Treadstone Associates
Article · 6 min read

How to tell a real AI use case from a polished demo

A short set of questions that tend to separate the AI tools worth paying for from the ones that only work in the sales call.

Treadstone Associates · Updated 2026

Key takeaways

  • • A demo is built to look good in fifteen minutes; your actual workflow was never built to be watched.
  • • Ask what happens with messy, real input, not just the vendor's clean sample data.
  • • A real use case survives being run by the person who will actually use it, not just by the vendor.
  • • If nobody can name the specific task it replaces, it is not a use case yet, it is a feature.

A demo is optimized for the demo

Every vendor demo is, understandably, built around the inputs that make the product look its best. That is not dishonest, it is how a fifteen-minute pitch has to work, but it means a smooth demo tells you almost nothing about how the tool behaves on the version of the task your business actually has.

The useful question is not “did it work in the demo,” it is “what would it do with the file, the email, or the record we got last Tuesday, the messy one, not the tidy one.”

Ask about the messy input, not the clean one

Bring your own example, ideally an ordinary one rather than your worst case, and ask the vendor to run it live. Watch what happens when a field is missing, a name is spelled two different ways, or the format is not quite what the tool expects.

A tool worth adopting degrades gracefully, flags what it is unsure about, and hands the odd case to a person. A tool that only performs on curated examples is not ready for your workflow yet, whatever the demo suggested.

Hand it to the person who will actually use it

Before committing to anything, put the tool in front of the actual employee who would use it day to day, not just the manager who evaluated it. That person will notice, within minutes, whether it fits how the work really gets done, in a way a sales call never surfaces.

If the person doing the work says it would slow them down or that they would not trust its output without redoing it themselves, that is a more reliable signal than any feature list.

If you cannot name the task, it is not a use case

A genuine use case can be stated as one sentence naming a specific, recurring task, and the person or step it currently sits with. “AI for customer service” is not that sentence; “drafting the first reply to a routine billing enquiry so a person only has to review and send” is.

If the pitch cannot be narrowed to that level of specificity, that is usually a sign the use case has not been found yet, even if the underlying technology is impressive on its own.

Bring us a demo you are not sure about.

A 30-minute call is enough to tell you whether it is a use case or a pitch.