Running your first live session

The default first move in an enablement role is a big session: everyone invited, an overview of what AI can do, some impressive demos, questions at the end. Attendance is high and the feedback forms are kind. A month later, almost nothing has changed, because people learned what the tool is, not how it fits into the specific thing they do on a Tuesday afternoon.

A live session that works is smaller and more specific. One team, one workflow, their own material, forty-five minutes, done together on screen. This playbook is the preparation, the plan and the follow-up.

The week before: prepare with the team, not for them

Pick the workflow and the team

Use a workflow you've already chosen and baselined (see Picking your first three workflows). Choose a team whose manager wants this. If the manager isn't attending, think hard about whether to run it yet: people take their cue from whoever decides what they work on.

Keep the group small enough that people can talk. Somewhere between five and fifteen is comfortable. If you have forty people who do the same workflow, run it three times.

Get real material

Ask the manager or a team member for two or three genuine examples of the work: last month's notes that became a report, a real customer email thread, the spreadsheet behind a recent summary. Check them against your usage policy. If they can't go into the tool as they are, either redact them together with the team or pick different examples. Don't fall back to made-up examples unless you have to. Real material is the whole point.

Do the workflow yourself, twice

Before the session, do the workflow end to end with the tool, using their material. Then do it again from scratch. You're looking for:

• The prompt or instructions that work, and the ones that don't
• Where the output goes wrong: wrong tone, invented details, missing a step the team always does
• How long it actually takes you, including checking

Ask one person to be your co-pilot

Find someone in the team who does this work regularly and ask them to sit in the driving seat for the second half of the session. It shifts the room from "the AI person showing us" to "one of us doing it", which is a much more convincing demonstration.

The session: 45 minutes, minute by minute

1. 1

   Step 1: 0–5 minutes.

   Why we're here, and what's allowed.

2. 2

   Step 2: 5–20 minutes.

   You do it, live, with their material.

3. 3

   Step 3: 20–35 minutes.

   They do it, with your co-pilot driving.

4. 4

   Step 4: 35–42 minutes.

   What changes on Monday.

5. 5

   Step 5: 42–45 minutes.

   One question each.

0–5: why we're here, and what's allowed

Open with the workflow, not the technology. Something like:

We're here to look at how you do the monthly client report, because you told me it takes most of a morning. By the end, you'll have done one with AI and you'll have a guide to do the next one. First, two minutes on what you can and can't put in.

Then the two minutes: which tool, which account, what's fine for this workflow, what's not. Use their examples. This removes the biggest silent blocker before you start.

5–20: you do it, live, with their material

Share your screen and do the workflow start to finish. Talk while you work:

• Say what you're giving the tool and why. "I'm including last month's report as an example, because otherwise it'll guess the format."
• When output comes back, read it out and critique it. "This bit is good. This number isn't in the notes, so it's made it up. I'm going to delete it and tell it not to do that."
• Show the bad output on purpose. Either the one you kept or a fresh one. Then show how you fix it. People trust a demo more when they see it fail and recover than when everything works first time.

20–35: they do it, with your co-pilot driving

Hand over. Your co-pilot shares their screen and does a second example, while everyone else follows along on their own machines if they can. Your job now is to coach, not to drive:

• Ask the room what they'd check in the output before sending it.
• When someone says "it got that wrong", treat it as useful. "Good. What would you tell it?"
• Write every question on a visible list. That list is the first draft of the guide's FAQ.

35–42: what changes on Monday

Agree, specifically, what happens next:

• Who will do this workflow with AI this week, and on which real instance
• Where they'll ask questions (a chat channel, not your inbox)
• When you'll check in

Ask the manager to say this part if they will. It lands differently from them.

42–45: one question each

Go round quickly: "What's one thing you'd check every time?" It reinforces the habit that matters most, and tells you whether the session landed.

Scripts for the awkward moments

"It's quicker to just do it myself."

It might be, for you, for this one. Can we time the next real one both ways? If it's not faster or better, that's worth knowing and I'll drop it for this workflow.

"It made something up. How can we trust it?"

You can't trust it unchecked, and you shouldn't. The question is whether drafting plus checking beats writing from scratch. Let's look at what checking takes.

"Is this going to replace us?"

I can't speak for every decision the company will make. What I can say is that today's aim is giving you back the parts of this job you don't like. The judgement in that report is still yours, and the tool can't do it without you.

Senior person takes over with a tangent.

That's a good one. Can I note it for the follow-up? I want to make sure everyone gets hands-on before we run out of time.

The week after: follow up or it didn't happen

• Same day: send a one-page guide for the workflow. The instructions that worked, what to check in the output, what not to paste in, and the questions from the session with answers.
• Two or three days later: check in on the chat channel. Who's tried it on real work? What went wrong? Every question is a gap in the guide.
• End of the week: a fifteen-minute retro with the team. How long did it take compared with the baseline? What would they change? Update the guide.

Mistakes to skip

• Generic demos. Writing a poem or planning a holiday shows the tool works. It doesn't show it works on their job.
• Too many workflows in one session. One, done properly, beats four skimmed.
• Talking for forty minutes. If people haven't touched the tool by the halfway mark, the session is a webinar.
• Hiding failures. People will hit them on day one. Better they've seen you handle one.
• No follow-up. Without the guide and the check-in, most of the session is forgotten within a week.

At the end you should have

• One team that has done a real workflow with AI, together, on their own material
• A named person doing it for real in the next week
• A one-page guide, updated after the retro
• A list of questions that tells you what the next session and the next guide need
• A before-and-after estimate to add to your progress report

If you'd rather not build the slides from scratch, the free deck Prompting on real work follows this session structure and you can present it as it is. The Role Book has the full session script library, the guide template and the retro questions, plus how to hand sessions over to champions once you've run a few yourself.