How to know where to start with AI in your company

Start with the work that is difficult to move forward

“We want to use AI” expresses an intention, but it does not yet define what needs to change. Two companies with the same applications can have completely different problems: one takes too long to respond, while another responds quickly but with incomplete information.

The first step is to describe a recognisable situation. What does someone have to search for repeatedly? Where is the same data copied? What is left waiting? Which task comes back because a check was missing? Choosing one of these situations makes it possible to investigate something real, not a list of features.

Reconstruct a case from beginning to end

Choose a recent and sufficiently common case. Record what started it, who was involved, which tool each person used, what information they needed and what result completed the work. Include waiting and corrections: they are also part of the process.

Do not confuse the written procedure with what actually happened. An instruction may say “record the request”, while in practice someone copies it into two spreadsheets and sends a message. That difference helps reveal where the effort lies.

Choose a trial you can interpret

A first scope should be meaningful, but contained enough to understand its result. Preparing context and a draft response can be a useful trial. “Automate all sales” combines too many decisions to know what improved and why.

Before building, define what you will observe: time spent, errors, percentage of cases needing correction or response quality. Record which part a person will continue to perform. Do not assume that all current time will disappear.

Decide what you need to verify first

If you still cannot describe the process or distinguish its opportunities, start by organising the situation. If the objective is clear but there are doubts about data, access or dependencies, those elements should be checked. If the evidence is already sufficient, a first implementation can be designed.

The journey is not an obligation to buy phases. A good decision may be to simplify a task, improve a document or measure more before introducing AI. The final question is simple: which small improvement deserves a trial, and what would we need to see to keep it?

Exercise for your next meeting

Bring a recent case and ask the team to reconstruct it without discussing new tools. Mark one repeated search, one wait and one correction. Then choose which of the three should be investigated first and explain why.

The examples are illustrative. They do not represent an active integration, a client case or a guaranteed result. Application depends on the process, sources and agreed scope.