We run innovation workshops with companies that want to know where AI helps them. The first day is always the same: at the end there are thirty ideas on the wall, and everyone is excited. The second day is the important one.
Thirty ideas on the wall
Chatbot for customer service. Summarise documents automatically. Write proposals faster. Check invoices. An assistant for sales. All good ideas, all heard before.
The problem is not that the ideas are bad. Fact is: thirty ideas mean none gets done. Or the one gets done that was pitched the loudest.
The three questions
On the second day we go through the wall. Every idea gets the same three questions. Whichever cannot answer one comes down.
Which decision becomes faster or better with this? Not which process gets automated. Which decision. If the answer is "none, but it saves time", that is usually an idea for a script, not for AI.
Do we have the data, and who understands it? A name. If the data sits in three systems and nobody knows which one is right, that is a data project. You can do that, but it is not the idea hanging on the wall.
What happens when it gets it wrong? Who notices, when, and what does it cost until then? If the answer is "that must not happen", the idea is not ready yet. AI gets things wrong. The only question is whether you can live with it.
What is left
After the three questions, four or five of thirty ideas are still on the wall. Sometimes three. Once, one.
Those are the ones where someone could say: this decision, this data, this person knows it, and if it goes wrong, that person sees it the next morning and corrects it. With those you can build something in four weeks that runs.
And the twenty-five that came down? They are not gone. They are just not up yet. Some become data projects. Some become simple automations without AI. And some are ready in a year, once the first five have run and the company has learned what a system feels like that is sometimes wrong.
Why nobody likes hearing this
The second day is uncomfortable. On the first day everything was possible. On the second, almost everything is suddenly not up yet. I have been in workshops where the mood clearly turned.
But I have also seen the companies that skipped the second day. They started on ten ideas at once, had none finished after a year and then said AI does not work for them.
The three questions are not there to kill ideas. They are there to find the one that actually gets built.
Which decision. Which data, which person. What if it gets it wrong. Whoever can answer these three questions no longer needs a workshop. They can start.
Written by me. The thoughts, the values, the learnings, the mistakes: all mine. Grammar and spelling are corrected by our own twin model, trained on my texts. Sometimes a stumble stays in. That is mine too.