In a world where digital transformation is no longer a buzzword but a necessity, organizations are increasingly turning to hyperautomation to streamline operations, boost efficiency, and unlock new business value. At the 2025 European Power Platform Conference (EPPC), Cegeka’s consultants Katinka van Luijn and Elwin van der Laan shared a compelling case study of how we partnered with Bieze Food Group to do just that: start from Azure AI and then tuned the process with agents.
Curious for more? Continue reading this blogpost in which we highlight the session and case!
What is Hyperautomation really?
Hyperautomation is more than just automating repetitive tasks. It’s a strategic, business-driven approach to identifying, vetting, and automating as many business and IT processes as possible. It involves orchestrating a suite of technologies: RPA, low-code platforms, AI, and process mining, to create end-to-end automation that’s intelligent, adaptive, and scalable. In this case, the hyperautomation process spans multiple years, ensuring a long-term and evolving approach.
As Katinka explained, “It’s not about doing more, faster. It’s about doing the right things with the right tools.” Hyperautomation is iterative by nature. It’s a cycle of continuous improvement that includes six key stages:
- Centralize
- Govern
- Empower
- Innovate
- Orchestrate
- Improve
The challenge: food label chaos
Bieze Food Group, a European platform for fresh and frozen foods, faced a growing challenge. With every product delivery came a flood of food labels, each containing critical information like nutritional values, ingredients, and packaging details. This data needed to be documented and submitted to the GS1 portal, but much of it was missing or inconsistent. Manual processing was time-consuming and error-prone.
The opportunity? Use AI to extract relevant data from these labels and automate the submission process. But the complexity of food packaging, varying formats, languages, and symbols, meant that traditional OCR solutions weren’t enough.
The first step: Azure AI and OCR
The initial solution used Azure AI’s OCR capabilities to extract data from food labels. This information was processed through Power Automate and stored in Dynamics 365 and the GS1 portal. While this setup achieved decent accuracy (85–90%), it required extensive manual training and struggled with new or unstructured label formats.
As Elwin noted, “OCR works great for structured documents like invoices. But food labels? They’re all over the place. Different layouts, fonts, icons: it’s a nightmare to train for every variation.”
Enter agentic AI: a smarter, more flexible approach
To overcome these limitations, the team experimented with agentic AI, specifically, using prompt-based models to extract structured data from label images. Unlike OCR, which relies on fixed field structures, agentic AI could interpret symbols, perform contextual reasoning, and adapt to new formats without retraining.
The results were impressive. With refined prompts, the agent achieved near-perfect accuracy, even recognizing icons like microwave symbols and Nutri-Scores. It could output structured JSON data, calculate nutritional values per portion, and handle both text and image inputs.
This shift from static OCR to dynamic agents marked a turning point. “We didn’t need to train a model anymore,” said Katinka. “Just refine the prompt, and the agent learned on the fly.”
Orchestrating the flow
With the agentic model in place, the team integrated it into Power Automate flows, replacing the Azure OCR step. This allowed for seamless orchestration of label recognition, data extraction, validation, and GS1 submission, all within a single automated pipeline.
The flexibility of the agentic approach also empowered non-technical users to contribute. With no need for model training, business users could tweak prompts and test outputs, shortening feedback loops and accelerating innovation.
From labels to sustainability insights
But the journey didn’t stop there. Recognizing the environmental cues embedded in food labels, (ingredients, portion sizes, packaging materials), the team explored how this data could be used for CO₂ footprint analysis.
By mapping label data to external databases like RIVM and Ecoinvent, the agent could estimate emissions per product. For example, a paprika-chili tortilla wrap with 45% vegetables was found to emit approximately 0.43 kg CO₂-equivalent per 200g portion. Surprisingly, paprika contributed more to emissions than the wrap itself.
This opened the door to new use cases:
- Comparing suppliers based on environmental impact
- Flagging high-emission ingredients
- Suggesting lower-emission alternatives
- Combining sentiment analysis with sustainability metrics
Lessons learned and looking ahead
The Bieze Food Group case illustrates the power of hyperautomation when paired with agentic AI. It’s not just about efficiency, it’s about adaptability, intelligence, and continuous improvement.
Key takeaways from the project include:
- Start with structured thinking, not just tools
- Use AI where it adds real value—especially with unstructured data
- Empower business users with low-code tools and prompt-based models
- Treat automation as an iterative journey, not a one-time project
- Look beyond process optimization to broader impacts like sustainability
As Elwin concluded, “AI and hyperautomation are not just tools, they’re enablers of innovation. They help us build solutions that evolve with the business.”
Final thoughts
In a rapidly changing world, organizations need more than automation: they need intelligence, agility, and vision. The partnership between Cegeka and Bieze Food Group shows what’s possible when technology meets strategy, and when innovation is driven by real-world challenges.
Ready to enable hyperautomation?
Hyperautomation isn’t the future. It’s happening now, and it’s reshaping how we work, think, and build. Curious to learn more about hyperautomation? Discover how you can enable hyperautomation with Power Platform.
Curious to get a more insights from the technical point of view of Hyperautomation? Then watch our five part videoseries Zero to Hyperautomation and go from laying your foundation for hyperautomation to strategic automation.