Manufacturers manufacture. With planning, delivery dates and line efficiency on the daily agenda, a jump to artificial intelligence looks like a distant debate. But the first step toward being ready for AI is not building a model; it is starting to collect data — and that step comes out of the work you already do.
Where the data comes from
Every virtual training session quietly produces a behavioural record:
- How long each step took
- Where the participant hesitated, where they backtracked
- Which type of error repeated, and how often
- Which participant crossed which threshold on which attempt
That record is needed for the training anyway. The difference is that once it accumulates, the same record starts answering a different question. Which procedure step systematically produces errors? Which maintenance operation takes longer than planned? Those are not training questions; they are process questions.
Why starting with a digital twin makes sense
The digital twin provides the skeleton the data attaches to. A measurement is not a row in a spreadsheet but a specific step on a specific part of a specific machine. Data that accumulates already labelled is ready input for a model later — trying to label it afterwards costs more than collecting it properly from the start.
A realistic expectation
This journey does not happen in one move, and should not be sold as if it does. There is a sequence: first report what happened, then relate why, then predict. Each rung is made possible by the data the previous one accumulated.
Only one thing matters today: start measuring. One machine, one procedure, one pilot scenario. The data starts accumulating that day.