What data does your first AI use case need?

1. Describe the task in one sentence

Define who will use the application and what work it should support. “We want to use AI” does not define a project scope. A more specific task is helping staff find approved information about a business process.

Also define which decisions remain with people and which information the application must not use.

2. Identify relevant sources

Create a small inventory for this particular task. Consider availability, freshness and business ownership.

• Which information is required?
• Where does it live, and in which format?
• Who may read or change it?
• Who can explain its meaning?

3. Review quality using examples

Select examples from actual workflows. Check completeness, inconsistencies and freshness. Time-based data needs a clear reference to a point or period in time.

Include difficult and incomplete cases. A carefully prepared showcase does not explain how an application behaves in daily use.

4. Plan access and usage together

Technical access does not mean that every user should see every piece of information. Permissions must be respected in processing and output.

Clarify who updates the content and what should happen when information changes or is removed.

5. Define evaluation before the pilot

A pilot needs questions or tasks whose outputs can be evaluated by the business. Include ambiguous situations and incorrect answers.

• What counts as correct and useful?
• How will errors be identified and recorded?
• When should the application ask for clarification or hand off to a person?
• What would rule out introduction?

A useful next step

Summarize the task, sources, owners and open questions on one page. This supports a decision about starting a pilot or addressing a specific data integration first.