AI Act: How Sparkasse Automates AI Processes Securely

S-Communication Services GmbH, known as S-Com, is the communication unit of the Sparkassen-Finanzgruppe, producing guide articles, campaign pages, and consumer information for sparkasse.de on behalf of about 340 Sparkassen in Germany. Two years ago, the project manager completely dismantled the editorial process, examining 17 individual steps, more than half of which involved manual handovers between Jira, Word, Slack, and the CMS. Widely used cloud-based automation tools failed early because they offered no control over server location and no self-hosting option. Every new tool at S-Com must pass an in-house compliance framework: an assessment with over 100 questions jointly approved by IT security, the works council, and data protection. This approach was part of corporate culture even before the AI Act became relevant for most companies. Only one automation tool remained, one that can be operated entirely within its own IT infrastructure. On that basis, an architecture was built that automates editorial work steps without giving up control over sensitive data. Today, more than half of the steps run automated, some still in test mode. Depending on the content piece, the process is 10 to 20 percent faster, and an external quality assurance service provider has already been saved. However, automation does not replace everything; some steps remain deliberately human. The time saved flows mainly into editorial care: tone, context, and a brand voice that cannot be arbitrarily replaced by a language model. The t3n PRO use case details the governance practices, criteria for which AI models are used for what, and why only part of the process is automated.