Predictive Maintenance Lost Ground: How Prescriptive AI Can Revive It

On October 19, 2026, IIoT World published a summary of a panel at the Industrial AI Summit 2026 that tackled an awkward question: why is predictive maintenance (PdM) no longer at the top of industrial AI use-case lists? The participants were Boris Scharinger (Siemens), Karthikeyan Natarajan (CEO of Infinite Uptime) and Jan Pingel (Ingersoll Rand), with Rob Tiffany of IDC moderating. The article is labeled as sponsored by Infinite Uptime, while stating that it is editorially independent, so the vendor's figures should be read as commercial claims.

A safe use case, but with a ceiling on returns

PdM has long been the ideal first AI project because it does not interfere with production. The flip side is limited financial return: according to the panel, for a $1 million asset the maximum benefit is about $12,000 a year, in line with the rule of thumb of roughly $1,000 per month for eliminating all unplanned downtime. With that ceiling, a vendor needs hundreds of plants and thousands of connected machines to build a sustainable business.

From anomaly to prescription

According to the panel, many implementations stopped at anomaly detection: a message along the lines of "something is wrong, please check." True prediction requires detailed classification, for example "bearing fault, replace within 10 days." This calls for analysis of vibration and sound, and for models trained on a large and diverse fleet of machines.

The operator trust threshold

The decisive issue is trust. The panel reports that operators who see 30–40% false positives abandon the system within a month. The threshold for lasting adoption is put at above 95% true positives and below 5% false positives. The context matters: the article cites roughly $100 billion in unplanned downtime costs in manufacturing.

The figures claimed by Infinite Uptime

The vendor reports the following results, which have not been independently verified:

• 99.97% accuracy across 55,000 prescriptions;
• 95% of prescriptions acted on by operators;
• 175,000 hours of downtime avoided;
• models built over 8 years across more than 1,000 plants in 27 countries;
• more than 600 plants in the process of expanding, and over 300 with 70–80% coverage of critical equipment;
• value to be demonstrated within 3–6 months.

The company is also adding so-called "Verticalized AI Shields": 18 asset categories contextualized with process data, inspection records, maintenance history, OEM procedures, and technician input in text, audio and video. Planned sectors include steel, chemicals, food and beverage, and cement.

The next pool of value: plant redundancy

The panel sees broader value in redundancy. Plants carry an extra 30–40% in costs for standby lines, spare equipment held for 12–18 months, and on-call engineers. An accurate estimate of remaining useful life could allow this redundancy to be reduced at the design stage. Scharinger expects PdM to return to the top three use cases in about five years: that is a forecast, not a fact.

What this means for monitoring and IIoT leaders

For those running monitoring and IIoT projects, the practical takeaways are:

• avoid systems that produce only anomaly alarms: they risk losing operator trust;
• aim for precise, actionable outputs, and measure false positives and prescriptions actually carried out;
• contextualize data with process information, inspections, history and OEM procedures;
• consider models trained on large fleets, not just on a single plant's data;
• build the business case beyond maintenance planning, including spare capacity and plant design.