An AI idea becomes easier to assess when it is connected to a specific task. Instead of beginning with a broad promise to automate a department, describe the information a person uses, the assistance they need, and how they would judge whether the output is useful.
Choose a bounded task
A task such as finding relevant material, extracting a few fields, or drafting a summary has a clearer boundary than a general request for intelligence. Describe the inputs, the expected output, and the person who will use it.
Check the information
Identify the sources that may be used and the permissions around them. Information can be incomplete, inconsistent, outdated, or unavailable through an application interface. Those conditions belong in the assessment rather than being postponed until development.
Collect representative examples
Use examples that reflect the work, including ambiguous and incomplete cases. Agree on what a useful answer would contain, when it should refer back to a source, and when it should ask for a person to review the result.
Plan the surrounding workflow
An AI feature still needs an interface, access controls, integration, feedback, and an owner. Decide where assistance belongs in the existing process and what happens when a result cannot be used.
Expand from what is learned
A focused evaluation should inform the next decision. The result may support a wider application, suggest changes to the data, or show that a simpler approach would fit the task better. Keep room for each of those outcomes.
Bring this to the conversation.
Use this checklist to gather useful context.