Before you talk to me about AI, look at how we enter a delivery note

Let’s stay grounded. A small company, a warehouse, a truck arriving. Someone has to make the system show that the goods have come in.

Often it’s the warehouse worker, starting from the delivery note (the “bolla”). They enter codes and quantities, line by line, into an old, slow management system with a procedure full of steps. I haven’t counted them: let’s say about forty, and the number is there to give the idea, not a measurement.

When a lot of goods arrive with a long packing list, half a day disappears there. And for all that time the warehouse worker isn’t doing warehouse work. It’s not their fault, nor the fault of whoever wrote the procedure years ago with the tools they had.

Before innovation, I would break the problem down. I’d put a layer on top of the management system that cuts the steps to a few: upload the file, check, confirm. No AI, just a simple flow running on top of another one, but covering less ground.

How to actually do it, I don’t know yet. It depends on the system: it may have a file import, an API, or the only option is automating the interface, which is the most fragile. That’s where the hard part is, not in the idea.

You’ll tell me: AI can read the delivery note (PDF or photo) and extract the data. True, and it’s a sensible use.

But if the flow behind it stays slow and full of steps, I just have an AI that reads the note better and leaves me with the same mess. Flow first, then AI where it helps. Maybe even to read the delivery note.

If the software gets a code wrong, who notices?

This isn’t a post against AI. It’s against AI as the first move.