1

Task fit

Test the exact work: research, drafting, analysis, multilingual service, document handling, coding or tool use. Do not substitute a generic leaderboard.

2

Data boundary

Compare training use, retention, processing location, access, deletion and the controls available to your organisation.

3

Operational control

Review identity, permissions, logs, connectors, approvals, monitoring and incident support.

4

Evidence and effort

Record test evidence, review effort, integration burden, failure recovery and the cost of maintaining safe use.

FREQUENTLY ASKED

Two points worth making clear

Why are there no fixed scores here?+

Model features, prices and performance change quickly. A dated, task-based evaluation is more useful than a permanent-looking score.

Can one organisation use several assistants?+

Yes. Many teams route different data and tasks to different approved systems, provided the boundaries are clear.

NEXT STEP

Apply the framework to your actual task

Desk AI can help structure evaluation questions and controls. It does not make procurement decisions or retain chat data.