By Malcolm Schulstad & Aliesha Aden
Most conversations about predictive maintenance jump straight to AI, analytics, or sensors. Very few start with the thing that actually determines whether these systems work on site: Whether anyone bothers to close the loop and provide feedback.
Across the thousands of assets we’ve monitored, one pattern shows up again and again. Predictive maintenance doesn’t fail because the technology isn’t good enough. It fails because validation and feedback aren’t baked into the process.
It’s the quiet reason why many ‘transformations’ stall after the first wave.
When AI Ambitions Meet Messy Reality
Several years ago, we kicked off an internal project known as the Galaxie AI Initiative.
The goal was ambitious but logical: use the behaviour of certain assets to predict the behaviour of others, group similar assets, and let machine learning pick up patterns that humans couldn’t.
This was a solid concept on paper. But it exposed a bigger problem: some of the most basic data feeding our models wasn’t reliable.
One key input was the simple ON/OFF state, whether an asset was actually running. That should be black and white. But it just wasn’t.
When we dug in, we found:
- Over 30% of running-state indicators were incorrect
- Some were misconfigured from the start
- Others drifted slowly over time
- Exceptions and edge cases chipped away at accuracy
- And we didn’t have enough validated data to override the noise.
But even the best algorithms can’t compensate for unreliable inputs.
The lesson was clear: If your inputs aren’t accurate, and nobody validates your outputs, then you’re not doing predictive maintenance… You’re doing guesswork.
The Real Bottleneck: Time and Workload
Every modern monitoring platform includes a way for teams to validate alarms and confirm what was actually found. On paper and in theory, this should create a healthy feedback loop. But in reality, it runs head-first into the pressures of a busy site.
This sound familiar? Maintenance and reliability teams are already stretched, often working through long backlogs and juggling urgent jobs. When time is tight, the immediate value of ‘updating the system’ isn’t always obvious. People naturally focus on the hands-on work in front of them, and the validation step becomes “something to get to later”… or not at all.
As a result, only a handful of champions consistently keep the system updated, and even they can struggle to maintain it long term.
Without regular validation, the platform stops learning. Alarm accuracy plateaus, users notice that the insights aren’t improving, and trust quietly erodes. Engagement drops, the system is used less and less, and eventually it fades into the background.
Not because it failed dramatically, but because no one had the time to keep the feedback loop alive and the insights gradually stopped being relevant or trusted.
Why Feedback Loops Matter
Predictive and prescriptive maintenance depend on a simple cycle: the system raises an alert, someone investigates, the outcome is recorded, and the model improves. When that loop is broken, everything downstream suffers.
In many organisations, the loop breaks early. The system generates an alert, a technician goes out to inspect the asset, and work is carried out. However, the findings never make it back into the platform. With no recorded outcome, the system continues operating blind.
Over time, alerts feel inconsistent, accuracy stalls, and confidence gradually drops. Usage drops, and the platform struggles to deliver the value it was designed for.
A healthy loop looks very different. The system raises an alert with a diagnosis, a technician investigates, and the outcome is captured quickly as to whether:
- The diagnosis was correct
- A different fault was found
- The issue was process-related
- Nothing was wrong at all
That small action gives the system the data it needs to refine its logic. Accuracy improves steadily, insights become more precise, and teams begin to rely on the alerts because they see them getting smarter.
This is the point where predictive maintenance stops being a side project, and becomes a genuine decision-making tool that supports day-to-day operations.
What the Data Shows When Feedback is Embedded
Within our platform PlantOS™, we designed validation as a core workflow, rather than as an afterthought. Across a sample of more than 14,000 monitored assets, the impact is clear.
The system generated just under 11,000 fault notifications. When those alerts were actively validated, 768 alarms were marked inaccurate and investigated further. Of those, 102 were different faults, 170 were due to process or environmental causes, and 498 were recorded as No Fault Found (NFF). That NFF rate is under 5% of all alarms raised.
Fast-forward across a 12-month period, the system missed 51 issues, less than 0.5% of the total events recorded. Not perfect, but pretty accurate. And every ‘incorrect’ alarm became training data that strengthened the system over time.
This is what modern maintenance looks like when feedback is part of the workflow.
How to Make Validation Business as Usual
If feedback is essential, the real challenge is making it sustainable in environments where everyone is already stretched. This isn’t something that gets fixed with a single training session. It becomes effective only when validation fits naturally into the way teams already work.
A strong validation culture starts with treating the outcome of an alert as part of closing the job.
If an alert triggered the work, the job isn’t truly finished until the technician records what they found. That outcome should link directly back to the original alert so the system can learn from it. Without that integration, the step gets skipped and the feedback loop breaks down.
It also needs to be quick.
If providing feedback takes too many steps or requires a written explanation every time, people simply won’t do it. Most outcomes should be recordable in seconds, whether the alert was correct, whether a different fault was found, whether the issue was caused by process, or whether there was no fault at all. Any extra detail should be optional.
Showing people that their feedback actually changes something.
When updates disappear into the system with no visible impact, engagement drops fast. But when teams see fewer nuisance alarms, more accurate diagnoses, clearer fault types, or examples of avoided downtime that came directly from technician input, they understand the value of validating their work. The system should make it obvious that it improves because people participate.
And finally, none of this works without change management.
Predictive maintenance is as much about people and process as it is about technology. That means setting expectations with leadership, giving teams context about why validation matters, supporting the champions who model the behaviour, and aligning everything with the CMMS so validation doesn’t exist in isolation. When these pieces come together, predictive maintenance stops feeling like extra admin, and becomes a smarter way to run the plant.
Modernising Maintenance Starts with Validation
It’s easy to get caught up in the promise of AI, analytics and digital transformation. But underneath all the technology, the fundamentals are still simple: if your data is wrong, your predictions will be too. If nobody validates the outputs, nothing improves, and if nothing improves… people stop trusting the system.
For most teams, this isn’t a technology gap, rather, it’s a workflow gap. The real shift happens when validation becomes part of how the work is done, not an extra task squeezed in between competing priorities. When outcomes are captured consistently, the system learns, alerts become more accurate, and people start relying on the insights. In other words, the whole maintenance program becomes easier to manage, not harder.
This is where the value emerges: fewer surprises, smarter decisions, and assets that last longer because issues are caught early. And it doesn’t require a major transformation. It starts with small, consistent habits.
At MOVUS, we’ve built PlantOS™ to support this kind of workflow. Validation is woven into the process, made quicker for technicians, and visible in the results. When feedback is easy and the improvements are clear, teams engage, and predictive maintenance becomes something that genuinely helps rather than something else to manage.
Modernising maintenance isn’t about adding more technology. It’s about closing the loop so the technology can do its job.
So the real question is: what would improve most on your site if your feedback loop was working as well as your assets need it to? Get in touch with us, we would love to hear about your experience!