AI and Data in Industry: Moving from Pilots to Scale

AI is everywhere in the conversation, but in industrial operations, the reality is more nuanced. Many companies are still figuring out how to move from promising AI pilots to solutions that work reliably in production and can be scaled across the business.

In this interview, we look at where industrial AI stands today, what it takes to scale, and why the data foundation matters just as much as the AI itself.

Meet Javier Lozano Díaz, Digital Transformation Lead at Orange Business

How is AI changing the industrial environment, and where are we right now?

Javier: Right now, we are earlier than the noise suggests. Most industrial companies are still investigating how to enable AI in the best and most efficient way: which architecture, which data foundation, which governance model, which use cases to start with.

That is not a failure. It is the normal shape of a technology whose enabling layer, industrial data, is still being built. But it does mean the distance between companies talking about AI and companies running AI in production is still wide.

What is genuinely changing is not that AI replaces roles. Every job in a plant is a bundle of tasks, and AI redistributes that bundle.

Take a maintenance technician: an assistant can surface the right procedure and the history of similar failures in seconds, so that task is augmented. An agent can pre-classify alarms and draft the event report, so that task is automated. And nobody is going to change a bearing remotely, so that task remains untouched.

At the same time, new tasks appear: validating what an agent proposes, curating the data model, supervising the quality of automated decisions.

Industry has an advantage that purely knowledge-based sectors do not. A large share of the value is created in the physical world, so the people closest to the process become more valuable, not less, once they have better information in their hands. The real constraint is not model capability. It is data readiness, trust and adoption in operations.

And the skill profile shifts with it. Value moves from knowing the plant by memory towards being able to frame a problem well and challenge the answer a system gives you. That has consequences for hiring, training and team design, and most companies have not started that conversation yet.

What does it take to scale AI in industrial companies?

Javier: I think the first thing is knowledge of operations. Technology will be less and less of the barrier. The same thing happened with code: writing it used to be the exclusive territory of people who had learned to write it, and that is no longer true. What does not get democratised is understanding how a plant actually works, its processes, its constraints and its priorities. That is where the differentiator sits.

The second is less isolated focus on AI and more focus on the industrial data foundation. Before scaling AI, IT and OT have to exchange data in a way that is contextualised, governed and reusable across lines, systems and plants. Without that foundation, a use case stays a demo.

And then there is technology, people and adoption. Sales, service, production, maintenance, quality and finance have different objectives, but they can all benefit from the same data. Breaking those silos requires the right architecture, clear processes, clear ownership and real adoption by the people who will use the solution.

So the approach needs to be ambitious, but pragmatic. Far more is possible with AI today, yet the distance between a good idea and an industrial solution running sustainably is still underestimated. We keep seeing companies stuck in pilots and PoCs that never scale.

What makes a good Data & AI strategy different?

Javier: I would say the first thing is to connect IT, OT and business from the start. Data & AI should not be developed department by department. Production, maintenance, quality, laboratory, controlling, purchasing, sales and service create far more value when their use cases share data and context.

The opening question is a simple one: what information from one area would let another make better decisions or automate work?

Designing that view up front is much cheaper than reconnecting solutions that were built as independent silos. Reworking what already exists costs more time, more money and more political capital inside the organisation.

The other important point is not to copy the competitor’s strategy. Copying the competition means starting from behind. Take inspiration from the market, of course, but the architecture, the priorities and the roadmap have to answer the real context of your company: processes, existing systems, maturity, people and business objectives. That is where the reason for a customer to choose you appears.

Data & AI should not be developed department by department. Production, maintenance, quality, laboratory, controlling, purchasing, sales and service create far more value when their use cases share data and context.

What are the most tangible AI applications in industry today?

Javier: It depends on the sector and the maturity level. Energy, pharma, logistics and food do not prioritise the same things. But three come up across all of them.

The first is industrial knowledge assistants for operations, maintenance and engineering: natural language access to technical documentation, procedures and incident history.

Then there are agents that analyse downtime, alarms and events, supporting root cause analysis and cutting diagnosis time.

And the third is real-time traceability and contextualisation, from shop floor data to business data, without manual reconstruction, so decisions happen faster.

What has changed recently?

Javier: I would say the biggest change is the cost of creating a new use case.

Until now, every use case required building and manually extending its own data dictionary. Architectures such as Industrial Data Platforms, Unified Namespace and contextualised industrial data models allow companies to establish a shared semantic layer that can be reused across applications in real time.

And the direct consequence is that every new analytics, automation or AI application you put on top costs less and arrives sooner.

---

Interested in exchanging ideas on industrial data and AI?
Connect with Javier Lozano Díaz on LinkedIn and continue the conversation.