Clinical AI · Team exercise
Before you automate a task with AI
Use a short team discussion to decide whether an AI tool has a clear, safe job to do—and what still needs a person.
What you’ll have at the end
A one-page task description, risks to check and a decision about the next step.
How to begin
- Describe the task without naming a product. Who does it, what starts it and what should be different when it is finished?
- Walk through ordinary and awkward examples with the people who do the work. List missing information, exceptions and decisions that need clinical judgement.
- Agree what the tool could produce and who must review it. Decide what happens if the output is wrong, incomplete, delayed or unavailable.
- Check data handling, clinical safety, accessibility, procurement and governance with the people responsible in your organisation before testing a tool.
- If the idea is approved for exploration, use fictional or approved test data first. Compare the output with a person’s work and record errors, time spent checking and unexpected effects.
What to measure
- Outcome: did the task become more useful or reliable for the person receiving the result?
- Process: how often the output needed correction or could not be used.
- Balancing: checking time, new work, safety concerns and effects on access.
Before using real data
This guide is an early discussion exercise, not approval to procure, deploy or use an AI system in care. Follow NHS clinical safety standards and local governance before any real-world testing.
Agree scope, data handling, supervision and escalation with your practice. Use approved systems. Do not upload patient information to this website or a public AI tool.