Agentic AI Use Case: How a Data Agent creates the basis for Digital Commerce

When AI structures product data: ZEG automates with Data Agent

In many industries, the quality of product data determines whether digital business models scale or get bogged down in operational costs. Heterogeneous data sources, historically evolved structures and manual maintenance processes slow down growth, efficiency and time-to-market.

It was precisely this challenge that valantic tackled together with ZEG Zentraleinkauf Holz + Kunststoff eG and Pimcore: The aim was to develop a resilient data and process architecture with which AI can structure, check and automatically process product data on a large scale.

At the heart of the solution is an Agentic AI application implemented together with ZEG, which is based on the open Pimcore platform and was integrated into the existing system landscape by valantic. The aim was to prepare heterogeneous product information in such a way that it could be used for ZEG’s digital sales channels – comprehensible, controllable and with clear quality mechanisms.

In this interview, the project participants share their experiences from the transformation project.

Insights into the Agentic AI use case from:

• Martin Kolakowski, Executive Board, ZEG Zentraleinkauf Holz + Kunststoff eG
• Philipp Krüger, Vice President Product Strategy, Pimcore
• Sebastian Drickl, Managing Partner, valantic

In conversation: AI data management & automation in digital commerce

Sebastian (valantic)

Martin, at ZEG you are working on further expanding your digital sales and retail channels. Why does structured product data play such a central role for you?

Martin (ZEG)

Our product range is very heterogeneous, from flooring and doors to blockboards. This means different suppliers, data sources and attribute logics. In total, we are talking about around 170,000 items that need to be maintained with technically precise product information. Much of this information was previously available in unstructured Excel and PDF files or manufacturer documents, which made it considerably more difficult to compare and use the data digitally.

Sebastian (valantic)

What impact did this have on your processes and daily work?

Martin (ZEG)

On the one hand, the manual effort required for data maintenance was enormous – in terms of time, personnel and money. Secondly, the data quality was very inconsistent. We couldn’t use product data in the way we needed it for digital sales channels. In addition, standards such as BMEcat or E-TIM are not yet used consistently in our industry. This has made the scalability and automation of processes even more difficult.

Sebastian (valantic)

This is exactly where valantic started the project: Not with the isolated question of which AI model should be used, but with the development of a clear target image for the data structure, integration logic and quality assurance. Philipp, why is data structuring a central prerequisite for automation from a technological perspective and what role does an open platform architecture play in this?

Philipp (Pimcore)

The project made it clear that pure AI use without a robust data architecture falls short. Automated processes can only run reliably if data is properly categorized and processed in a technically consistent manner. Specialized data agents can provide targeted support for data preparation and structuring. Pimcore offers an open platform architecture that can be used to flexibly integrate agent-based systems and custom models. The advantage of this is that companies retain data sovereignty.

Key learnings: Data agent best practice with "human in the loop"

Sebastian (valantic)

Together, we developed a data agent for ZEG and established an application for AI data management that combines technical logic, data mapping and quality assurance. Philipp, how does this approach fit into Pimcore and make it resilient for productive use?

Philipp (Pimcore)

In the project, a data agent based on Pimcore takes on tasks such as structuring, attribute assignment, inconsistency detection and optimization tips. For example, it helps with the assignment of attributes, recognizes correlations in documents – even with different spellings – makes inconsistencies visible and provides optimization suggestions.

Sebastian (valantic)

Martin, how do you deal with the automated suggestions?

Martin (ZEG)

Transparency and traceability of AI recommendations are crucial. For us, the “human in the loop” principle is important: the AI system takes over the pre-processing, shows anomalies, justifies decisions and provides recommendations with a confidence value or an assessment of reliability. An employee has the final say and must actively confirm the recommended changes. The Data Quality Score visible in Pimcore enables our editorial team to identify where action is required and which data records need to be checked before publication.

Sebastian (valantic)

How did you gain trust in the AI recommendations and how has the collaboration with the data agent changed over time?

Martin (ZEG)

In the beginning, we checked every AI suggestion manually in order to estimate the hit rate. Gradually, we were able to build up confidence in the figures and significantly reduce the manual work.

Sebastian (valantic)

What are your most important findings from the project and this Agentic AI use case?

Martin (ZEG)

To narrow it down to three takeaways:

1. A strategic objective is essential. At the beginning, we had to clearly define what we need and use which product data for, such as for searches, comparability or digital advice in sales.
2. Humans still have the main responsibility, but their workload is significantly reduced. The data agent performs preliminary work like a detective: it uncovers blind spots in data structures and creates better conditions for data quality and automation.
3. The successful use of Agentic AI means more than just implementing the technology: data structures, processes, governance and integration must be considered and implemented holistically. It virtually “forces” you to set up clean data structures. This may be challenging at the beginning, but it pays off in the long term. In our case, product data was transformed from an operational bottleneck into a strategic basis for digitalization and growth.

Sebastian (valantic)

Philipp, what strategic added value do you see in such projects?

Philipp (Pimcore)

In my view, the project is a prime example of how structured data creates a resilient foundation for automation and digital sales channels. And it shows the potential of data agents if they are embedded in existing processes and system landscapes with foresight and do not work in isolation. Added value is created where structured data, transparent decision-making logic and clearly defined quality mechanisms work together.

Sebastian (valantic)

At valantic, we see it as our task and core competence to create precisely this basis for digital innovations and to provide holistic support for transformation projects: from data strategy to integration, automation and continuous optimization for effective and sustainable use.

Thank you for taking the time to share your experiences!

From an Agentic AI use case to sustainable business impact

The project with ZEG is an example of what is important for Agentic AI applications in practice: real added value is created through the interaction of a clean data structure, well thought-out system architecture, comprehensible decision-making logic and clearly defined quality processes.

For companies, this means that if you want to use agentic AI effectively and develop a viable business case from it, you need to talk not only about models, but also about processes, integration and controllability – regardless of the respective application scenario.