Agents for the corrugated board industry

For decades, WITRON’s Corrugated Packaging Management System (CPMS) has been the benchmark solution for many market-leading companies in the corrugated board industry. Today, more than 10,000 users across over 250 corrugated plants worldwide rely on the system. CPMS spans the entire value chain - from order entry to production planning and final delivery. Now, the development team led by Dr. Stefan Bauer is taking the next major step: the integration of artificial intelligence.

Listen to the podcast here: Agents for the corrugated board industry

Before addressing specific use cases, it is crucial to understand what generative AI truly excels at — only then can we leverage its strengths to full effect.

Dr. Stefan Bauer, Business Unit Director CPMS, WITRON

For many years, CPMS has utilized classical optimization methods. “Traditional AI has long been embedded in numerous areas of the system, as we optimize business processes using heuristic methods,” explains Stefan Bauer, CPMS Business Unit Director at WITRON. These intelligent algorithms provide a proven foundation, especially in production planning and route scheduling. With the introduction of Generative AI (GenAI), WITRON is now equipping CPMS with a new level of capability. However, Stefan Bauer underscores an important point: “Before addressing specific use cases, it is crucial to understand what generative AI truly excels at — only then can we leverage its strengths to full effect.”

Large Language Models (LLMs) open up entirely new ways of processing language, text, images, and statistical data. Stefan Bauer and his team performed a structured evaluation to determine:

• Which business processes can truly benefit?
• What level of integration effort is required?
• What specific value does the functionality deliver to customers?

In industrial environments, near-perfect accuracy has traditionally been the benchmark. Generative AI, however, operates differently. “No one expects AI to be correct in every instance - yet in most cases, it provides substantial support,” Stefan Bauer explains. He refers to the well-known 80/20 principle: In many management strategies, reaching 80 percent automation is considered sufficient when the efficiency gains clearly exceed the residual effort. The key is a controlled, well-orchestrated integration of the technology into existing processes.

Consistent presentation of information

WITRON deliberately adopted a pragmatic approach from the very beginning. “We wanted to gain experience first, which is why we started by enhancing documentation support in multiple international languages,” Stefan Bauer explains. CPMS provides extensive “Books Online,” offered mainly in English so far. With LLM technology, WITRON can now automatically translate this content into our customers’ native languages – in a consistent and context‑aware manner.

Documentation support is just the beginning. Our ambition is to evolve the bot into an agent that actively works for the user.

Dr. Stefan Bauer, Business Unit Director CPMS, WITRON

But the initiative goes far beyond translation. “We merge various documentation sources, enabling the AI to identify the corresponding terminology and produce a consistent and coherent presentation of the information,” says Bauer. The solution is integrated into the platform as a chatbot. Users can interact with the system in their own language and receive context-specific answers related to configuration and functionality. From a technical perspective, WITRON relies on a Retrieval Augmented Generation (RAG) architecture supported by sophisticated prompt design. The system selectively draws on validated documentation sources, significantly reducing the risk of hallucinations. For WITRON, the chatbot represents only the first step. “Documentation support is just the beginning. Our ambition is to evolve the bot into an agent that actively works for the user,” Stefan Bauer emphasizes.

CPMS processes three central entities:

• Customer order
• Production order
• Delivery order

While production and shipping already rely on established algorithms and heuristics, WITRON sees particularly strong automation potential in customer order management. 

APIs for the agent

In many corrugated plants, a significant share of the order entry process is still handled manually. Despite the availability of EDI and web-based ordering systems, one to two thirds of all orders are still submitted as PDF documents or e‑mail requests. This is where the AI agent comes into play. “The bot can analyze international order documents, identify key terms, and create an order within the system,” explains Stefan Bauer. Importantly, the agent goes far beyond merely extracting data.

Through an existing API layer, it interacts directly with the CPMS database:

• It checks whether the customer exists
• It validates delivery addresses
• It verifies product master data and relevant status information
• It compares information with existing datasets

The user reviews the proposed result, makes adjustments via text or voice input, if necessary, and confirms the order. The result is a hybrid workflow that combines AI-supported analysis with human approval. The business benefits are substantial: Reduced manual effort, accelerated processing times, and lower error rates - while at the same time easing the workload of the sales back office.

WITRON’s AI strategy is built on an open system architecture. CPMS serves as the digital hub across the entire value chain. “All systems - Yard Management, CRM, Finance - must be integrated. This level of connectivity is orchestrated through a digital hub like CPMS,” explains Stefan Bauer.

Customers already use the API layer today, for example to support:

• Delivery date inquiries
• Inventory availability checks
• Third-party pre-sales applications

AI agents access the same infrastructure and extend it with intelligent automation. WITRON currently relies on ChatGPT-based LLM technology. “The features we require are very well covered,” says Stefan Bauer.

The decision was made deliberately in favor of a long-term, stable partnership within the Microsoft ecosystem. “We want to focus on a core technology that combines generic AI functionality with a strong and substantial partner,” emphasizes Stefan Bauer.”

WITRON regards AI as an innovation investment in its early stages. “In the beginning, AI cannot be calculated separately,” explains Stefan Bauer. As functional depth grows and direct customer value increases, the company plans to introduce add-on licenses for specific AI features. As a result, AI evolves from a simple technological add‑on into a distinct and scalable business model.

Agents and forecasting

Stefan Bauer and his team are exploring additional automation scenarios over the long-term. “The more functions the agent can access, the more we can combine them into workflows,” says Stefan Bauer. The system is expected to identify usage patterns and automatically perform repetitive operations. If users frequently apply the same filters or consistently select data in a particular way, the system will be able to apply this logic autonomously.

WITRON also sees future potential in analyzing time-series data from machines and production processes. “We are not active in that area yet, but we clearly see the potential for improving forecasting in the future,” explains Stefan Bauer. A combination of agent-based architecture and specialized forecasting models is a logical next step.