Practice position
AI assists. People decide.
Use approved systems, minimise information, check important output and keep accountable professional judgement human-led.
Define the theory of change
State how the tool is expected to change a workflow and how that change might produce a benefit. Identify assumptions: staff use it as intended, output is accurate enough, review time is available and people trust the process.
Choose balanced measures
Combine process measures with outcomes, error rates, staff time, rework, accessibility, service-user experience and unequal effects. Include balancing measures so an apparent saving in one team does not hide extra burden elsewhere.
Evaluate the whole system
Performance depends on prompts, source documents, permissions, training, local workflow and human review as well as the model. Record changes during the pilot and avoid attributing every difference to the technology.
Practical checklist
- Set a baseline
- Define intended and unintended outcomes
- Measure error and human-review burden
- Include staff and service-user experience
- Agree continue, change and stop criteria
Authoritative starting sources
NHS real-world AI evaluation lessons ↗NICE Evidence Standards Framework ↗Written by Tony D’Agostino / TD Consultancy. Reviewed . This page provides general professional education, not legal or clinical advice.