The Sentence That Ends Projects Before They Start
Many predictive maintenance initiatives don't die from budget or technology. They die from one sentence in the first vendor meeting: "Before we can detect anything, we need to collect six to twelve months of data."
That means: a year of cost without value, a year of project momentum evaporating, a year in which the managing director asks at every status meeting what all this is actually delivering. No wonder many plants walk away at that point.
The good news: that sentence describes one particular class of approaches — not the technology itself.
What Works Without Any History
There is a category of monitoring that needs no historical data and no trained model: statistical monitoring of every signal against its own normal range.
Our system learns each signal's baseline from the first readings in live operation — and monitors continuously from then on. Three independent statistical methods run side by side: one for sudden spikes, one for slow sustained drift, one for abrupt level shifts. The drift method is the decisive one: it accumulates small deviations instead of waiting for one big one — and thereby sees exactly the creeping developments that fixed alarm limits are structurally blind to (why those are the most expensive failures).
Concretely: value from week one, with no data collection project. No waiting period in which nothing happens.
What Honestly Comes Later
To avoid any false impression — the staged approach has two stages, and the second one does need data:
Stage 1 — immediately: the universal watchdog described above. It doesn't know your machine, but it recognizes when it behaves differently from its own past. That covers a surprisingly large share of real-world cases.
Stage 2 — as soon as your data supports it: machine-specific trained models that have learned the particularities of your line. If you already have data in PLC logs, MES, or a historian — and most plants do, often without knowing it — that training can start immediately. If not, stage 1 collects the data as a side effect while already monitoring.
The difference to the "collect first" approach: collection is not a pre-project here. It is a byproduct of running value.
How to Spot the Dubious Promises
Two warning signs in a vendor meeting:
- "We first need X months of data" without the counter-question of what data your controllers already produce. Anyone who doesn't ask about PLC, MES, and historian first is planning your project past your existing assets.
- An accuracy promise before ever seeing your data. Accuracy depends on your machines, sensors, and data. The only honest order is the reverse: measure first, then quantify — our Predictive Maintenance module (fixed price after scoping, 7 weeks) ends with an accuracy report for your own equipment. If it comes out poorly, that is what it says.
Whether your existing data is enough for stage 2 is exactly what the AI Readiness Assessment answers — €9,500 flat, 1–1.5 weeks, open-ended. Or just ask us directly.