At its core, predictive maintenance involves using historical data, time series, anomalies, and operational data to anticipate future problems and thereby avoid them. This is nothing new. What is new is the ability to do it better, and the question is why so many people still aren’t doing it.
What do AI and machine learning really bring to the table?
In practice, machine learning significantly enhances predictive maintenance. Whereas traditional models require us to define in advance what the model should look like, AI-based methods can identify patterns and correlations in large datasets on their own.
“Machine learning can identify patterns and spot correlations that we have trouble seeing—and it can do so in a very short amount of time,” says Robert Velén. However, it’s not just about faster analysis; it’s about a shift in how we build and use models.
“Without machine learning, we have to decide what the model should look like. With AI, the model has ‘rubber legs’ and can adapt.”
Despite this, many initiatives are still in the pilot phase. A recurring pattern is that the work often begins with vibration and noise measurements, not least because sensors have become cheaper and easier to implement. It’s a logical first step. But that’s also where many stop. Starting with what’s easy isn’t the same as creating real value, and it’s precisely in the transition from pilot to widespread implementation that many organizations get stuck.
From Pilot to Breakthrough—What's Holding Us Back?
The biggest obstacle to a real breakthrough is rarely the technology itself. It’s how organizations handle data. Today, the industry sits on enormous amounts of historical data, sometimes going back several decades. The problem isn’t a lack of data, but rather that it’s often fragmented, unstructured, and difficult to use in practice.
In other words: organizations overestimate the value of the data they have and underestimate the work required to make it useful. It’s about integrating multiple data sources:
- Process control systems (e.g., temperature, pressure, flow rates)
- Maintenance Systems and Deviations
- Manual measurements and lab data
- Vibration and Sound Sensors
Getting them to “speak the same language” is not a technical side project; it is a fundamental prerequisite. At the same time, there is another, often underestimated, challenge: trust.
When early pilot projects fail to deliver perfect results, there is a risk of creating lasting skepticism. What should be viewed as a learning phase instead becomes an argument against moving forward. This is problematic because it is precisely in the early stages that the models improve the most.
The possibilities are greater than many people realize
Many organizations still view predictive maintenance as a way to prevent breakdowns. That’s part of the picture, but it’s far from the whole story. The real potential lies in fundamentally changing how maintenance is planned and managed. By adopting a more data-driven and structured approach, organizations can shift from reactive to proactive maintenance, where decisions are based on actual insights rather than assumptions.
With the right approach, organizations can:
- Improve availability and uptime
- Reduce unplanned downtime
- Optimize Scheduled Maintenance
- Reduce costs and resource consumption
In practice, this means that maintenance is performed at the right time—neither too early nor too late. This not only reduces costs but also improves resource utilization and contributes to more sustainable operations. However, this requires viewing predictive maintenance as more than just a technical initiative. It is an operational issue.
Our Outlook for the Future
We see that the next step in this evolution isn’t about doing more—it’s about doing it right. Many organizations have already begun this journey, but get stuck in isolated pilot projects or technology-driven initiatives that aren’t linked to business value.
The real shift only occurs when we stop asking what technology can do and start asking what the business needs. This means taking a more integrated approach to data, analysis, and decision-making. Because the question is no longer whether data is available, but whether it is being used to make better decisions.
The key is not just to collect more data, but to:
- Structure it correctly
- Build Trust Within the Organization
- And link insights to concrete business decisions
This also requires more than just technology: it requires the courage to challenge established ways of working. Moving from experience-based decisions to data-driven ones is not a technical change—it is a cultural one.
“Ultimately, it’s not about the technology itself, but about how we use data to make better decisions—that’s where the real value lies,” concludes Robert Velén.
The organizations that succeed are not necessarily the ones with the most data, but rather the ones that best understand how to use it.
