Preparation, the week before: environment built, one real product configured end to end, stock on hand, cost setup done. And a rehearsal run, alone, so that nothing fails live. A pilot that breaks on screen loses more credibility than not running one. You are inviting the plant to watch the new system work; if it does not work, you have not run a neutral session — you have actively damaged the project's credibility, and you will not get a second one.
Who attends: an operator, a supervisor, the materials manager, the engineer who owns the BOM, and the controller — the controller especially, because costing objections found on day two are cheap and found in UAT they are not.
Day one: walk the master data. Show the item, the structure, the work definition. Ask "is this how you actually make it?" repeatedly, and write down every "well, not quite." Every one — the "not quite"s are the entire product of day one, and they arrive in half-sentences that are easy to let pass.
Day two: execute. Create a work order, release it, issue material, complete operations, complete the work order, run the costing processes, show the cost. Then show the journal entries to the controller. That is the moment the finance half of the room engages.
Three facilitation rules, and the middle one is hard. Let the operator drive the screen for at least one transaction — not you narrating, them clicking; you will learn more in those two minutes than in the rest of the day. Write down every objection. Do not defend the design in the room. That is genuinely difficult, because some objections will be based on a misunderstanding and you will know the answer — defend it once and the objections stop coming, and the objections are what you are there for. And when something cannot be done, say so plainly rather than promising a workaround.
The output: a prioritised list of design changes, and a signed acknowledgement of what was demonstrated. That document is your protection for the rest of the project — the record of what the client saw working and agreed to, on a date, and the thing you reach for in month seven.
And the failure mode: running a pilot with your own demo data instead of the client's. It is faster, it is safer, nothing breaks — and it proves nothing, and everyone in the room knows it. The part number on the screen is the whole point.