Manufacturers need to prioritise data context before harnessing AI in factories

Manufacturers risk pouring money into artificial intelligence without seeing meaningful returns unless they first give their factory data structure and context, according to Mario Morales, the former global data and AI products lead for Manufacturing & Procurement at PepsiCo.

Speaking at a Mobile World Live Unwrapped Digital Industries keynote, Morales argued that better connectivity is only the starting point in the digital transformation of factory operations. Modern prod...

uction systems can generate a flood of readings on temperature, vibration, humidity and equipment faults, but those signals are often detached from the operational realities that give them value. Without that extra layer of meaning, he said, businesses struggle to know whether a measurement relates to a particular product, batch, line or shift.

His warning reflects a broader challenge now confronting industrial AI projects: raw data is not the same as usable data. Specialists in manufacturing data contextualisation often describe the problem in terms of identity, location, semantics, quality and operating state, arguing that these relationships need to be established at source if information is to be interpreted reliably across different applications and sites. In practical terms, that means linking a sensor reading or machine signal to the asset, production run and business process it actually represents.

Morales said every signal has to be tied to something tangible on the shop floor, and that companies also need consistent definitions, trusted measurements and access for the people expected to act on the information. In his view, that requires the right platform design, proper data governance and a working model that connects technology, process and operations.

The scale of the issue is particularly acute in factories with ageing equipment. In consumer goods manufacturing, Morales noted, machinery can be decades old, and the same product or process may produce slightly different outputs depending on the line or plant involved. That makes it harder to build AI models that can be copied from one site to another without extensive rework.

Industry commentary increasingly supports that caution. McKinsey has said manufacturers face persistent barriers from broken sensors, missing data points and incompatible systems, while other industrial technology specialists argue that clean data alone is not enough for AI readiness. Standardisation and contextualisation, they say, are what make factory information useful at scale, especially when applications must compare data across multiple plants, production lines or business units.

Morales suggested the current excitement around AI can obscure how much work is still needed before industrial systems can benefit. Too often, he said, businesses assume they can simply feed factory data into a model and extract insight, while underestimating the effort required to make that data meaningful in an industrial setting. He framed the challenge less as a technical limitation than a cultural one, saying the technology itself is capable, but companies must adapt their business processes to take advantage of it.

For organisations beginning that journey, his advice was straightforward: organise the data, add context and build the foundations before expecting AI to operate at scale. In manufacturing, he implied, intelligence starts not with the model, but with the meaning attached to the data going in.

Source: Noah Wire Services