Which task would you most like to be rid of tomorrow?

The employees doing the work can show you a specific task from start to finish. That gives you a starting point. Where do they search for information? What do they retype every time? Who makes the final corrections?

A short list of possible tasks is enough to begin with. Each task needs a description that lets a colleague see where it starts and ends. A draft reply to a particular type of question is easier to test than the whole customer service operation.

The result must be possible to check

These questions help you narrow down the tasks. An unclear answer needs resolving before the trial starts.

• Can you find examples that reflect everyday work and may be used in the trial?
• Do you know whether customer data, personal data and prices may be shared with the AI tool you want to try?
• Can an employee who knows the task check the result?
• Can you catch and correct mistakes before they affect a customer or colleague?

If the task needs information you cannot yet access, it is not ready for a first trial.

Which task is worth a trial?

You can compare the remaining tasks by how often they happen and how much working time they take. Keep working time and waiting time separate. A faster draft, for example, will not shorten the wait for a delivery date from the warehouse.

The aim needs to be concrete too. It might be less time spent writing a reply or fewer missing details in a handover. Both give you something to check.

Count the time it takes to read and correct the draft too. A quick draft only helps if the whole task improves once corrections are included.

An example of the decision

Invented example: A company considers both letting AI send customer replies and letting AI draft replies to questions about delivery.

The company chooses the drafts. An employee has access to current delivery information and reads each reply before sending it. The trial will test whether the employee can finish the task faster without getting more of the facts wrong.

Automatic sending would need a different trial. The company does not need to decide on that to find out whether drafts help.

The decision in a short message

The decision should fit in a short message: Which task are you testing? Who is responsible? What should improve? Which mistakes would mean stopping the trial?

The first task needs a possible benefit you can examine properly. The other ideas can wait, with a short note on what is missing.