AI in insurance: When a 70s tariff meets Google Gemini
Large insurers are past the hype and deep into day-to-day AI. At Signal Iduna, board member Johannes Rath is pushing a practical approach: make AI useful, prove it works, then scale it. Simple idea, serious execution.
US tech, EU-grade guardrails
Signal Iduna runs its internal AI on Google Gemini and Google Cloud. The setup is strict: no access to personal customer data, pseudonymized inputs only, processed securely within their own infrastructure. It's a strategic partnership, not a casual plug-in.
If you care about enterprise controls, review Google's approach to security and compliance: Google Cloud Security.
Horizontal: one AI platform for everyone
In October 2025, Signal Iduna rolled out Gemini Enterprise across the group. Every employee has access to a central AI platform and can build agents that others can use. Internally, the platform is called "Co SI" - complete with an avatar.
Vertical: a health insurance agent built from messy data
Claims volumes in health insurance have surged. Service teams were stuck searching through hundreds of tariffs and thousands of documents. Some tariffs date back to the 1970s - only available as low-quality scans. Students were hired to transcribe the oldest PDFs so the AI could learn from clean text.
The result is a health-specific agent trained across 600 tariffs and deployed to employees to answer policy questions quickly and accurately.
Results that matter
- Referral rate between service reps dropped from 27% to 3%.
- Processing speed increased by 37%.
- Net Promoter Score doubled in the rollout phase.
- Response quality is above 85% across simple and complex inquiries.
There's always a human in the loop. On niche questions, the AI often beats manual search - it can save the 15-minute hunt. The stance is clear: AI takes over tasks; people keep judgment, responsibility, and tact.
Claims, fraud, and the "Zero-Touch Claim" goal
AI already supports damage recording, processing, and fraud detection. Humans make the final call. The long-term aim is straightforward: push toward "Zero-Touch Claim" - fully automated handling for a real-time experience.
Input quality makes or breaks automation
Over 60% of health invoices now arrive digitally, but Germany has no standard format. That variety triggers manual re-entry, stretching a one-day process into a week. Signal Iduna puts AI upfront to normalize and validate inputs. Better data in, better agents out.
Sales enablement without the bloat
Intermediaries use AI to prep meetings and compare tariffs, including new health products. Less time on admin, more time with customers.
Voice is coming
The team is running proofs of concept with speech generation. Rath personally tests products like Perplexity, You.com, and ElevenLabs to see how voice will land in insurance and finance.
From "use" to "case"
Inside the company, the rule is: First use, then case. Employees use "Co SI" for everyday work - summarizing emails, answering questions, drafting. The organization observes actual usage to spot patterns worth scaling. Over 110 "AI Champions" support teams locally and help surface cases with real ROI.
People, trust, and change
About 30% of employees will retire over the next decade. Signal Iduna has a works agreement in place: no dismissals due to operational reasons linked to generative AI until the end of 2028. The message reduces fear and keeps people engaged in building the new workflows.
What's next: sector agents and friction-killers
Two tracks guide the roadmap. Build more sector-specific agents (health, motor, etc.) with deep domain logic. And build small agents that remove bureaucracy and reduce friction across the business. Vertical knowledge, horizontal efficiency.
Practical checklist for insurers
- Data governance: keep personal data out of training; use pseudonymization; run inside secured enterprise infrastructure.
- Fix the inputs: transcribe old PDFs, standardize fields, and improve OCR. Clean inputs raise automation rates.
- Start internal with a human in the loop: let teams answer faster while maintaining accountability.
- Measure what matters: referral rates, cycle time, answer quality, NPS. Scale where metrics move.
- Build capability: a company-wide AI platform, internal champions, and clear business cases before you roll out agents.
If you're assessing enterprise AI options, review Gemini for enterprise use and its admin controls.
Tools mentioned
- Google Gemini and Google Cloud
- Perplexity, You.com
- ElevenLabs (voice)
Build skills across the org
Want a structured way to upskill teams by role? See curated AI courses by job here: Complete AI Training - Courses by Job. For technical and infrastructure teams, consider the AI Learning Path for Network Engineers as an example of a role-specific training path.
Bottom line: AI starts paying off when it's close to the work. Clean the data, put an agent where the friction is, track the numbers, and let your people lead the change.
Your membership also unlocks:
