I built an AI-assisted meeting follow-through system using Make, OpenAI and Google Workspace

[Kanika]

I would like to look at something that is frequently ignored in the automation of meetings. It is useful to record what has been discussed, but how can we ensure that the decisions actually result in action?

I created a workflow that uses AI to link together the notes from meetings with action tracking, deadline monitoring, and change management.

How it works

1. Meeting Notes Intake: Meeting notes are entered via a Google Form.

2. AI-assisted extraction: OpenAI looks over the notes to detect the action items that have been suggested and arranges them in a uniform format.

3. Human review: The proposed actions are placed in a Review Queue so that a person can check and approve them before they are included in the Action Register.

4. Deadline monitoring: Keep an eye on the deadlines that are coming up and prepare draft emails to be sent via Gmail for review.

5. Change Log: The updates relating to the tracked actions are kept so that any changes made to the original plan can be seen.

6. Follow-up dashboard: This provides a central overview so that you can keep track of the actions still needed, the deadlines, and the progress.

   
   
   
   

One aspect that I especially wanted to look at was the Change Log.

The plans for a project rarely remain precisely the same as those agreed upon in the initial meeting, for priorities may change, deadlines may be altered, and the actions may be modified.

It is more difficult to tell whether the present action plan still agrees with the original decisions if there is no record of the changes.

The Change Log offers insight into the way the plan develops. It enables people to see what changes have been made and to decide whether the current commitments are still in line with the agreed direction. It does not automatically make that decision. The aim is to provide people with a clearer record based on which they can work.

A few design decisions behind the workflow:

The AI suggests courses of action, but a person has to approve them.

• Emails sent as follow-ups are prepared in draft form and are not automatically sent.
• Changes are kept on record rather than simply overwriting the original plan.
  The dashboard shows progress on various actions in one place.

I used Make, OpenAI, and Google Workspace, focusing on practical automation and human supervision rather than building another meeting summariser.

I’m attaching a brief walkthrough which shows how the workflow fits together.

I’d be interested to hear from other Makers:

What is your approach when it comes to altering action items or deadlines after a meeting? Do you keep a separate change history, or do you make the changes directly in the action tracker?

[Mahdi_Eqbal]

I’d keep the approved action as a separate record from the AI’s original suggestion, then treat later changes as revisions rather than overwriting it.

For each change, I’d capture the previous value, the new value, who approved it, when it took effect, and why it changed. That makes it much easier to distinguish a legitimate deadline extension from an overdue action that was quietly rescheduled.

One edge case worth handling is when meeting notes mention an action that’s already open. Matching it to the existing action before creating a new one could prevent duplicate commitments across recurring meetings.

[Stoyan_Vatov]

Why are you using polling triggers instead of instant ones? And why are you parsing the output of the OpenAI module separately? Its a built in function.

Both of these are wasting operations.

[Kanika]

Thanks for sharing this! I reviewed your suggestions against the workflow design, and quite a few of these controls are already part of it.

The Change Log is designed to capture the Action ID, change date, field being modified, previous value, new value, reason for the change, and the person making the update. The Action Update Form collects these details before Make updates the Action Register and creates a separate Change Log entry.

The intention is exactly what you described: retain a traceable history of changes rather than silently overwriting the original information. This makes it possible to understand how an action or deadline has evolved.

Your suggestions about explicitly capturing the approver and effective date are useful additions to consider. I also account for possible duplicate actions through AI-generated duplicate flags and human review, although automated matching against existing open actions is something that could be developed further.

Appreciate the thoughtful feedback!