For anyone sitting through an AI demo this week who wants to see an ordinary Tuesday, not the vendor's best day.
When to use it
A vendor booked a demo and you have twenty minutes to prepare.
The demo already happened, everything looked flawless, and something tells you questions were missing.
You were asked for an opinion on a tool you have only seen in a presentation.
The sheet
Before the questions, the move: bring your real file. One from last week, with messy data and missing pieces, like yours always are. Ask them to run it right there. If they agree, you learn more in five minutes than in the whole presentation. If they refuse, you learned something too.
#
Question
What to listen for
Warning sign
1
What does it cost me at my volume in month six?
A figure calculated with your real volume, not the list price.
"It depends on usage", with no number at all.
2
When it gets something wrong, who will notice?
A review step with a named person, on your side or theirs.
"The model hardly ever gets it wrong."
3
How do I get out if it does not work?
Exit notice, what happens to your data and what leaving costs.
Annual contract with no exit clause.
4
Where does my data end up?
Country, cloud provider and whether it is used to train models.
"The technical team can cover that later."
5
Who provides support, and how fast?
A channel, hours and a written response time.
A generic inbox and "as soon as possible".
Real file I brought: ____________ · Did they run it? Yes / No · Weakest answer: question no. ___
How to fill it in 5 minutes
Pick the file the day before. Make it last week's and let it carry the usual problems: empty cells, mixed formats, duplicate names.
Print the table or keep it open. The five questions go in order; the first three weigh most.
Ask question 1 with your real volume in hand. The first month is always cheap.
For question 2, ask for a name. If nobody reviews, the error reaches the client.
Write down the literal answer, not your impression. "We can cover that later" is also an answer, and it counts.
The most common mistake: letting the demo set the pace. If they walk you screen by screen, the list never gets asked and you leave with the same information you came in with.
The line that is almost always missing
The real file. Almost everyone arrives at a demo with questions and nobody arrives with data. A demo runs on clean information prepared for that day; your operation does not. Asking them to run your file turns the presentation into a test, and the vendor's reaction to that request is already part of the answer.
Vic Sol · Decision Desk · vicsol.org/resources/ · Free to use with attribution. This sheet structures a conversation; it is not legal, financial or audit advice.
Open the .docx in Word (or the .xlsx in Excel) or print it before the meeting.
Fill the fields with the team; leave the blanks visible, they are the discussion.
Close with one owner, one date and one next action. Nothing else.
Prompt
Generate this sheet for your case
Paste this prompt into ChatGPT, Claude or Gemini together with your proposal, process or policy. It reproduces the sheet with your data and flags what is missing.
Act as an operations director with 15 years of experience who has sat through dozens of software demos and who judges AI tools by how they behave on an ordinary Tuesday, not on their best day.
Context I am giving you:
- Company and the area that would use the tool: [e.g. collections team, 200-person company]
- What the tool does according to the vendor: [ ]
- Real monthly volume we would process: [e.g. 1,200 requests]
- Type of file or data we work with: [e.g. spreadsheet with empty fields and mixed formats]
- Vendor proposal or product sheet, if I have one: [paste text or write "none"]
What I need: prepare me for the demo with exactly this structure:
1. Which real file I should bring and why: describe what it should contain (messy data, missing fields) and how to ask them to run it on the spot.
2. The five questions, in this order, adapted to my case: cost at my volume in month six; who notices when it gets something wrong; how I get out if it does not work; where my data ends up; who provides support and how fast.
3. For each question: what answer would be acceptable and what answer is a warning sign.
4. One follow-up question for each typical dodge ("it depends", "we can cover that later", "it hardly ever fails").
Output format: a table with the columns Question · Acceptable answer · Warning sign · Follow-up, preceded by a short paragraph about the real file. Maximum 400 words. No introduction.
Rules:
- Do not invent figures or vendor names. If data is missing, write [DATA PENDING] and tell me how to get it.
- Flag every assumption as an assumption.
- If critical information is missing, ask me at most 3 questions before answering.
- Answer in plain, direct business English, with no sales adjectives.
What to check in the answer:
Does question 1 use your real volume, or did it slip back to list price?
Does question 2 demand a named person, or does it settle for "the system alerts you"?
Does the real-file paragraph describe a file like yours, problems included, or a clean one that makes the vendor's day easy?
Vic Sol