How to move past IT entanglement to unlock AI-based Asset Management
This article was previously published in the March 2026 issue of APUEA magazine.
Modern energy facilities are more automated and interconnected than ever. From rooftop solar and district cooling in bustling Asian cities to large-scale WtE plants and grid-scale renewables, these facilities face unique challenges: intermittency in solar and wind output, rapid asset aging in humid/tropical climates and cascading risks across urban grids. Yet they remain heavily dependent on scarce high-skilled personnel. In this high-stakes environment, where technical problems can cascade into unplanned outages, environmental non-compliance, financial penalties and reputational damage, can AI help transform how we manage the physical assets that power our world?
When companies explore AI for asset management, conversations default to predictive maintenance. Let’s step back. According to ISO 55000:2024, Asset Management is the “coordinated activity of an organization to realize value from assets.” Maintenance fits within this framework as an operational enabler. Predictive maintenance is a specific technical activity that may or may not suit a given asset. True AI-powered asset management is far broader. Roughly speaking, predictive maintenance is the domain of the technician and machine supplier; asset management, that of the manager and plant owner.
Energy companies are well-resourced, with deep operational experience and robust IT infrastructures. They have long relied on EAM and CMMS systems, often supplemented by in-house mobile apps and portals, yet remain dependent on countless Excel sheets and paper records. They have experimented early with IoT, digital twins and now AI. Yet many have observed how legacy IT and entrenched practices obstruct real progress.
This article draws on four decades digitalizing asset management, two in Asia.
Context Matters
Despite the promises, few AI attempts by asset owners advance beyond pilot stage. Makers of standardized equipment like turbines have developed mature predictive maintenance approaches. But from the plant owner’s perspective, scaling remains elusive. Variability in equipment, site conditions and integration challenges complicate deployment.
Most efforts focus solely on real-time sensor data. Alerts are generated, but what should a technician do? A vibration alert doesn’t distinguish imminent failure from sensor drift. Without understanding the failure mode, cause and past interventions, the alert remains untrusted. False positives compound the problem: teams that respond to nothing wrong learn to ignore the system.
Black-box AI deepens distrust. Teams cannot verify recommendations and when predictions prove wrong, there’s no path to understand why. Experiments with LLMs (and most energy companies have tried plugging ChatGPT into IoT data) introduce hallucinations that seem sensible to non-experts but make engineers wince.
This brings us to the core issue: context matters more than raw real-time data. Relying solely on sensors fails long-term due to noise, false positives, cost and incomplete failure mode coverage (corrosion, human error, lubrication issues). Real-time inputs need rich historical and operational context.
This holds for predictive maintenance and even more for AI support of asset management. If maintenance fails without context, strategic asset management fails without an even broader context: capital plans, risk registers, financial constraints, regulatory obligations, organizational objectives. AI is not the starting line, it is the finish line, achievable only after the data foundation is sound.
The Three Bottlenecks
Three bottlenecks prevent AI-powered asset management:
-
Incomplete as-built data
Plants inherit poor databases from construction, gaps in hierarchies, naming conventions, failure history. This undermines every strategic decision: risk assessments, life cycle analysis, investment prioritization. -
Data silos
Information scatters across ERP, CMMS, Excel, paper, in-house apps, IoT platforms. The result? Painful mobile apps no one uses, broken sync, half-finished Power BI dashboards reliant on Excel exports, shared Google Sheets and (most tellingly) unauthorized WhatsApp use to request work orders with pictures and voice messages. Field data (closest to physical reality) is often missing. This contradicts ISO 55000’s requirement for coordinated activity grounded in verifiable field data to prove strategy implementation and drive continuous improvement. Adding AI on top of fragmented systems only amplifies these existing issues, creating more noise without real insight. -
Incumbent lock-in
Strong legacy suppliers and entrenched vendor relationships create patchwork environments where new entrants face insurmountable procurement and IT compliance hurdles. IT teams often block integration to protect existing systems, while procurement processes favor incumbents and demand exhaustive compatibility checks that make meaningful change impossible. For asset management, this means organizations struggle to introduce innovative solutions without massive disruption or outright rejection.