The impact of AllianzGPT

Officially launched on September 1, 2023, AllianzGPT now serves over 60,000 active users, with the goal of being available to all 158,000 Allianz employees worldwide. Built on Generative Pre-trained Transformers (GPTs), the platform uses deep learning to process and generate human-like text by analyzing vast datasets to understand context, meaning and relationships within language. Unlike public AI models like ChatGPT and DeepSeek, AllianzGPT integrates specific internal content, offering employees tailored responses, document analysis and actionable insights while meeting Allianz’s standards of Ethical AI.

“Tasks that previously took hours or even days, such as document summarization, translation, and complex comparisons, are now being completed within minutes or even seconds,” says Firas Ben Hassan, Deputy Head of Data Science Services and Program Lead of AllianzGPT. “This optimization significantly enhances productivity and allows our employees to focus on higher-value activities.”

AllianzGPT’s strengths include navigating intricate network structures, such as folders and subfolders, in seconds to create engaging, personalized responses for improved communication. It also optimizes processes and assists with communication by crafting individual responses that make customer interactions more dynamic and effective. 

“Classic chatbots were quite boring, and you pretty quickly hit the limits of the questions they could answer,” he explains. “Now, with GPTs, they are more personalized and can also be quite funny in their humor. You can hold human-based types of conversations with it.”

The development of AllianzGPT began in June 2023 and progressed rapidly, thanks to a collaborative effort involving Allianz teams from Germany, India, Spain, and external partners. Despite the swift deployment, significant care was taken to align with compliance and data privacy standards. 

The initial version was based on OpenAI’s GPT-3.5, with subsequent enhancements addressing user feedback, expanding functionalities, and improving integration with Allianz systems. Despite the speed of the rollout, the platform strictly adheres to compliance and data privacy standards.

During the continuous development, Allianz added new and updated models. As of early 2025, AllianzGPT offers OpenAI GPT-4o for chat and DALL-E for image creation. The recent addition of the new DeepSeek model to AllianzGPT enables employees to safely try and assess its capabilities and qualities, using less energy and at a lower cost, helping Allianz to stay on track to reach the group’s sustainability goals.

Just as GPT-4o and DALL-E, DeepSeek on AllianzGPT is a ring-fenced internal version and is separate from the public cloud services of their vendors. All information that’s shared with any of the models in AllianzGPT stays within the Allianz Azure Cloud. The Azure cloud environment, provided by Microsoft, ensures data confidentiality.

AllianzGPT is available in two tiers. AllianzGPT Standard is open to all employees for general productivity tasks. AllianzGPT Premium (also known internally as Ask Me Anything) is a specialized version that supports specific functional areas like audit, Risk Consulting actuarial and many other business areas. As of early 2025, Allianz operates sixty instances of AllianzGPT Premium across Allianz.

While GPTs offer remarkable capabilities, they have notable limitations. These models can ‘hallucinate’ non-facts, confidently presenting incorrect or fabricated information and insisting on the accuracy when queried. They can also fail basic logic tests because their responses are based on language patterns rather than true reasoning.

Firas says Allianz has a three-pronged approach to ensuring that such flaws are not problematic. The first is awareness training for employees to explain potential pitfalls and strategies for validating AI-generated content. The second is ensuring that AllianzGPT is explainable AI, that is, that it provides detailed references, such as document sources and page numbers, for each response for users to cross-check. Finally, disclaimers of potential risks are visible to all users to remind them of the need to verify outputs.

Another key benefit of AllianzGPT is employee upskilling: Allianz employees can use AllianzGPT as much as they want. The tool is freely accessible within Allianz, and the use does not require a license or approval. With this, employees have a low-threshold opportunity to start working with Generative AI in a safe, secure, and compliant environment.

With a growing list of enhancement requests, AllianzGPT is poised for continued evolution. “We’re exploring advanced capabilities, including voice integration and agentic AI, which moves beyond generating content to taking actions,” explains Firas. “However, any new features will be carefully assessed for compliance, ethical considerations and business impact before implementation.”

As AI technology evolves, AllianzGPT shows how companies can adapt this technology to their business sector, delivering value through innovation while upholding ethical and secure practices. “This is not about jumping on the AI bandwagon; it’s about delivering tangible impact and measurable return on investments,” says Firas adamantly. “AllianzGPT represents a step forward in innovation, empowering our teams while upholding the highest standards of responsibility and security.”