Prescriptive Maintenance in Real Plants: Takeaways from the Industrial AI Summit 2026

On 8–9 October 2026, IIoT World published a Q&A drawn from the session "Unlocking Plant Reliability with Verticalized Prescriptive AI" at the Industrial AI Summit 2026. Eight audience questions were answered by Boris Scharinger, author of *Industrial AI: From Pilot to Profit* and active at Siemens in R&D Excellence, DataOps and Master Data Management, and Karthikeyan Natarajan, CEO of Infinite Uptime, a provider of prescriptive maintenance AI.

This is neither research nor a new dataset: it is an exchange between two points of view and should be read as such. It does, however, offer practical guidance for those who need to take maintenance beyond simple failure prediction. In the article, prescriptive AI is the kind that tells operators which component to replace, which maintenance action to perform, and when.

Data first, or use case first?

Scharinger calls the question "almost ideological": CFOs tend to favour use cases because of return on investment, while CTOs favour a data strategy as a strategic capability. In his view, the data layer is a foundational capability whose cost cannot be charged to the ROI of a single use case.

Natarajan, by contrast, argues that waiting for clean data is "a multi-year tax": most plants will never have a complete history for equipment that is decades old. He suggests starting with the most critical and best-instrumented assets, letting data and the AI layer mature in parallel. He claims a first prescription within two weeks and plant-wide extension within 90 days; these are vendor claims and have not been independently verified.

Brownfield plants: sensor retrofits

For Scharinger, retrofitting sensors, including computer vision, is an excellent way to bring older plants closer to Industry 4.0, and it can also reduce dependence on machine manufacturers. Natarajan points out that vibration analysis catches mechanical degradation, while vision and thermography detect process and quality issues that vibration cannot see: combining techniques broadens coverage.

How to approach diagnosis

Both play down the role of anomaly detection alone, which Scharinger calls "the beginner's start": anomalies still need human review, and classification requires labelled data, even though AI can speed up labelling. Their positions differ on the details:

• Scharinger recommends keeping anomaly/classification AI separate from consultation AI (RAG with LLMs over manuals and known issues).
• Natarajan proposes a sequence: physics-based failure signatures for initial classification, review of every prescription by a reliability expert before it reaches the shop floor, and RAG+LLM only as an explanation and search layer, not as the diagnostic engine.

Replace or predict?

Scharinger offers no universal threshold. He recommends a quantitative decision model for each use case, including:

• the cost of unplanned downtime;
• the remaining useful life lost by replacing early;
• the cost of planned replacement, including scheduled downtime and labour.

He suggests using AI to prototype the model, possibly with Monte Carlo simulation to account for uncertainty. He also estimates that auditing a predictive maintenance algorithm (business hypothesis, risk assessment, simulations) takes about 2–3 people for 6 weeks; at Siemens, he says, this cost was not charged to the business units involved.

On slow adoption, Scharinger points partly to psychology: in low- and medium-risk areas management has grown used to downtime, while in high-risk areas redundancy is already built in. The version of the article we consulted is incomplete, so the rest of this answer and the remaining questions are not covered here.

Key takeaways

For those running plants with ageing machinery, the message is concrete: retrofit sensors, start with critical assets, keep a person in the verification process, and build a cost model before investing. It is also worth remembering that one of the two speakers is a vendor, so the timing promises are commercial in nature.

The topic is closely tied to MES and IIoT: prescriptive maintenance depends on connected machine data and on the ability to feed results back into operations, turning a prescription into a planned intervention.