Condition monitoring becomes useful when the process can turn a signal into diagnosis and timely action.
A predictive-maintenance system can produce thousands of readings without preventing a single failure. The operational value appears only when the organisation converts a signal into a timely diagnosis, a controlled maintenance action and verified equipment performance. A recent case from Domtar’s Kingsport, Tennessee, paper mill shows how much of that value depends on the process surrounding the technology.
Business Insider reported on 1 September 2026 that Domtar had already installed AI-assisted vibration sensors from Waites Sensor Technologies. When a motor failed, reliability engineer Matthew McLaughlin was asked to review one day’s data from 450 sensors. He estimated that a manual review would have taken 32 weeks. The data existed, but the original working method could not use it at the required speed (Business Insider, 2026).
The case matters beyond maintenance. It shows a common implementation pattern: a tool increases information faster than the organisation redesigns ownership, review, response and learning. The corrective action was not another dashboard. Domtar changed the operating routine around the existing system.
What Changed in Domtar’s Maintenance Process
More data exposed a new process constraint
Before condition monitoring, the practical constraint may be detecting deterioration. After hundreds of sensors are installed, the constraint can move to interpretation and response. The organisation now sees more possible problems than one engineer can review manually, so adding data without changing the process may increase the backlog rather than reliability.
Domtar treated that mismatch as a process-design issue. The plant began using analyst support already included in the sensor provider’s service. Monthly discussions became weekly calls, and the mill shared evidence from thermal imaging, infrared cameras and ultrasonic meters. Business Insider reported that the team saw significant improvement within three months of this more active routine.
A review cadence turned alerts into shared decisions
The weekly call was useful because it established a regular point for evidence to become a decision. The provider’s analyst learned the facility, its equipment and staffing conditions, while Domtar contributed local knowledge and additional diagnostic data. AI performed the primary analysis, but provider analysts, on-site specialists and maintenance staff worked together to diagnose the condition.
That combination is stronger than either extreme. Fully manual review could not keep pace with the data, while an automatic alert could not understand every operating context or maintenance trade-off. The redesigned process assigned complementary roles: technology screened the signal, specialists interpreted it, and the plant owned the action.
Visible follow-through built trust in the system
McLaughlin’s team tracked network and equipment actions, including response time, and kept action items under 30 days old. He also sent daily summaries of alerts and responses to experienced employees. Over time, managers stopped calling to ask whether each alert was being handled because the routine made follow-through visible (Business Insider, 2026).
Trust did not come from asking employees to believe the technology. It came from repeated evidence that signals were reviewed, action was taken and results were communicated. This is an important continuous improvement principle: confidence grows when the control loop operates consistently, not when a launch presentation promises accuracy.
Design Predictive Maintenance as a Closed-Loop Process
Define the path from signal to verified action
A useful condition-monitoring process needs more than thresholds. It should define which assets and failure modes are covered, who receives each category of alert, how quickly it must be triaged, which evidence confirms the diagnosis and who can authorise intervention. It should also distinguish a monitoring issue, such as a damaged sensor or missing data, from an equipment condition that requires maintenance.
The process should end with verification. After lubrication, alignment, repair or replacement, the team needs to confirm that the condition measure returned to an acceptable pattern and that the intervention did not create another problem. Closing a work order without checking the operating result records activity, not reliability.
Manage the queue, not only the individual alarm
As coverage expands, alerts form a queue. Teams need a method for prioritising that queue by safety, production consequence, lead time for parts, rate of deterioration and confidence in the diagnosis. A high-severity condition on a non-critical asset may need a different response from a modest but rapidly worsening condition on the process constraint.
Ageing rules are equally important. Domtar’s attention to keeping actions under 30 days old provides a simple control, although every organisation should set response limits according to asset risk. Reviews should examine overdue actions, repeated deferrals and alerts that reopen after closure. Those patterns can expose weak planning, spare-parts availability, access constraints or an incomplete diagnosis.
If you want to build the analysis, redesign and control skills behind this kind of result, explore STEP Bootcamp’s practical Continuous Improvement Training. The 28-day programme combines live classes with applied assignments and personalised feedback.
The signal-to-action loop needs defined evidence, ownership, response and verification.
Combine evidence without creating another data burden
Thermal, ultrasonic, vibration and operating data can strengthen diagnosis when each source answers a defined question. Combining data simply because it is available can recreate the overload that the system was intended to solve. The team should know which additional evidence distinguishes likely causes and which readings do not change the decision.
A practical standard might specify the evidence required for common failure modes, the conditions that trigger specialist review and the minimum information attached to a planned job. This reduces repeated interpretation and makes good diagnostic practice transferable across shifts and sites. It also creates a clearer training need than a general instruction to become data driven.
The Continuous Improvement Lessons Extend Beyond Maintenance
Build routines around decisions rather than dashboards
Gartner’s 1 September survey found that only 22% of organisations had successfully scaled AI across multiple business units or adopted an AI-first approach, even though 85% of functional leaders planned to increase AI spending in 2026. High performers continually tracked returns and reallocated or stopped weak initiatives, while low performers often did not know the return (Gartner, 2026). Domtar’s case gives that broader finding an operational form.
The mill created a cadence, clarified roles, tracked actions and communicated results. Those elements can be applied to many technology-enabled processes. A demand forecast needs a decision owner and exception rule. A quality alert needs containment and cause analysis. A service recommendation needs an approval boundary and customer-outcome measure. The dashboard matters, but the operating routine determines whether the information changes performance.
Measure capacity protected and failure prevented
Business Insider reported that Domtar now monitors 748 sensors and that the mill estimated 1,546.65 hours of unplanned downtime had been saved. The article also reported maintenance refinements such as changes to lubricant viscosity and an instance in which a drive-belt supplier had sold no fan belts to the facility for a year. These are useful signals, although the figures were reported by participants in the implementation rather than produced through an independent evaluation.
A strong benefits method should state the counterfactual and avoid double counting. For an avoided failure, the team should document the evidence that failure was developing, the expected outage duration, the production constraint affected and any costs displaced by the intervention. It should separate theoretical availability from actual good output and recognise planned maintenance time. This makes the result credible enough for larger operating and capital decisions.
Use Kaizen to improve the response system itself
Predictive maintenance creates an ongoing source of process-improvement opportunities. Teams can study false alarms, late responses, repeated faults, hard-to-source parts, unclear work instructions and repairs that do not hold. Small changes to notification rules, planning windows, evidence requirements or standard jobs can reduce future response time and variation.
Good Kaizen Training teachs professionals to improve this response system repeatedly rather than treat installation as the finish line. The most transferable lesson from the Domtar story is straightforward. Better sensing creates potential value; a disciplined signal-to-action process converts it into operating value. Organisations that improve both the technology and the routine around it are more likely to gain reliability, capacity and trust at the same time.
To develop a repeatable way to turn operational signals into verified improvement, review STEP Bootcamp’s Process Improvement Training. It helps participants analyse real problems, design practical changes, measure value and build controls that sustain the result.
References
Business Insider. (2026, September 1). A paper manufacturer got more out of its AI sensors with a simple administrative fix. View source Gartner. (2026, September 1). Gartner Survey Finds Only 22% of Organizations Have Successfully Scaled AI Across Multiple Business Units. View source
Leave a Reply