From Data to Deployment: How AI Is Taking Shape in the Packaging Industry

In the packaging industry, as in many other industrial sectors, the use of artificial intelligence has become part of overarching digitalisation strategies. The focus is less on the technology itself and more on how information can be structured, made accessible and used reliably.

Stefan Riedl, Partner Manager and Senior Sales Consultant at Cosmo Consult, summed this up in his presentation in the INNOVATIONBOX at FACHPACK 2025: “Companies are information-driven. Keeping pace with digitalisation and understanding and using new technologies is essential to win customers and remain competitive.” Digitalisation, he explained, means transforming analogue information in a way that allows it to be processed efficiently – a prerequisite for the meaningful use of AI.

In the packaging and print environment, AI applications are currently used primarily where large volumes of documents, classifications and master data need to be processed. Large language models, natural language processing and image analysis are integrated directly into existing ERP, supply chain and collaboration systems. Riedl describes how this makes it possible “to speak to software in natural language – without having to program or use codes – and to receive understandable answers.” Typical use cases in this context include assigning customs tariff codes for packaging, working with ECMA catalogues or analysing technical documents.

A key prerequisite for reliable results is control over the data sources being used. “AI delivers better and more traceable answers when it accesses clearly defined databases rather than the open internet,” says Riedl. He also makes it clear that AI does not deliver absolute truths, but rather “the answer the system considers most likely.” Results therefore always need to be reviewed in their professional context and primarily serve to improve access to information and support workflows in the packaging environment.

AI in Packaging Machinery Manufacturing

While the focus in packaging is on information-driven processes, the emphasis in packaging machinery manufacturing shifts more strongly towards connectivity and automation. In his presentation at FACHPACK, Reinhard Schlechter, Segment Marketing Manager for Packaging Machinery at Schneider Electric, clearly positions AI as the next step in development. At the same time, he stresses that although AI is widely discussed, the industry must first address fundamental digitalisation tasks: “Everyone talks about Industry 4.0. In my view, we are still a long way from 4.0. Before we start talking about 5.0, we should first do our homework.”

In the packaging environment in particular, legacy production structures with different generations of control systems and a wide range of manufacturers represent a major challenge. Different control generations, systems and suppliers add complexity. According to Schlechter, the prerequisites for AI applications only emerge once these heterogeneous systems are connected via suitable platforms.

Only then can the full benefits of AI in machinery manufacturing be realised – for example where conventional automation reaches its limits, such as in image-based quality inspection, condition monitoring or software-supported process optimisation. At the same time, AI must remain verifiable in industrial environments. “ChatGPT always gives an answer – even if it’s complete nonsense. We cannot afford that in an industrial context.” Instead, AI results need to be safeguarded through validated libraries and clearly defined data sources.

Pharma: Quality without Compromise

In pharmaceutical packaging, visual quality inspection plays a central role, as it is directly linked to product safety and regulatory approval requirements. Accordingly, the use of artificial intelligence in this field is implemented in a highly controlled and structured manner. Robert Kaiser, Technical Director at Octum GmbH, describes his approach as “an insight into the black box of AI system development in pharmaceutical quality inspection.” The systems used are AI-based image processing solutions that are not built around a single model, but designed as complete systems. “When I talk about an AI system, I mean a complete solution with several AI models integrated and linked by rule-based algorithms,” Kaiser explains.

The development of such systems follows a clear sequence. First, requirements and a technical concept are defined. This is followed by an intensive analysis of the available image data, because “this is the part you have to deal with most during AI system development.” The data are divided into training, validation and evaluation datasets, as well as a separate acceptance dataset. This acceptance dataset is “separated very early in the process” and later serves as an independent basis for system approval

Ensuring the traceability of AI decisions is essential. To support this, Octum has developed a modular toolkit of methods, tools and guidelines that address transparency and explainability: “Why was this decision made, and what was the reason behind it?” AI is thus integrated as a controlled, explainable and qualifiable tool within existing quality and validation processes.

Across all three perspectives, a shared understanding of AI in the packaging industry emerges: it is not an autonomous decision-making system, but a tool for the structured use of data. AI supports people in analysing complex information, automates repetitive tasks and integrates into existing processes. Its value is realised where data quality, system integration and clearly defined use cases come together – and where AI is understood as part of an overarching industrial concept.

 

Author: Alexander Stark, Editor FACHPACK360°