Getting started with AI in your business: choose a problem before a tool

First, describe the work you want to improve

A business can buy new tools without changing the way it works. I prefer to ask: which task keeps recurring, who handles it and what gets in their way? Answering that makes the conversation about the business.

Take supplier quote comparisons. The useful output is not an elegant summary. It is a clear view of equivalent products, quantities, delivery times and terms, with missing information identified. Without defining that first, it is hard to judge what the AI produces.

A first exercise: compare suppliers

1. Choose documents that you can use. Gather the quotes and the terms relevant to that purchase.
2. Define the output format. For example, a table showing the product, quantity, currency, lead time and source document.
3. Make the uncertainty explicit. Ask it to flag missing fields and keep each supplier’s terms separate.
4. Check the result. A person checks the figures against the original documents before deciding.

This is a practice exercise, not a reported customer outcome. It is also the kind of business problem we explore in my AI course for businesses.

How can you tell whether the first use is worthwhile?

Measure the whole process: preparation, generation, review and corrections. Compare the effort involved before and after, the errors you find and whether the output helps you make a decision. A fast draft that takes longer to review may not be an improvement.

Repeat the exercise with different documents. One good result is not enough to understand the conditions in which it works. Keep the failed cases too: they often reveal rules that have not yet been made explicit.

When should you take the next step?

Once the team can explain what the tool does, where its information comes from and who checks the result, you have a basis for expanding its scope. You might turn the request into reusable instructions or consider an integration. If every attempt depends on improvisation, make the process more consistent first.

Experience builds when we preserve what we learn in the process. In my work with AI, that continuity matters as much as the tool itself. The next step is to understand what an agent needs to work in a business.