The note
AI works best when the job, data and owner are clear.
A new AI tool can make a weak process move faster. That does not make the process good.
Start with the job
- What starts the job?
- Who owns it?
- What data is needed?
- What choice gets made?
- What happens next?
- What can go wrong?
If the team gives five different answers, map the work first.
Check the data
AI needs useful input. Old, missing or mixed data will lead to weak output.
Decide which tool holds the main record. Do not let five tools each claim to be right.
Set a safe limit
Write down what AI may do on its own. Write down what needs a human check.
The higher the risk, the stronger the human check should be.
Run a small test
- Use a small set of real tasks
- Save the input and output
- Check errors by type
- Ask the team what slowed them down
- Fix the route
- Test again
Name the owner
A person must own the tool after launch. They should check errors, costs and changes.
AI is part of the work. It is not the owner of the work.