Generative artificial intelligence (GenAI) can speed up processes, support data analysis and prepare decisions. However, the prerequisite for this is that the content generated is correct. This study explains how insurance companies can objectively measure, improve and successfully use the quality of AI products.
To this end, over 200 texts generated by GenAI were compared with those of human experts. They then assessed how accurately the models work in areas such as actuarial services, accounting and risk management. The results obtained can also be transferred to banks, asset managers and other knowledge-intensive sectors.
Two key findings at a glance:
1. Precision does not come about by itself.
Simple commands are not enough. Only with structured prompt engineering and meta-prompting does the accuracy increase from 57 to 98 percent compared to simple prompting. If you are strictly dependent on correct content, you should create prompt databases and proceed methodically.
2. RAG changes the rules of the game.
Retrieval Augmented Generation (RAG) brings internal knowledge into the AI process. This makes results more specific and factual. A basic prerequisite for using AI in critical business processes.
Dr. Fabian Bohnert
Director, Financial Services - Insurance
KPMG AG Wirtschaftsprüfungsgesellschaft
In the study, you will learn how you can systematically evaluate and use generative AI - not as an experiment, but as a component of digital excellence.
