When a company decides to "do something with AI", the first instinct is to pick a tool: ChatGPT, Copilot, Gemini, a chatbot for the website. It's understandable, but it's the wrong starting point. The right tool depends on what you want to improve, and until that is clear every subscription risks becoming one more expense that nobody uses.
Step one: find the work that weighs
Every business has tasks that repeat and that nobody enjoys: answering the same customer questions over and over, copying data from an email into a management system, preparing quotes that look alike, summarising long documents. They are the ideal candidates. To find them, just ask the people who work with you what they would happily remove from their week, and how much time it takes. Their answers are almost always more precise than any theoretical analysis.
Step two: choose one case and measure it
From the tasks that emerged, choose just one: the one that takes the most time and is easiest to describe. Before changing anything, measure how it is today: how many hours a week it takes, how many errors occur, how long the customer waits. Without this starting point, in a month you won't be able to tell whether AI brought a real benefit or just something new.
Step three: test small, with the right people
At this point the tool almost chooses itself, because you know what it has to do. Often what the company already has is enough, such as Microsoft 365 or Google Workspace, which include artificial intelligence features. Run a trial of a few weeks with two or three motivated people, on real cases, and note what works and what doesn't. Involving the people who will actually do that work, and training them first, is the difference between a project that lasts and one that is dropped after a month.
Step four: compare the numbers and decide
At the end of the trial, put the starting figures side by side with the new ones. If the time saved and the quality of the result justify the cost, extend it to more people and move on to the next case, perhaps with a custom solution. Before using customer or employee data, also check what the AI Act requires of a small business. If they don't, you have spent little and learned a lot, which is still a good result.
One last piece of advice
Don't start with the most ambitious project. A small first case that works convinces the team more than any presentation, and builds the right habit: using artificial intelligence to solve concrete problems, not to chase the latest novelty. If you want to take this path with someone at your side, that is exactly the work we do in our introductory consultation.
