The quickest way to make an archive demonstration unhelpful is to begin with everything. Every folder, every format, every possible question—and no agreement about what success would look like.
A better starting point is a small question with a real consequence. “Can we find the interview where this topic was discussed?” is more useful than “Show us what AI can do.”
Choose a question people already ask
Ask the person who works with the collection which requests are difficult to answer. Perhaps names are spelled differently across records. Perhaps the filename gives no clue to the image. Perhaps someone remembers a sentence but not the recording.
Choose a task that occurs often enough to matter. Write down how it is handled today, including the manual checking that happens after a candidate result is found.
Bring the difficult examples too
A sample made only of clear scans and clean recordings tells you little about the rest of the archive. If you have permission to use them, include a few representative difficulties: a faint page, a revised copy, an unusual layout, an interrupted conversation or a file whose title is misleading.
Keep the sample manageable. Its purpose is to reveal the workflow, not to complete a migration during the first meeting. Sensitive or restricted material needs handling arrangements before it is shared.
Define a useful result
Success might mean retrieving a relevant record and opening the correct page. It might mean finding a passage and replaying enough audio to judge what was said. It might mean organizing likely image variants for a reviewer.
Do not reduce the evaluation to a stopwatch. A fast result that takes several minutes to verify can be less useful than a slower result with a clear source trail.
- State the task and the reviewer.
- Identify the permitted material.
- Agree on what the result must include.
- Record uncertainty and missed examples.
- Decide what evidence would justify a larger evaluation.
Leave with a next decision
The first demonstration should help you decide whether a scoped evaluation is worthwhile. It does not need to settle every deployment detail. It should make the remaining questions visible: formats, hardware, permissions, support and expected review effort.
For a VAIMS introduction, a description of your problem is enough. Our website form does not accept collection uploads. We can begin with a demonstration collection and discuss an approved sample separately.
