Representative End User Clients
Representative Automation Clients
Representative Software Clients
ABB, a Gold Sponsor of ARC Advisory Group’s 24th Annual ARC Industry Forum in Bengaluru, was represented by Vinod Ninan, Head of Product & Portfolio Management, ABB Automation. In his presentation on scaling manufacturing with Industrial DataOps and agentic AI, Ninan examined why many promising industrial AI pilots struggle to scale and how manufacturers can build a more repeatable foundation for deploying AI across operations.
Manufacturers are not short of data. Operational systems, historians, enterprise applications, engineering systems, maintenance records, documents, alarms, and increasingly connected assets are producing more information than ever. The challenge is turning that data into trusted operational context that AI applications can use consistently.
Ninan’s presentation can be viewed on YouTube or here:
Why Industrial AI Pilots Struggle to Scale
A successful AI pilot does not necessarily translate into a scalable enterprise capability.
ABB identifies three recurring obstacles: fragmented context, one-off data pipelines, and low operational trust. Individual projects may connect to the data they need and produce useful results, but repeating that engineering effort for dozens or hundreds of applications quickly becomes difficult.
The underlying issue is not simply data availability. An AI model may see a tag, an alarm, or a trend without understanding the asset involved, its maintenance history, surrounding process conditions, operating procedures, or relationships with other equipment.
That missing context becomes particularly important when AI moves beyond analytics and begins recommending operational actions.
Manufacturers therefore need a foundation that allows industrial information to be contextualized, governed, traced to its source, and reused across applications rather than rebuilt for every new project.
Making Industrial Data AI-Ready
ABB positions Genix Industrial DataOps as part of this foundation.
Industrial DataOps brings together the processes required to collect, contextualize, govern, and operationalize industrial data so that it can be reused across analytics, AI applications, and agentic workflows.
This creates what Ninan described as contextual intelligence: operational data is no longer treated as isolated values but connected with engineering knowledge, asset structures, maintenance information, workflows, and other relevant sources.
ABB also argues that industrial AI should not rely on large language models alone. First-principles engineering models continue to provide physical understanding, operating constraints, and domain accuracy, while data-driven models can identify patterns and adapt to changing conditions.
Combining the two can provide a stronger foundation for prediction, diagnostics, and optimization than either approach operating independently.
Moving from GenAI to Bounded Agentic AI
As industrial AI evolves from providing information toward supporting action, governance becomes increasingly important.
ABB uses the term bounded agentic AI to describe agents that can understand a situation, investigate information, recommend an action, and support execution while remaining within defined operational boundaries.
Security, guardrails, traceability, and human oversight are built into this model. The objective is not unrestricted machine autonomy but an environment in which agents can take on more of the investigative and coordination work while people retain control of critical decisions.
ABB Genix Copilot illustrates this approach by bringing together operational data, enterprise knowledge, domain expertise, analytical models, industrial applications, and workflows.
The resulting interaction can move beyond answering a question. An agent can explain what is happening, prioritize possible responses, guide the user through available options, recommend the next action, and support the associated workflow.
ABB Genix Copilot brings together operational data, enterprise knowledge, domain expertise, AI models, and workflows within a governed environment designed to help users move from information access to confident action
From Asset Signals to Maintenance Action
Ninan illustrated the approach through asset performance management.
Consider a centrifugal pump showing abnormal behavior. Detecting the signal is only the beginning. The system must connect that information with operating conditions, alarms, maintenance history, engineering knowledge, and the relationship between the pump and surrounding equipment.
A bounded maintenance agent can then gather supporting evidence, compare the behavior with known failure patterns, retrieve relevant procedures, and recommend an inspection or corrective action. Where required, an engineer can approve the recommendation before the workflow proceeds.
The value is therefore not simply predicting that an asset might fail. It is reducing the time between an abnormal signal and a trusted maintenance action.
A second example involved Genix Datalyzer, which brings together data from industrial analyzers to support fleet visibility, analyzer health monitoring, anomaly detection, emissions visibility, calibration insights, and maintenance decisions.
Agentic capabilities can further help users investigate abnormal conditions, retrieve relevant documentation, summarize likely causes, and recommend the next step while retaining human approval for corrective actions.
Scaling by Pattern, Not by Project
ABB describes the path toward autonomous operations as a progression:
Connect → Contextualize → Predict → Assist → Act → Autonomous.
Each stage depends on the capabilities established before it. AI cannot reliably support autonomous workflows if the underlying industrial data remains fragmented, poorly contextualized, or difficult to govern.
ABB describes autonomous operations as a progressive journey built on connected and contextualized industrial data, analytical AI, generative AI, and agentic workflows
For manufacturers, the larger challenge is therefore not how many AI pilots they can launch but how effectively successful approaches can be reused.
Ninan emphasized the need to scale by pattern rather than by project. Reusable data products, semantic models, agent patterns, workflows, and governance mechanisms can allow organizations to replicate successful approaches across additional assets, plants, and regions without rebuilding the entire solution each time.
Industrial DataOps provides the trusted foundation, while bounded agentic AI adds the ability to investigate, recommend, and support action. Together, they provide a path for manufacturers to move beyond isolated AI experiments toward scalable industrial intelligence.