Choose by use case

How do you select an AI tool based on comparable tests rather than a general impression?

1250 € per session, excl. VAT

In short

This half-day course of 3 hours teaches participants how to choose between generative AI tools for everyday tasks. Participants define their criteria, run comparable tests and document the differences. They leave with an evidence-based selection matrix that distinguishes between response quality, correction effort and usage constraints.

Why this training?

A fluent response can conceal omissions, fabricated information or an unusable format. Comparing style alone leads to favouring an appealing tool rather than one suited to the task and the organisation’s constraints. A defensible choice rests on common inputs, explicit expectations and traceable observations. The course turns scattered tests into a decision that can be reviewed, separating observed results from advertised or unverified capabilities. Continuing case study: The fictitious organisation Avelune Documentation must choose between two authorised tools to summarise an internal note, draft an email and extract information from a file. Participants use entirely fictitious documents and recommend a choice for each use case, without seeking a universal winner. Transfer to the workplace: In the following week, reuse the matrix for an everyday task with fictitious or anonymised data and an authorised tool. Check whether the ranking remains relevant, then record the changes in context that would require a new test.

Objectives of this training

  • Describe three use cases with their input data and expected results.
  • Build an assessment grid with weighted criteria and exclusion conditions.
  • Run an identical protocol on two authorised tools.
  • Score the faithfulness, completeness and correction effort of the responses.
  • Justify a recommendation for each use case using the evidence gathered.

Who is it for and what are the prerequisites?

Roles and tasks concerned

  • Staff who use AI to draft, summarise or extract information.
  • Managers and designated contacts seeking to justify a tool choice for common use cases.

Prior skills

  • Have prior experience of making simple requests in a generative AI assistant.
  • Be able to identify sensitive information and consult internal usage rules.

What you will work on

  1. Define what needs to be compared

    Objective: Turn each task into a testable need. Inputs, expected format, unacceptable omissions and data sensitivity; the difference between personal preference and a verifiable requirement.

  2. Build a fair comparison

    Objective: Set the rules before seeing the responses. Weighted criteria, a common scoring scale, exclusion conditions; accessible settings and test conditions to record.

  3. Test and characterise differences

    Objective: Measure quality without confusing fluent writing with accuracy. Checking against source documents, identifying fabricated information, adherence to the format and counting the corrections needed; variability between responses.

  4. Decide with explicit caveats

    Objective: Formulate an actionable recommendation for each use case. Weighted calculation, quality–effort trade-off, absence of evidence and limitations of small samples; documentary information still to be verified.

What you take away

  • Scoping sheets for the three use cases.
  • Comparison protocol with requests and test conditions.
  • Weighted matrix with annotated responses.
  • Recommendation note for each use case and observed limitations.

Method and assessment

  • Short teaching inputs followed by immediate practice on the continuing case study.
  • Individual work followed by work in pairs, with peer review of outputs.
  • Guided group discussion: what works, what needs correcting and what remains a human decision.
  • Preparation: Check access to two authorised tools; if unavailable, provide a set of prepared and anonymised responses for comparison. Prepare three fictitious documents, their reference responses and a shared blank assessment grid.

Observable criteria: The same documents and initial instructions are used for each tool. ; Weightings are set before testing and add up to 100 %. ; Each quality score cites a passage or an observable error. ; Any exclusion condition takes precedence over the overall score. ; The recommendation distinguishes between observations, uncertainties and outstanding checks..

Frequently asked questions

Can a single tool be enough?

Yes, if testing shows that it meets the requirements of the selected use cases; this is not assumed.

Are costs compared?

Only if verified contractual terms are available; no pricing is assumed.

Can work-related documents be used?

Only if anonymised, in accordance with internal rules and within authorised tools; otherwise, use the fictitious case study.

Does the ranking remain valid over time?

It depends on the versions, access arrangements and tasks tested. A significant change warrants further testing.

Last updated: