Council Post: From Data To Decisions: How Context Can Turn Industrial AI Into A Scalable Advantage
Heiko Claussen is Chief Technologist at Emerson’s Aspen Technology business, leading its AI research and technology strategy.
getty
As industrial organizations continue to accelerate AI adoption, many are hitting the same wall: More data isn’t translating into more value. Despite years of investment in collecting operational data, much of it remains siloed, taken from narrow operational regimes, misaligned and stripped of context. The problem isn’t a lack of data; it’s a lack of the right data and an understanding of what it all means.
Without the context required to generate meaningful insights, AI can't scale reliably across the enterprise. Context is the set of semantics and relationships that tells AI, or any relevant enterprise application, what a measurement means, when it's valid and how it relates to the current operating state.
A key misconception is that AI systems inherently understand the environments they operate in. In reality, they don't. Rather, they infer context based on the data they're given. In generic AI systems, this inference is typically driven by statistical patterns, historical correlations and heuristics. However, in industrial environments, that approach can introduce risk, as the physical systems, safety constraints and nonlinear behavior require careful considerations.
Consider a simple example: A pump reports an RPM of zero. A generic AI system may infer that the pump is offline and disregard signals from other sensors. However, the real issue may be a faulty RPM sensor (while the pump continues operating normally). When the model’s assumptions don't match the asset’s true operating context, the recommendation can be technically plausible yet operationally wrong.
Without visibility into the underlying assumptions driving the conclusion, operators may gain false confidence in an incorrect assessment, which can lead to missed anomalies or inappropriate actions. Industrial operations regularly deal with calibration issues, time misalignment between sensors and siloed systems. If they don’t have the proper context, AI models may be comparing signals that were never meant to be analyzed together, leading to errors in interpretation.
AI purpose-built for industrial environments addresses this challenge by combining data insights with domain expertise and asset-level knowledge. Rather than defaulting to the most statistically likely conclusion, it also evaluates a broader range of possible states and uses relationships between signals and known operational ranges to determine what's really happening.
Data As The Enabler
Many companies have implemented industrial AI without a perfect data foundation, primarily by leveraging AI solutions backed with strong domain expertise and first-principles models. However, as organizations expand their AI use beyond isolated applications and into real-time, field-level decision making, the limitations of fragmented or low-quality data become clear. The absence of contextualized field data makes it difficult to truly scale AI across operations, sites and geographies, so organizations are effectively "flying blind," unable to see what's happening now or act with confidence.
In this sense, data determines whether AI is effective in operational conditions. Contextualized data elevates AI from hand-built models to scalable, enterprise-level capabilities because the AI has context for the asset’s true type and operating conditions.
By grounding models in high-quality operational data, companies can continuously adjust for changing process conditions such as fouling, catalyst deactivation or other nonlinear effects that are difficult to capture with first principles alone. This is one way operators are closing the "simulation-reality gap"—the disconnect between theoretical models and real-world asset behavior.
Contextualized historical data can also boost the accuracy and reliability of predictions related to specific equipment failure modes, energy demand and renewable generation. In each case, the underlying driver is data that's aligned, contextualized and connected, enabling AI to interpret signals correctly and support decisions that remain valid as conditions change.
From Data Foundation To Enterprise Value
Beyond enabling AI, a data management platform can unify data across edge, on-premise and cloud environments. It becomes a data foundation that can help standardize data access, context and governance so organizations can connect systems more easily.
A unified data fabric can also improve the cybersecurity posture of industrial operations, where more connections create more opportunities for attack. Traditional point-to-point integrations often rely on numerous protocols, interfaces and custom connections that must each be maintained, monitored and secured throughout their life cycle. However, a contextualized data fabric reduces the engineering effort to onboard new assets while making it easier to apply consistent security policies, monitor data flows and manage access controls across the enterprise.
With access to the right data, new levels of automation become possible. Asset templates, for example, can define the specific sensor relationships required to detect failure modes for certain equipment. This enables AI systems to automatically identify which models to deploy, build them dynamically and maintain them over time.
A contextualized data foundation allows the broader incorporation of AI through the use of enterprise operations platforms. By integrating data from across the enterprise into a unified, contextualized framework, these platforms enable the move beyond isolated AI use cases to truly connected operations. This is where the greater value of AI emerges—not just in individual models but in the ability to deploy, adapt and scale insights across the business.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?