How to choose your first AI task

Find a task you know well.

Talk to a colleague who regularly spends time reading, comparing or organising information. They might be reviewing supplier proposals or sorting customer enquiries by topic. You need access to the material and a clear idea of what a good result looks like.

Ask about the work: Which parts take time? Which information must always be included? If those questions are hard to answer, spend some time understanding the workflow before choosing where AI could help.

Write down three possible tasks and choose one together. Consider how often it comes up, what makes it difficult and who can check the result. Choose something that matters to your work, where any errors can be spotted and corrected before anyone uses the draft.

Try AI on part of the task.

Imagine you need to compare three supplier proposals. You have to understand the offers, clarify what is missing and eventually choose a supplier. In this fictional example, the experiment starts with gathering the information already in the proposals into one overview.

Agree on the columns before you begin: scope, price, timeline and conditions. Ask AI to show where each piece of information appears in the proposals and flag anything missing. This gives the team a draft to review. You still need to resolve the questions and choose the supplier yourselves.

Write down what you want to find out. For example: Can AI help us produce a comparison that takes less time to finish and is at least as useful as the one we make today?

Save the instructions you give AI. You will need them when you repeat the experiment.

Check what you are allowed to use.

Your company needs to have approved the tool for the task you want to do. You also need to know which information you may enter. Having a licence does not settle that. If you are unsure, ask the people responsible in your company before using company material.

You can start with fictional proposals where you know the content and what the result should contain. That lets you try out the workflow. When you move on to real documents, you will need to check how it handles differences in the material and the requirements that apply to it.

Choose someone with the right expertise to review the draft before it is shared or used. They are responsible for checking the information against the original sources, even when AI has added references. Important conditions must be included, and you need to check whether anything has been missed or invented.

Count the time spent checking and correcting.

Describe how you do the task today and the quality you expect. Then try the new approach on several comparable tasks. One good result can give you something to investigate further, but you need to see whether it holds up before changing an established workflow.

Count all the time involved: preparation, instructions to AI, review and corrections. Also check whether the result is accurate, includes the necessary information and is useful for the task. If checking takes longer than doing the task yourself would have taken, the quick draft has not saved you work.

Agree on when to review the experiment, and write down why you want to continue, change it or stop. If the approach works, save a short guide with the material, instructions and checks. The next colleague should be able to try the same task without starting from scratch. You can update the guide as you learn more.