(ERGO headquarters in Düsseldorf. Source: ERGO homepage.)
For Tim Stettner, Data Analytics Consultant & Strategist at ERGO Group AG (Düsseldorf, Germany), the moment generative AI crossed into the enterprise world was unmistakable.
Tim Stettner, Data Analytics Consultant & Strategist. ERGO Group AG.
“November 2022 was like the iPhone moment for large language models,” Stettner says. “It became clear that this technology was going to change how people interact with information, even inside big corporations.”
That realization set ERGO—one of the major multi-line insurance groups in Germany and Europe and subsidiary of Munich Re (Munich, Germany)—on an ambitious path. By early 2023, the company was laying the groundwork for what would become ERGO GPT, an enterprise-grade generative AI platform designed to empower employees, streamline operations, and unlock years of institutional knowledge often buried in documents and systems. Not that the company was new to turning data into valuable information, using innovative analytics technology.
As a matter of fact, that´s how Stettner himself began his journey about a decade ago, building dashboards that caught the attention of ERGO’s management. He could rely on the extensive analytics and AI infrastructure by SAS (Cary, N.C.). Both companies share a long history of analytics innovation: ERGO has partnered with SAS for over 35 years, beginning with data warehousing and business intelligence and continuing to today’s solutions that use machine learning and AI. “This clearly had opened the door for more innovation,” Stettner says.
Adding generative AI such as ERGO GPT to the existing data solutions, however, presented a different level of opportunity—and complexity. “We saw the potential, but we also knew that using public AI tools directly wasn’t an option in a corporate environment,” Stettner says. “We needed a secure, compliant way to tap into this power.”
Building ERGO GPT from the Ground Up
By January 2023, a team of about 50 specialists at ERGO was designing a secure AI platform using Microsoft Azure (Redmond, Wash.) as the infrastructure. Their goal was to create a system capable of understanding natural language, navigating internal documents, and generating new text—all while remaining safely inside ERGO’s digital walls. To ensure adoption, ERGO trained 80 to 90 “Multiplikatoren”—evangelists or, as the German phrase has it, “multipliers”—employees drawn from business units who can act as AI champions for their colleagues.
“They’re not just IT people,” Stettner says. “They come from business areas and know the day-to-day work. We teach them how to use the technology, how to craft good prompts, and how to help others get value from it.”
By May 2024, ERGO GPT was live for around 28,000 employees and sales staff, assisting with tasks ranging from summarizing long email threads to navigating complex internal policies. One of ERGO GPT’s most transformative contributions has been its ability to untangle ERGO’s dense knowledge repositories. “We have huge amounts of information stored in documents and databases—guidelines, tariff structures, procedures—and finding the right piece of information was always a pain,” Stettner says. “Traditional search tools just don’t cut it for this level of complexity.”
Real-Life Use Cases—and Realistic Caution
He offers a vivid example. Curious about the company’s travel policy, he decided to test ERGO GPT by asking whether a multi-leg trip from Düsseldorf to Vienna to Madrid qualified for business-class airfare. “The document didn’t explicitly list my route or the flight durations,” he says. “But the system combined our travel guidelines with pre-trained public knowledge like airport codes and flight times and came back with an answer.”
The verdict? The trip was right at the threshold of ERGO’s business-class policy—a result that left Stettner both amused and impressed. “It told me that Vienna to Madrid is typically two hours and fifty-five minutes, but with a slower plane, it could be over three hours, so business class might be allowed,” he says, laughing. “That’s the kind of nuanced help you’d never get from a static FAQ.”
Stettner emphasizes how different this is from traditional approaches to corporate knowledge. “A few years ago, you’d build an FAQ with 10 or 20 questions and hope people would ask those exact things,” he says. “Now, people can ask anything. There’s effectively an infinity of possible questions the system can handle.”
The technology has already produced significant benefits such as fast document summarization and insights extraction. Since rollout, 1.5 million prompts have been entered into ERGO GPT. Yet Stettner remains clear-eyed about the limits of the technology. “Summarizing an email chain is helpful, but sometimes the system might miss the underlying sentiment,” he says. “It’s great for a first impression, but you might still want to read it yourself before responding.”
Overall, ERGO takes a human-centric approach to its AI and GenAI cases, which means that people are at the heart of its AI efforts and always have the final say. This is particularly important as AI and GenAI applications do not only offer huge potential in end-to-end processes, but also pose challenges related to, amongst others, data protection, copyright, hallucinations or discrimination.
Rethinking the Way Work Gets Done
One of Stettner’s key lessons from the initiative has been the importance of involving governance bodies from the beginning—including ERGO’s workers council (Betriebsrat) and compliance teams. “At ERGO, all AI applications undergo a comprehensive testing and review process before they are put into productive use, involving all relevant stakeholders such as legal, compliance, IT, risk management and governance bodies,” he says. “This ensures that our employees’ interests are taken into account at all times and lays the foundation for a trusting and therefore successful working relationship.”
Looking ahead, ERGO plans to expand ERGO GPT capabilities even further. A priority for 2025 is increasing the platform’s use of retrieval-augmented generation, or RAG—a technique that allows large language models to combine internal proprietary knowledge with external data to produce context-rich answers. “This project has made us a different kind of company,” Stettner says. “We’re more efficient, more digitally fluent, and better equipped to navigate complexity. We’re not just adding technology to our processes—we’re rethinking how we work.”
By 2030, ERGO plans to invest around €130 million in the development of its GenAI platform. “Generative AI isn’t just another technology—it’s a whole new way of interacting with information,” Stettner says.
SAS Expands AI Portfolio with Digital Twins and Quantum Tools