Where should an industrial company start with AI?

For an industrial company, AI is not automatically a new system. It is a way to reduce the work that comes from searching for, repeating and interpreting information. If that is forgotten, the project starts from a tool and ends as a trial that does not change the day.

The most sensible start is therefore narrow. Where does time go into assembling, summarising or forwarding information that already exists? Where does the same question return week after week? That is where AI can help, if the information is in a sufficiently usable form and responsibility stays with the person doing the work.

Start from work that is already done

A good first use is rarely “AI in production”. A better one is a concrete task: a draft offer, a summary of a maintenance report, a description of an exception, a document search, or compiling a weekly status. In these cases AI does not decide on anyone’s behalf. It shortens the part that is now manual assembly.

• repeated writing, where the content already exists elsewhere
• searching in several places before a decision
• reporting that takes time but does not require a new measurement

Bring AI into the work that is already done — not as a separate programme beside it.

Do not start from data that nobody uses

A wide data lake without a use is a slow way to learn. If production data, maintenance history or project information is not yet used in everyday work, AI will not fix that. First the information has to reach the place where it is read. Only then is a model useful.

When the first use works, the next step is easier to judge. Then it becomes clear whether an integration into the current system is needed, cleaner data, or simply a better way to ask what already exists. An industrial company does not have to start with AI. It should start with one job where AI saves time without turning the operating model upside down.