AI was the talk of the day at all the manufacturing companies I visited in recent months. Predictive maintenance, smart scheduling, automatic quality checks. The ambitions are high, so are the expectations.
When I then ask further questions, an awkward silence often follows. The question I ask is invariably the same. What does your data look like? At best, one word falls: fragmented. More often than not, the data is in an Excel file somewhere. Or in a system that records all sorts of things, but not at the level you need to extract actionable insights from it. And at worst, the data is in the head of the senior operator who has been on the production line for more than 12 years.
No working AI without good data
I'm really not the first to say this, but an AI model is only as good as the data it is trained on. Good to reiterate that here. In practice, this means that if your ERP system does not record which machine performs an operation, with which parameters, at what time and with which material you actually have nothing. You won't then be able to recognise a pattern, identify a deviation and certainly not be able to make a sensible prediction.
The manufacturing industry has been struggling with this problem of missing data for a long time, even before the advent of AI. When AI was still in its infancy, it didn't really matter either. But now that AI applications are at everyone's fingertips, it becomes painfully obvious that the data needed to put AI to good use just isn't there.
The solution lies on the shop floor
The question then naturally arises: where is this good, structured data supposed to come from? Not from a fancy dashboard, nor from a BI tool. But simply from the shop floor itself. From the daily recording of what actually happens in the production environment. Processing times, scrap, material consumption, machine status, but also deviations from the parts list.
I see it every day when I visit a manufacturing company. For example, a milling machine that consistently takes a bit longer than planned for a particular material. Nothing dramatic. But if you don't register that, you never see the pattern. Moreover, your planning is structurally incorrect. Your margin is smaller than you calculated. And you don't know why.
The companies that will soon be able to really do something with AI are probably not the ones talking the loudest about it now. They will be the companies that have been consistently tracking what happens on their shop floor for years. Preferably in an ERP system that treats shop floor registration not as an afterthought, but as the core. One where an operator records for each work order what actually happened: which machine is active, what material is used, how much time it takes, what deviation was found. Not in a separate system. Not in Excel. Just in the heart of your operations.
By: Chris de Vries, sales engineer at ECI Software Solutions
