A German health insurer now runs its GenAI member service and 2,400 staff on one system, covering 5 million calls a year.

AOK PLUS is a public health insurer in Germany. It covers 3.4 million members across Saxony and Thuringia. For years, it split GenAI member service across two systems. A GenAI voice assistant answered incoming calls. Separately, a contact center of 2,400 staff handled everything else. Now the insurer runs both on a single platform from the software company NiCE.

Why the split hurt GenAI member service

The two-system setup created a familiar problem. When the assistant could not resolve a call, it passed the member to a human agent. However, the context did not travel with them. As a result, members had to explain their problem twice. Costs climbed too, because every handover consumed staff time that the assistant had already spent.

One platform, one handover

On the unified platform, the assistant now identifies why someone is calling. Then it routes the call to the employee whose expertise fits, and the context follows. Moving to the new system also meant transferring more than 1,400 phone numbers. The insurer did this without a single service interruption. Meanwhile, the system kept answering calls throughout the migration.

GenAI member service on a sovereign cloud

The combined system now handles all 5 million member contacts the insurer receives each year. It answers more than 95% of them. In addition, AOK PLUS is the first health insurer in Saxony and Thuringia to run GenAI member service on a European sovereign cloud. Member data therefore never leaves EU jurisdiction. In a sector where privacy rules often stall AI projects, that combination is what other insurers will study. Other large public institutions face the same constraint, as the Pentagon’s own GenAI rollout showed earlier this year.

Why It Matters

  • Privacy rules do not have to block GenAI in regulated industries. A sovereign cloud is a proven route around that objection.
  • The gain came from closing a handover gap, not from the AI itself. Look at the seams between your AI and your people, because that is usually where the experience breaks.
  • This insurer added the assistant first, then unified the platform a year later. Sequencing that way avoids betting everything on one large migration.
  • At this scale, ask a vendor about operational continuity, such as moving 1,400 phone lines without downtime. That matters more than what the AI does in a demo.