The Heise Group has been successfully implementing AI in various departments for two years. With their own AI platform and structured training, they have achieved significant efficiency improvements. In a recent discussion, Kim Scheurenbrand talks with colleagues who were part of this process.
Benjamin Danneberg, a partner at DEEP CONTENT by Heise and editor-in-chief of Heise KI PRO, shares his experiences and insights. He discusses how the Heise Group introduced AI, the principles they followed, and the reasons behind developing their own AI platform, Heise I/O, instead of using ChatGPT. Max Schreiner, senior editor at THE DECODER and co-author of the Heise KI Update, talks about working with generative AI in editorial processes, improving prompts, and handling AI hallucinations.
The Heise Group developed their own AI-powered publishing platform, Heise I/O. Danneberg explains that the company adhered to clear principles: transparency, responsibility, and a focus on proven efficient processes. They chose to develop their own platform instead of using ChatGPT because they needed better process control and the ability to integrate various language models and tools. The ChatGPT framework was deemed unsuitable for professional work.
Over 500 employees actively use AI, resulting in the release of resources equivalent to more than two person-years of work. The company conducted two AI weeks with 15 courses and demo stations, totaling over 35 training sessions and more than 210 hours of training. The program included onboarding workshops for a unified knowledge base and the appointment of AI ambassadors in teams.
Practical applications demonstrate the benefits: The podcast “kurz informiert” saw an 88% increase in views with AI support. Producing the Botti newsletter saves 12 minutes per issue, equating to 1.5 workdays per month. Transcriptions that used to take four hours are now completed in minutes.
The Heise Group plans to expand Heise I/O with full multimodality, including image generation and text-to-audio/video functions. They also aim to further develop AI-based research tools, including existing RAG/chatbots.