Everything you run from the front of the room on Day 2 — Excel, PowerPoint, and Security & Ethics — then wrapping up. Each demo uses its own data and prompts, deliberately different from the trainee lab, so your live run never hands out the answers. Tap any prompt to copy it.
Differs from trainees Trainees analyse 4 branches over Jul–Dec 2025; you demo 5 branches over Jan–Jun 2026. Open the demo workbook and click into the table before you start.
The data has a planted story: Salmiya drops ~29% in March (branch renovation) then recovers by May; Fahaheel grows every month (Jan ≈ 4,076 → Jun ≈ 5,293, about +30% over the half). If Copilot names Salmiya + March as the sharpest drop, it read the table correctly. If it says anything else, open a PivotTable and show the room the truth.
Open the growth column's formula out loud: "could you defend this number to Finance?" That single move — reading the working — is the whole point of the Excel session. An answer you can't check is a spell, not analysis.
Differs from trainees Trainees restructure the Branch Network Review deck; you demo on a different mess — the Digital Channels deck — so your live fix doesn't hand out their answer. Doc→deck reuses the shared earnings release.
The demo deck has planted messes: the agenda slide is in the wrong place, the summary sits at the top instead of the end, and there's a duplicate outage slide. A correct restructure moves the agenda to slide 2, the summary to the end, and merges the outage duplicate. If Copilot misses the duplicate, catch it live — that's the teaching beat.
On Move 1, open one generated slide against the earnings PDF and check a number aloud. A deck inherits the document's facts only if a human verifies them. Never let a slick-looking slide ship an unchecked figure.
Differs from trainees Trainees fix an unsafe prompt in the lab; you run both prompts live so the room sees the boundary before they practise it. Run the safe one for real; read the unsafe one aloud and stop before you paste.
The unsafe prompt fails twice: it leaks real customer data and it lets a model shape a lending decision. Ask the room: "who is accountable if that letter is wrong or unfair?" The answer is always a human — that's the line Copilot never crosses.
Facilitator role This is a close, not a demo — you're steering the room through two links, not running a prompt. Put both on screen and give the room quiet time to do each one before you send them off.
Point the room to the Wrap up exercise in the Day 2 lab (or share the link). It's the same skills map they took on Day 1, so it measures the shift across the two days — frame it as a measure, not a test.
Once the assessment is in, move the room to the feedback survey — a few minutes while the two days are still fresh. Say plainly that it shapes the next cohort's workshop.
Land the two-day arc: Copilot went from an icon they scrolled past to a tool they can drive — bank-safe, with a human accountable. Confirm both the assessment and the survey are submitted before anyone leaves the room.