Der ROI von KI

Manufacturing companies spend 100,000-200,000 francs on a single AI project with no guarantee of any ROI. Three experiments at 5,000 francs each would have answered that before the big bet.

The question I get asked the most is: What is the ROI on AI?

It can be calculated. Just not at the stage where a company is merely interested in AI.

It becomes answerable once you know what you want to optimize and what that costs you today.

So this is the homework:

• Take one process that eats a lot of time and repeats constantly.

• Split it into substeps, each with a clear, measurable outcome.

• Then treat it like teaching an intern a complex task.

An intern would do one step, then check in: "Is this the right direction?"

You would then say, "Keep going," or look at the output and say, "No, redo it."

You have checkpoints, a supervisor, and quality assurance at every step.

By the end of that exercise, you know the complexity, where things go wrong, and how much time every step really takes.

Multiply that by how often the task runs per year, then add the people involved and the external costs.

Now you have the full calculation of what this process actually costs you today.

You would be surprised how many companies have never seen that number. They've worked the same way for 30 years and assume it's the only way.

Once you have that number, go back through the substeps and ask whether an AI can do each specific task.

You judge its output exactly the way you would judge the intern's.

It either can do the step or it can't. There's usually no in between.

Each of those little tests is an experiment.

And once you know where AI can help, run the same process with AI and measure the difference.

That difference is your return on investment, proven step by step instead of betting everything on one project.

Because if the process is unknown and the instructions are unknown, you can try whatever you want.

But you won't be successful.