Put AI to work
with a clear purpose.
Start with the work you want to improve. Bring the people, data, infrastructure and operating consequences into the same decision.
Swipe or use arrow keys to explore the diagram.
Read the diagram explanation
Agree the outcome and accountable owner.
Check data, access and dependencies.
Define the scope and permitted actions.
Match evaluation and oversight to the consequence.
Keep a way to stop, recover or do the work another way.
Include human review, corrections, dependencies and fallback effort when judging the benefit. Increase oversight as the consequence of error increases.
The opportunity is broader
than a chatbot.
AI can help people find knowledge, analyse information, recognise patterns or support decisions. The useful question is where it improves the complete piece of work.
Start with an outcome
Identify a real source of delay, effort or poor decisions. Establish how the current process performs and what an acceptable improvement would look like.
Design for the people doing the work
Agree who uses the result, who checks it and who can challenge it. Include the effort needed to review and correct outputs when measuring the benefit.
Understand the information and dependencies
Check which data is appropriate to use, how access is controlled and which suppliers, systems and connections the workflow depends on.
Make ownership clear
Name the person accountable for the workflow and the people authorised to change, stop or retire it. Build skills as the way of working changes.
Quality and safety
follow the consequence.
A written summary and a decision affecting a live operation need different checks. Match the evaluation and human oversight to the task.
Make the pilot
small enough to learn.
Choose a bounded task with a useful outcome. Compare it with the current process, including cost, review effort, errors and the impact on people. Expand only when the evidence supports it.
Related guidance: NIST AI Risk Management Framework, a voluntary resource for considering trustworthiness across the AI lifecycle.
Part of the same system.
Connect the design decisions across these areas.
What work could be better?
Bring the task, the operating constraints and the result you would like to improve.