How to Use AI to Organize Operations

Most operational problems do not start as problems. They start as a detail nobody wrote down.

A customer mentions a change on site. The foreman texts the office. Someone says “I’ll get that to you” in a meeting. Three weeks later the invoice is disputed, the change was never documented, and everyone remembers the conversation differently.

Ask any room where important information gets lost and you hear the same list: emails, texts, field notes, meetings. As President and COO of Uniti Med and GQR Healthcare and COO of Nebula, a big part of my job is making sure what gets decided actually gets done. This is the fifth area of my workshop Take Back Your Time with AI: turn scattered information into clear next steps.

Where does important information get lost?

In the gap between hearing something and writing it down. Information arrives in pieces, from different people, in different formats, and nobody’s job is to put the pieces together. AI is very good at exactly that. Give it the pieces and ask for three outputs: an action list, a change-order draft, and a follow-up email.

It is the same in every industry. A property management company gets maintenance notes from techs, emails from tenants and texts from owners. A medical practice has huddle notes, voicemails and portal messages. The format changes. The gap does not.

How do you turn meeting notes into an action list with AI?

Record the meeting, or have someone take rough notes, then run this prompt on the transcript or notes.

Turn these meeting notes into a clear action list.

FOR EACH ACTION, INCLUDE:
- Task
- Owner
- Due date
- Current status
- Information still needed

ALSO IDENTIFY:
- Decisions made
- Unanswered questions
- Risks or delays mentioned
- Items requiring customer follow-up

RULES:
- Do not assign an owner unless the meeting names one
- Do not create a deadline unless one was stated
- Mark missing information as NOT ASSIGNED
- Keep decisions separate from proposed ideas
- Do not invent commitments

MEETING NOTES:
[PASTE NOTES OR TRANSCRIPT]

The rules are the reason this prompt works. Left alone, AI will tidy up a meeting by assigning tasks to whoever seems likely and adding sensible-sounding deadlines. That is worse than no list, because it looks official. Marking missing owners and dates as NOT ASSIGNED does the opposite: it shows you, in black and white, which decisions the meeting never actually made.

Everyone leaves with the same next steps. The gaps are now visible before they become delays. Send the list to everyone who was in the room within the hour.

How do you turn field notes into a change-order draft?

Changes on a job are where money gets lost, usually because the record is incomplete by the time anyone writes it up. This prompt organizes the record while the details are fresh.

Using the approved field notes, photos, customer request, and contract information provided, create an INTERNAL CHANGE-ORDER DRAFT.

INCLUDE:
- Project and date
- Who requested the change
- Original scope
- Requested change and reason
- Labor, material, and equipment information provided
- Possible schedule impact
- Supporting photos or documents
- Information still needed
- Review and approval fields

RULES:
- Do not decide who is responsible
- Do not determine contract entitlement
- Do not calculate or invent price
- Do not promise a schedule
- Mark missing information as TO BE CONFIRMED
- Clearly separate facts from assumptions

Look at what the rules keep away from AI: deciding who is responsible, determining contract entitlement, pricing the change and promising a schedule. Those are business and legal decisions. The draft’s job is to organize the record before details are forgotten. The record is organized. The decisions still belong to the team.

A property manager can use the same structure for an owner approval request: what the tenant reported, what the technician found, photos, options and what still needs a quote. A creative agency can use it when a client asks for “one more round” that is not in the contract.

Can AI turn approved details into an invoice?

Yes. Once the work and the numbers are approved, give the tool your approved line items, customer details, payment terms and logo, and ask for a branded invoice as a PDF, with the same rule as the quote prompt: do not calculate, change or invent any amount. In the workshop, I show approved details becoming a professional PDF invoice in minutes in Claude. Someone still reviews it before it is sent.

How does this example work, start to finish?

What does a person still need to check?

Where should you start this week?

Pick your most important recurring meeting. Record it or take notes, run the action-list prompt, and send the result to the room the same day. Count the items marked NOT ASSIGNED. That number tells you how much of your operation currently runs on memory.

The last area takes the next step: capturing the knowledge that only lives in one person’s head, in How to Create SOPs with AI. If your team struggles with follow-through after meetings, my piece on turning vision into action covers the leadership side.