Show who does what

Write down the steps from the start of the task to the point someone acts on the result. Show what AI does and what a person remains responsible for.

For a customer enquiry, AI could prepare a draft while a team member checks the facts and decides what to send. That person needs the original enquiry and supporting information alongside the draft.

Make someone responsible

Name someone to collect problems, keep instructions up to date and decide when a change needs another test. They do not have to do every task, but the team should know who to turn to.

Agree when they will review problems and whether the checks are still doing their job.

Further reading: NIST: AI Risk Management Framework Playbook

Plan for the awkward cases

Test missing information, conflicting instructions and requests the system is not designed to handle.

Make it easy to stop, ask for help or return to the existing process. An uncertain answer should not quietly become the next person’s problem.

Check whether it helps

Regularly review a small sample of real work. Check quality, corrections, time spent and cases that needed extra help. Ask the people using it what is working and what they have to fix.

Use that feedback to decide what to change. Expand only when the current process is working well. More users or more automation do not, on their own, mean better results.

Put it to work

Before rolling AI out further, agree who’s responsible, who checks the work and what happens when it fails.

  • Who receives the work, and what do they do next?
  • Which cases need a person’s help?
  • When will we check quality, time spent and whether people use it?

Ready to make your AI pilot useful?

Bring the workflow and what’s getting in the way to a free 30-minute chat. We’ll discuss ownership, checks and the next step towards everyday use.

Talk through your workflow