notes first. If you plan to record meetings, tell attendees beforehand and agree who can access recordings and how long they're kept.
Start with one routine meeting type, not a backlog of confidential meetings. Number the paragraphs in your notes, so every decision and action can be traced back to its source.
II. Separate discussion, decisions and actions
Good minutes have four clear parts: what was discussed, what was decided, the actions to follow up and the questions still open.
Here's part of Wei Jie's notes:
"Operations will check the revised supplier quotation. Finance needs to confirm the budget. Someone suggested Friday as a possible review date, but no date was agreed."
A careful action list looks like this:
| Action | Owner | Due date | Note |
|---|---|---|---|
| Check the revised supplier quotation | Operations (name to confirm) | Not agreed | Confirm who is responsible |
| Confirm the available budget | Finance (name to confirm) | Not agreed | Ask the chair to assign |
Friday belongs under open questions. It shouldn't become a deadline just because the table looks neater with every box filled.
III. Use a prompt that keeps gaps visible
Here's a prompt Wei Jie could use:
"Draft meeting minutes using only the notes below. Separate discussion, decisions, actions and open questions. For each action, include the source paragraph, owner and due date only where stated. Otherwise write 'to confirm' or 'not agreed'. Don't turn suggestions into decisions. List anything unclear for the chair to review. Mark the document as a draft."
AI can sound confident and still add details that weren't in the notes, so treat its output as a first draft.
IV. Check before anything becomes a task
Wei Jie checks the draft against his notes. The chair confirms the decisions and assigns any unowned actions. Each owner confirms their tasks and dates.
Only then do the actions go into the team's task tool. Creating tasks straight from an unchecked summary can send the same mistake to several people.
Keep the approved version with who checked it and when. If a decision changes later, record the change rather than quietly editing the original.
V. Test what gets left out
Missing items matter as much as wrong ones. A polished summary can still drop the one open question holding the project up.
Try a short test set of notes that includes:
An action with a named owner and a clear deadline
A proposal that was discussed but not approved
An action with no owner
Two different dates for the same item
A decision that depends on someone else's approval
Decide the correct answer first, then compare the AI's draft. Note any missed actions, invented commitments, wrong owners and made-up dates. Include checking time when comparing it with how minutes are done today.
VI. Make it repeatable
Give someone ownership of the minutes template, and agree where approved minutes are stored. A colleague should be able to repeat the process, find the original notes and see why an item was flagged.
Choosing between several tasks to start with? Our guide to business processes to automate with AI can help.
If you'd like your admin team to build workflows like this with guidance, explore our AI for Administrative Teams programme or book a discovery call.
Wei Jie is an illustrative example based on a typical SME.