Why Human Analysts Make AI Condition Monitoring More Reliable, Not Less

Fully automated platforms and AI-driven models not only provide alerts (and a lot of false ones), but they also provide a long list of items to check. But that’s it. No follow up and no one to ask questions about what’s most critical or where to take another look.

Back-and-forth interactions between vibration analysts and facility maintenance and reliability teams allow us to assist in the troubleshooting process and get down to the problem and resolution. A reliability partner like Waites is there every step of the way, answering every question and providing follow-up analysis.

Here’s why the human touch is essential in your predictive maintenance program.

If AI is so smart, why does Waites put a human between the algorithm and your dashboard?

Because the alternative is worse, and every maintenance team that has lived through it knows predictive maintenance needs a human touch.

Having a human in the loop avoids maintenance teams waking up to 40 alerts on a Monday morning, 38 of which are nothing. It avoids having a technician pulled off a real job to chase a vibration spike caused by a forklift driving past a sensor. Most critically, it prevents someone switching off notifications and missing the one alert that shouldn’t go unseen.

Some vendors describe Waites’ human-in-the-loop model as a dependency: “You’ll rely on their analysts.” Our customers put it differently:

“Waites has become a core part of our reliability program. Across 600+ assets, we’re consistently detecting vibration-related issues early and addressing root cause issues, avoiding 483 hours of downtime in one year alone.”

Matthew A. McLaughlin, CRE, Reliability Superintendent, Domtar

You’ll rely on an analyst the same way you rely on a radiologist to read an MRI. While the machine produces predictive data, an ISO-certified vibration analyst confirms what it all means, how urgent it is, and what to do next before the alert ever reaches your team.

We’ve proven that people are not a limitation of our platform. Their collaboration is the crucial reason it works.

Every alert Waites sends has already been reviewed by a ISO-certified CAT II-IV vibration analyst. That’s a service built on trust and a system solution that goes beyond installation.

What’s wrong with fully automated condition monitoring?

To put it simply, AI-only systems can generate false positives.

Fully-automated condition monitoring systems have false positive rates that overwhelm maintenance teams and erode trust in alerts.

AI is extraordinary at detecting anomalies. It is far less capable of deciding whether an anomaly matters. A vibration signature that looks like a stage-two bearing defect to an algorithm might actually be:

• A temporary process change on the line.
• A product or load change on the machine.
• Seasonal temperature swings affecting lubrication.
• A nearby machine transmitting vibration through the structure.
• A sensor that simply needs to be re-seated.

An algorithm sees a pattern, but a vibration analyst sees a machine, in a plant, in context. Without that second layer, the anomaly becomes an alert, the alert becomes a work order, and the work order becomes wasted wrench time.

Certified vibration analysts at Waites review every alert before it reaches the customer. This is a service, not a dependency.

What is alarm fatigue, and why does it kill condition monitoring programs?

Alarm fatigue is what happens when a monitoring system cries wolf often enough that people stop listening. It follows a predictable arc:

1. Week 1–4: The team investigates every alert. Many turn out to be nothing.

2. Month 2–3: Alerts get triaged casually, i.e.  "probably another false one."

3. Month 4+: Notifications are muted, dashboards go unchecked, and the program exists on paper only.

4. The failure: A real defect develops and the system flags it. Yet, nobody is quick to act on it because the system isn’t trusted anymore.

At that point, the plant has paid for sensors, software, and installation and experienced an unplanned failure anyway. While it’s true that the technology didn’t fail, lack of oversight and humans’ desire to trust in the new system did.

A condition monitoring system is only as good as your team’s willingness to act on its alerts. Human validation is how that willingness survives past 3 months.

What does a Waites analyst actually do when an alert fires?

“Human in the loop” is easy to say. Here’s what that process actually looks like when our AI flags an anomaly:

1. The AI detects and classifies. Trained on more than 13 trillion data points and 10 billion daily readings, Waites’ models detect early signs of wear, imbalance, or impact events and separate anomalies by degradation type, such as progressive, acute, or long-horizon.
2. An ISO-certified analyst pulls the data. A CAT II+ vibration analyst opens the full spectral and waveform data — not just the alert summary — and examines the signature against the machine’s history and baseline.
3. They apply machine and plant context. Is this consistent with the machine’s known fault frequencies? Did operating conditions change? Has this asset shown early-stage indicators before? Analysts often coordinate directly with your team to understand what’s happening on the floor.
4. They confirm, reclassify, or dismiss. False positives are filtered out before you ever see them. Real issues are confirmed and assigned a severity and urgency.
5. They write the recommendation. Specifically, what the issue is, why it matters, and what to do next, all delivered through the Waites Dashboard as a clear, prescriptive action item, not a raw anomaly score.
6. The system learns from the expert, not from a guess. Analyst-validated outcomes are fed back into the machine learning models, so the AI improves on verified truth rather than on whichever technician happened to dismiss an alert at 2 a.m.

That last step matters more than most buyers realize. Platforms that rely on plant staff to validate alerts are training their algorithms on unverified feedback. When a technician dismisses a real early-stage defect as a false alarm, the model learns to ignore that pattern — everywhere, for everyone. Expert validation doesn't just protect this week’s alerts. It protects the accuracy of the entire system over time.

Why the sensor spec race misses the point

Much of the marketing in this category is a race to bigger numbers: higher sample rates, wider frequency ranges, more resolution. However:

“Vibration analysis doesn't work because the sensor has the highest sample rate, frequency, or resolution. All the bells and whistles mean nothing if you don't have the right machine information and set the settings up to get good data. If everything is left at the default of all those high frequencies, you basically get bad, low-resolution data, which makes it terribly difficult for any sort of targeted diagnostics.” – Dave Porter, Senior Director of Reliability Services, Waites

In other words: an ultra-high-spec sensor left on default settings produces worse diagnostic data than a modest sensor configured correctly for the specific machine it’s mounted on. As Dave Porter puts it:

“We can do analysis better with a bottom-shelf, low-frequency sensor set up properly than anyone else can with the Ferrari of sensors, not set up. It's very common for us to take our 11.2 kHz sensor and set the frequency to 2,800 Hz or even lower, depending on the machine. It’s all about getting the right data at good resolution, not how high the frequency can go.”

This is the part of a human-in-the-loop approach that happens beforehand. Waites analysts look at each machine, gather the right machine information, and configure sensor settings to capture the best possible data for targeted diagnostics. The exception: advanced ultrasonic signal processing techniques such as the ImpactVUE^® high-frequency data acquisition solution genuinely require high sample rates, and even those must be configured correctly or the data is no good.

Specs detect. Setup diagnoses. A monitoring program built on fundamentals — correct configuration, good data, expert interpretation — will outperform a spec sheet every time.

Isn’t relying on Waites analysts a dependency?

No, it’s a real advantage. Every condition monitoring approach depends on human expertise somewhere.

Considering these three models, the analyst advantage is clear:

Waites’ expert-driven process detects 99.92% of downtime-causing issues before failure occurs — not only because of the AI but because of the people behind the sensors.

What results does expert validation actually deliver?

Across industries, the human + AI model produces outcomes that unreviewed alerts don’t:

• At Domtar's Kingsport packaging mill, a collaborative program pairing Waites analysts with the plant’s reliability team cut unplanned mechanical and electrical downtime from 10% to roughly 3% over 18 months. Across 676 deployed sensors, the team prevented 254 downtime-causing events, avoided 1,138 hours of downtime, and saved $9.17 million in downtime-related losses.

• At an energy facility, Waites analysts tracked a slow-building vibration trend on an induced draft fan for months, then escalated the alert to critical when levels spiked. Inspection confirmed the fan wheel had separated from its hub mount. The facility replaced the wheel, shaft, and bearings under controlled conditions, avoiding 96 hours of downtime and saving $406,000.

These cases required a human in the loop to confirm the finding, rank its urgency, and tell the team on the floor exactly what to do—and early enough to make a real difference in the bottom line.

FAQ: Common Questions About AI Condition Monitoring

Do I need to hire my own vibration analyst to use Waites?

No. That’s the advantage of using Waites’ system solution. ISO-certified CAT II+ vibration analysts are built into the service, experts who review every alert, configure sensors correctly for each machine, and deliver prescriptive recommendations through your custom dashboard. Customers who already employ reliability engineers or analysts find Waites complements them: your in-house experts spend their time on confirmed, prioritized issues instead of screening raw anomalies. Customers without in-house analysts get expert-level coverage without a $100K+ hire, certification costs, or turnover risk.

What happens if Waites analysts are unavailable? Am I stuck?

You’re never waiting on a single person. Waites operates as an extension of your team with its own analyst and support teams behind the dashboard, so coverage doesn’t disappear with vacations, illness, or attrition (the single-point-of-failure risk that comes with hiring your own analyst). Meanwhile, monitoring itself never pauses: sensors stream continuously and the AI detects and classifies anomalies around the clock, with acute events prioritized for the fastest review. And your team always retains full access to live data and dashboards, with expert validation as a layer on top of your visibility, never a gate in front of it.

Does human review slow down alerts?

Detection is instant and automated, and the AI never sleeps. Validation is prioritized by degradation type, so acute events get the fastest turnaround while long-horizon wear is reviewed on an appropriate cadence. In practice, the minutes spent on expert review are trivial next to the hours teams lose chasing false positives from unreviewed systems — and next to the weeks or months of warning that early-stage fault detection typically provides.

The alternative (no human review) leads to alarm fatigue, missed real failures, and eroded confidence in the system. Waites' analysts reduce the burden on the customer's team, not add to it.

Can’t AI just replace vibration analysts eventually?

AI keeps getting better at detection, and Waites’ models improve daily. But diagnosis requires context no sensor transmits: what changed on the line, how this machine behaves under this load, whether that signature is a defect or a process artifact. Just as importantly, the accuracy of the AI itself depends on expert-verified feedback. Models trained on unverified dismissals learn to miss real faults. Waites believes the future of predictive maintenance isn’t AI instead of analysts. It’s AI that makes analysts dramatically more scalable, and analysts that make AI dramatically more trustworthy.

How is this different from fully automated competitors?

Fully automated platforms deliver every anomaly their models flag directly to your team and leave interpretation to you. Waites delivers expert-validated alerts: every notification has been reviewed by a certified vibration analyst, assigned a severity, and paired with a prescriptive recommendation. One model sells you detection, while the other delivers decisions.

AI-Powered Prescriptive Maintenance with a Human Touch

AI solutions are good at detecting and predicting problems, but it’s the hardworking people behind the advanced sensor tech who determine what’s critical. Waites analysts, genuinely awesome humans, confirm action items, prioritize them, and help your maintenance team resolve them quickly. By working together, real collaboration is what delivers the reliability you can trust.

See expert-validated condition monitoring in action.