The manufacturing operation is becoming smarter with data, edge computing, and agentic AI working in sync. The industrial CIO stands at the most significant turning point in the history of enterprise technology leadership.
For years, the role was defined by a singular expectation, i.e., to ensure 100% uptime for IT infrastructure and applications while operating as a cost center buried deep in the organizational chart. That era is over.
Now, the CIO community has evolved into forward-looking core strategic leaders navigating IT as a profit center that aligns with people, process, and technology with their respective organization’s vision and objectives.
In an interaction with CIONOW, Sumit Duttagupta, CIO at Haldia Petrochemicals, discussed the transition in the CIO’s role from implementing Sensors to Boardroom data architecture to deploying an Agentic AI system.
In your view, what specific behaviors or decisions signal that a CIO has moved from support function to profit-center leader in an industrial setting?
The industrial CIO has shifted from a support and cost-centric mindset, ensuring 100% uptime of IT infrastructure and applications, to a forward-looking core strategic leader driving IT as a profit centre and aligning people, process, and technology towards the company’s vision and objectives. In the present fast digital and smart manufacturing world driven by AI, it’s imperative for CIOs to be in sync with new technologies and manage change within the organization, leading to agile digital outcomes.
Where do you see the biggest gap between hyper-automated smart factories (Industry 4.0) and human-centric, resilient ecosystems (Industry 5.0)?
Moving beyond Industry 4.0 means entering Industry 5.0, a paradigm shift that elevates the industrial CIO from a builder of hyper-automated smart manufacturing to a champion of human-centric, resilient, and sustainable ecosystems, guiding the focused deployment of GenAI/Agentic AI/Cobots. This will require greater emphasis on upskilling people so they can collaborate more effectively with the AI world.
While Industry 4.0 focused heavily on machine-to-machine communication, data collection, and minimizing human intervention, the next phase pivots toward human-machine collaboration, powered by the AI ecosystem.
How do you ensure security and data integrity while enabling near real-time flow from OT to IT?
This is very crucial for any smart digital outcomes and is a very foundation for an agile AI framework. The sensors-to-boardroom connection is a key outcome of this plant and business interface of datasets in a near real-time environment.
A strong framework is needed to ground with a strong master data foundation and ensure seamless connectivity and interface in a secure environment. ISA 95 is one such methodology that can be deployed as a guiding principle to have a seamless data ecosystem. The ecosystem of your instrumented data, IIOT data, DCS/SCADA data, historian data, MES data, and interfacing it with business data from ERP/CRM/SCM systems needs a sound architecture that is sustainable and scalable. The essence is to find the right technical framework that can host this seamlessly. We use the MCUBE AI-based Analytics Platform.
Taking an example of what we have gone through in our implementation in the petrochemical industry, where the manufacturing process is continuous in nature, is worth mentioning to get a better practical insight through these five steps:
Step 1: Integrate your DCS/SCADA/IIOT data to your historian system only (ensure only a read-only view from the historian for interfacing with IT layered systems). Deploy edge computing devices at the plant level. Security at the physical level is of paramount importance.
Step 2: Ensure that your core manufacturing process (bare minimum critical to operations) is fully automated, which impacts Operational Integrity and ensures safe and reliable operations and maintenance. Create a manufacturing execution system stack.
- Automate the manual Logbooks into E-Log to ensure better visibility across your shift operations
- Automate the manual Field Round Logbooks (LLF) to ensure sync-up with field and control room operations with better integrated insights
- Rationalize and automate your DCS Alarms by deploying an Alarm Management System
- Automate the Lab Sampling through a Laboratory Information Management System (LIMS) with online analysers
- Deploy real-time optimizers for Key process plants and ensure near real-time recommendations
- Reconcile your Production Data and ensure a daily Hydrocarbon Balance from Load Port to Bagging and Flare Management.
- Ensure near real-time monitoring of Power and Utilities with an overall complex-wide balancing
- Ensure a cause & effect relationship for effective process monitoring of your plants
Step 3: Once the above items are automated, develop an “Integrated view of your operations – right from Feedstock loading to Process Plant operations to bagging” – Create an “ONLINE PLANT INFORMATION SYSTEM” with a contextualized view developed in sync with your business data from ERP / 3rd Party systems.
Step 4: Now plan to monitor the following:
- Track Lost Opportunities in Dollar value and ensure action tracking
- Monitor your HOURLY & DAILY EBITDA and give a sense to your plant & senior stakeholders – how they are performing
- Ensure you stay focused for effective planning and a supply chain model based on this integrated view, and arrive at an Integrated Margin Management to forecast scenarios.
Step 5: Now, once you are on track for a certain period, explore ways to utilize the analytics workbench and data and information foundation created to harness the potential of Gen AI and deploy Agentic AI for further autonomous automation.
What does your evaluation framework for agentic AI look like, and how do you continuously test that agents are safe, accurate, and valuable?
Today, everyone is talking about AI, Agentic AI and automation through the lens of digital or business transformation. The real practical value of AI lies in people transformation, with roles in an enterprise as the key focus. Let me illustrate this in detail:
Legacy enterprises ran on siloed, disparate databases. Three key AI changeovers unified them into a common data fabric, enabling Agentic AI to reason and act across the enterprise.
Let us understand what the three key changeovers are that this AI world has brought.
First: AI Infra
- Higher processing power (GPUs/TPUs)
- Scalable cloud data platforms
- Consolidation of siloed databases
- Foundation for unified data fabric
Second: LLM → SQL
- Natural language to SQL translation
- Conversational data access for all users
- Eliminates dependency on schema experts
- Faster, self-service insight generation
Third: Transformers
- Handle unstructured data natively
- Unify text, images, logs, PDFs, sensors
- Vector embeddings as common language
- Enables true cross-domain agentic reasoning
Agentic AI systems go beyond traditional automation — they perceive, reason, decide, and act independently within defined guardrails, continuously learning from their environment.
Perceive & Reason
- Real-time sensor fusion
- Anomaly detection
- Context understanding
- Multi-source data synthesis
Decide & Act
- Autonomous decision-making
- Process parameter adjustment
- Workflow orchestration
- Proactive risk mitigation
Learn & Adapt
- Continuous model refinement
- Feedback loop integration
- Performance self-assessment
- Predictive foresight
So, from a common database to autonomous execution, there are two phases that move the enterprise from talking to doing.
Common DB → GenAI – Shift from retrieving records to generating insights.
Gen AI → Agentic AI – Move from talking to doing — agents that reason and execute.
Finally, this AI and Agentic AI world will lead to an autonomous Agentic World where the People & Roles will be key to success. The information ecosystem should drive on these following 3 foundational pillars:
- Tool Integration — secure APIs
- Security & Governance — strict RBAC
- Continuous Evaluation — eval pipelines
What controls do you put in place to mitigate data poisoning, adversarial AI attacks, and automated ransomware in an OT context?
In today’s AI-driven world, the OT/IT interface has dramatically expanded the cyberattack surface.
Historically, factories relied on air-gapping (physical isolation) for security. Today, AI models require continuous, bidirectional data streams between sensitive plant floors and cloud environments. This makes traditional perimeter-defence models obsolete.
The fusion of AI with IT-OT convergence introduces three critical threat vectors: Data Poisoning, Adversarial AI Attacks, and Automated Ransomware. To secure this pipeline without choking the data flow needed for AI, industrial CIO/CISO must implement an adaptive security framework based on IEC 62443 standards and Zero Trust Network Architecture.
What happens when a digital initiative meets its technical targets but fails to move the business KPIs you committed to?
The measurement of digital transformation outcomes is very critical. There is a saying that “unless you measure, the improvement opportunities are never realized”. The outcomes need to be measured both in quantitative and qualitative terms. In most cases, outcomes are measured based on certain baselining. The outcomes are monitored on a regular basis based on key operational KPI’s and values vetted by the finance team.
What is the future of technology-led industrial transformation?
The future of technology-led industrial transformation in an AI world moves beyond discrete and isolated smart factories into a reality of autonomous, self-healing, and universally hyper-connected value chains.
As we look beyond, the role of the industrial CIO shifts from managing digital tools to orchestrating a fully integrated cyber-physical organism.
To realize the vision of the factory floor as a fully integrated cognitive infrastructure, the industrial CIO must stop treating IT, OT, and AI as separate layers. Instead, they must blend them into a single, self-healing system where digital software loops and physical hardware actions are completely interdependent.
In this ecosystem, data acts as the nervous system, edge compute serves as localized muscle reflexes, and agentic AI functions as the central brain.
