‘AI doesn't come for free’

Since summer 2024, we have been reporting regularly here on //radar about ERGO GPT – our ERGO internal, secure alternative to Chat GPT. Now, the Konrad Adenauer Foundation has also published an interview on this topic with ERGO AI expert Antonia Schiller, which we would like to share with you.

How did ERGO GPT come about?

When OpenAI released its ChatGPT application at the end of 2022, we at ERGO also took a look at the technology and examined its potential for the insurance industry. The potential is great, as such models are capable of understanding and generating human language. However, it was also clear that we as a group needed our own internal solution. The insurance industry is one of the most heavily regulated industries in Germany. We work with highly sensitive data on a daily basis. For this reason, we cannot use public tools, but need a secure and internal version. This is how the idea for ERGO GPT came about. To do this, we use GPT models provided by OpenAI via Microsoft. However, we have developed a self-contained, data protection-compliant version with its own interface and functionalities, which we have been making available to our employees in Germany since May 2024 and are continuously developing to remain state of the art. With our ERGO GPT, we were actually one of the first companies in Germany to offer its employees such a tool at the time.

What tasks does ERGO GPT perform?

ERGO GPT is essentially a general-purpose AI. This means that ERGO GPT can assist with a wide range of tasks. These are primarily text-based activities, such as writing emails or summarising texts. However, ERGO GPT also offers support for creative work, for example in developing concepts or creating new content. It can also provide support in IT, for example with coding or in Excel. So there are many use cases, and each employee can see for themselves where ERGO GPT delivers added value for them individually, enabling them to complete tasks faster and more effectively.

How has your workforce responded to ERGO GPT?

It has become an accepted work tool and is used a lot. Over 60 per cent of the workforce use it regularly, and more than 2.5 million prompts have already been entered into ERGO GPT. This is certainly also due to our targeted empowerment measures. After all, in order for employees to use new technologies effectively, they must also learn how to use them. We therefore offer (online) workshops and training courses, theme days, lectures, prompting competitions and more. This adoption management is important so that employees can use new tools effectively and recognise their advantages.

In Germany, the issue of a lack of resources is often raised. How was that for you?

Our Chief Digital Officer Mark Klein always says, ‘AI doesn't come for free.’ If you want to successfully implement AI in your company, you have to make consistent and continuous investments. This includes investments in licence costs, platform development, software development and integration into backend systems, personnel and material costs, departmental expenses, hardware and much more. Of course, ERGO GPT also involves (ongoing) costs, even though we used existing GPT models and therefore did not invest any resources in the actual model training. Overall, however, GenAI is strategically very important for ERGO, which is why the group will invest around £130 million in expanding its GenAI platform by 2030.

What potential do you still see for GenAI in the insurance sector?

As I said, insurers have immense potential to gain from using GenAI. With hundreds of thousands of customer contact points every day that require language processing, insurance companies are ideally suited to using language models and tools based on them. GenAI could be used to supplement traditional AI and rule-based systems in data classification and extraction from documents. Further automation and acceleration of claims settlement through the analysis of images, videos and other data is also conceivable. The tools could conduct more natural conversations in the form of chatbots and virtual assistants and also handle more complex enquiries. In addition, the models could help to identify patterns and thus be used in the fight against fraud to ensure fair claims settlement. Furthermore, AI agents could be pioneering. They could perform clearly defined tasks independently, such as sending emails or updating databases. In the so-called ‘agentic economy’, specialised, autonomous AI agents would network with each other to solve and automate more complex tasks . So there is still a lot to imagine, because we are only just getting started in the field of GenAI.

Interview: Konrad Adenauer Foundation