Self-directed prototype using a simulated operations meeting. AI extracts draft records; people retain responsibility for confirming actions and authorising work. Test evidence does not establish production savings.

1. Business purpose

Turn meeting content into an actionable record without manually retyping decisions, tasks and approval requests.

The workflow separates three responsibilities: the Meeting Lead checks that the extracted record reflects the meeting; an Approver authorises work requiring further approval; the Action Owner carries out the work. One person may hold more than one role.

2. Data and processing approach

The Meeting Actions & Approvals Google Sheets workbook has three tabs:

A sample transcript represents notes becoming available. Automatic recording/transcription is upstream of this prototype.

AI returns a list of decisions and actions. The prompt excludes discussion without agreement, leaves unknown details blank and distinguishes preparing an authorised proposal from purchasing something that still needs approval.

3. Workflow structure

Zap 1: Meeting Notes → Decisions & Draft Actions

A new Meetings row triggers AI extraction. Looping processes each returned item; Paths routes decisions to the Decisions tab and actions to the Actions tab.

Draft actions are logged as Pending Review. The Meeting Lead checks descriptions, owners, deadlines and approval requirements before setting Confirmed.

Zap 2 : Confirmed Meeting Actions → Tasks & Approvals

An update to any column in Actions triggers a readiness filter. The row must be Confirmed, have an action, owner and due date, and have no Asana Task ID.

Two paths handle the row: