The Human Meeting Hack that Leveled Up the AI potential: Pt 1

One of my most favorite things is when AI gives me the biggest AHA moments like this one: Most of the time just THINKING about how to explain a process or workflow to AI, reveals the simplest human work hack.

Event / Source

Workflow discovery · Global enterprise team · September 2026

What it was about

A colleague at a global shoe brand was dealing with a recurring problem during the rollout of a new internal system.

Team members would say they didn’t know how to perform a task or had never been trained on a feature. But she remembered explaining many of those features in previous meetings. Why did this keep happening? Was this a human block or an information block?

The information existed. Finding it was the problem.

Our first solution didn’t require building an agent, connecting an API or writing a prompt. My colleague was spinning her wheels rarely making progress on the integration because of what looked like a potential resistance to accountabilty on behalf of the team or perhaps a fear of learning a new system, or true lack of knowledge.

Typical company methods for this brand are very similar to most companies when it comes to meetings: Set up a recurring ‘check in’ or ‘status’. Invite the team. Send an agenda. Sounds good. Until we consider how we are going to assess all the meetings in totality. Once we step back, all the meetings look the same. The transcript titles just have dates, the recaps are buried in emails. How do we reveal the trends?

So each subsequent check in was the same routine: The systems analyst would report on open tickets, employees would say they don’t know how to use the system, and would submit tickets without clear identification of the issues. Training would be scheduled again. Employees would attend and forget or skip all together. Rinse and repeat.

My colleague didn’t want to put anyone on the spot but the truth was, employees kept maintaining that, when confronted, they couldn’t complete a task because they didn’t know how. She had in fact given multiple trainings over the past months. The employees pretended they were not taught.

So we changed the meeting names.

Instead of generic titles, walkthroughs became things like:

System Walkthrough 2: Feature X

Now the meeting title itself describes what knowledge exists inside the transcript. Employees or their managers could search emails for that subject. Sharepoint drives could be catalogued by meeting category. This began the infrastructure needed for not only AI but the humans.

What I’m keeping

Sometimes the first step toward an AI-ready workflow isn’t adding AI.

It’s making the existing information easier for both humans and machines to understand.

Meeting titles, filenames, folders, project names, dates, categories and ownership fields may feel mundane. But they’re metadata. And metadata becomes increasingly valuable when AI systems need to retrieve the right information from hundreds or thousands of documents.

3–5 things I learned

• Good human organization becomes good AI context. Descriptive names and consistent categories make information easier to retrieve before AI ever touches it.

• Reduce the search space. If a question concerns Project X, the system shouldn’t need to search every transcript the company has ever created.

• Information architecture affects AI performance. Research on enterprise retrieval shows that metadata-enriched retrieval can outperform approaches relying only on document content.

• Fix repeated friction before automating it. If nobody knows where decisions, training or responsibilities live, automating the workflow may simply automate the confusion.

• Small structural changes can create accountability. “I think we discussed that somewhere” is very different from “See System Walkthrough 2: Feature X. You attended the session and the transcript is here.”

How this changes my work

I’m starting to see AI workflow design as something that begins before the AI.

First, look at how information moves through the organization.

What gets named?

Where does it live?

Who owns it?

Can someone find it again?

Can we tell a decision from a discussion?

Can we tell training from a general meeting?

Can we identify which project, client or system a document belongs to?

Only then ask what AI should do with it.

A better AI system may start with something as boring as a better meeting title.

** Stay tuned for more on this subject. I am obsessed with the superpower combination of AI + meeting transcripts. This is has been life changing.