AI in insurance: When a tariff from the 70s meets Google Gemini

Insurers and financial service providers are among the first real users of AI systems in everyday business. But what does that mean in practice, how do you deal with the shift in work priorities, and what about the great power of American corporations in this area? In an interview with heise online, Johannes Rath from the German insurer Signal Iduna, who is responsible for digitalization as a board member for "Customer, Service & Transformation" in addition to operational organization, talks about the practical use of AI in such a corporation – and the concerns of employees.

heise online: Let's start with the elephant in the room: Trump, tariffs, geopolitical risks. Your company uses Google Gemini and Google Cloud for its internal AI services. The internet giant is known to be a US company – does that worry you?

Johannes Rath: No. From the very beginning – in the partnership and in the way we use Google's services – we have placed great emphasis on everything taking place uncompromisingly at the high regulatory level of our industry.

For example: there is no access to personal customer data. The AI solutions are based exclusively on internal, pseudonymized data that is processed securely within our technical infrastructure. We are a strategic partner of Google Cloud – just like Deutsche Bank or Deutsche Telekom, for example. Therefore, there is no elephant in the room for me.

What does an insurance company like Signal Iduna do specifically with Google Cloud?

Google is our transformation partner in terms of our cloud strategy. We are migrating a number of applications to Google Cloud. In terms of content, we are in the middle of two major topics – one horizontal and one vertical.

Horizontal means: In October 2025, we were one of the first companies in Europe to introduce Google's Gemini Enterprise for our entire corporate group. This means that every employee now works centrally with the AI platform and is also enabled to independently develop AI agents that can then be used throughout the organization.

Vertical means: We started building a very specific agent in health insurance back in 2023.

Internally at first?

Exactly, for our employees. Background: In recent years, there has been a high increase in claims processing in health insurance. Of course, this also leads to a corresponding increase in service volume for insurance companies. What cost the most money and nerves was the searching and waiting: Customers, for example, were then on hold for longer and longer. And might have ended up with someone who couldn't fully resolve their issue. With hundreds of different insurance tariffs and thousands of contract documents, this can happen.

That's why we started developing and training our own agent with our health insurance data – across all 600 tariffs. That was enormous work, because it was also about the input quality of the data. For example, we still have tariffs from the seventies that are available as scanned PDFs, but are not machine-readable – and people still want to keep these tariffs today. The data quality of the scans was sometimes so poor that we decided to hire students to transcribe the oldest PDFs.

What measurable benefit does such a knowledge agent bring to health insurance?

After the agent was integrated into the application, the referral rate – from one customer service representative to another – decreased from 27 percent to 3 percent. At the same time, processing speed increased by 37 percent.

The results were really noticeable: we doubled our Net Promoter Score, a measure of customer satisfaction, during this phase. This was the first time we saw such a clear effect on a vertical topic.

And are the agent's results reliable – or do you also see hallucinations?

We have now achieved very good response quality. It remains important that our own employees train the agent and that we continuously improve the response quality.

How high is it actually?

Currently over 85 percent across all inquiries (simple and complex). We always have a "human in the loop" in the process – meaning an employee checks the AI's answer.

What's interesting is: Without the AI agent, the quality of answers to very specific questions is often significantly worse – people might search for 15 minutes.

Our conclusion: Artificial intelligence is increasingly taking over tasks, but it does not replace humans themselves. Experience, responsibility, and judgment remain crucial. And especially in our field: tact and responsiveness.

Does the customer see anything of the AI yet?

No. Currently, our employees use the agent to answer customer questions more quickly.

So it's about tariff details in their contracts? For example: "How many hours of psychotherapy are included in my tariff"?

It's usually more complicated, but yes. Normally, people have to search through documents for this – that's time-consuming and costs time, especially with complex topics.

Let's move on to claims, i.e., claims against an insurer. Are you already using AI to make decisions in case of damage?

AI is already an integral part of our claims processing today. We use AI to support damage recording, processing, or fraud detection. Our AI assistant provides the data-based foundation, and humans make the final decision.

For this year, we have set ourselves the clear goal of further advancing AI integration on the path to our long-term vision: 'Zero-Touch Claim' with fully automated claims processing for real-time service experience.

Speaking of input – what are the issues with invoices submitted by customers, for example?

Insurance is a paper-intensive business. But: More than 60 percent of health insurance invoices are now received digitally. Input quality is key: either the process runs smoothly and automatically, or manual re-entry is required. In Germany, there is no standard for what a health insurance invoice should look like. If manual re-entry is necessary, it takes longer: then a process is not completed in one day, but rather in a week.

Is that also simply due to bad scans?

More due to the diversity of documents. That's why we've put AI at the forefront for better input quality. The crucial point is: the better the data quality, the better the AI agents perform. In my opinion, there is still a lot of potential in this area.

Sales are important for insurance companies. Do you also use AI to find suitable tariffs?

Yes – the entire sales department has been able to use our AI solutions since the beginning. This allows intermediaries to better prepare for customer conversations or create tariff comparisons – for example, when introducing our new health insurance product. So, with AI, we reduce preparation times so that our intermediaries can focus on the customer.

AI can do more than just documents. Do you use speech generation systems on the customer side?

We are currently conducting the first proof-of-concept cases. I consider this one of the most exciting products for the insurance and financial industry.

Proof of concept is an important keyword. A lot of experimentation is happening with AI, which is also expensive. The question then is, what use cases ultimately remain in a company's everyday life.

A lot is talked about use cases. But I believe the actual "use" is crucial. Our approach is: First Use, then Case. In my opinion, that is the correct way to use AI in companies. We have named Signal Iduna's AI platform based on Gemini Enterprise "Co SI". Our employees can also use it to summarize emails, answer questions – just like you use a GPT.

That means – someone now has to constantly monitor within the company what people are doing with AI?

Exactly. We observe to understand how employees use this technology. That's quite fascinating, because it leads to the cases that we later implement on a larger scale. For this purpose, we have established over 110 "AI Champions" who help decentrally to use AI and identify the right cases. So, you have to give the organization the opportunity to build AI capabilities, recognize their value, and then specifically build AI agents.

We ensure that AI is used broadly – but before an AI agent is developed and scaled, there must be a corresponding business case.

Do you also see fear of AI among your employees? For example, that employees think their jobs will be taken away?

In the next ten years, about 30 percent of employees will leave our company due to age. Therefore, we are compelled to act. And we communicate this actively to our workforce. At the same time, we have concluded a works agreement that guarantees until the end of 2028 that we will not issue any dismissals due to operational reasons in connection with the implementation of generative AI.

To stay with practical applications – what AI agents do you want to build?

Vertically, we will build further "sector agents" – in areas such as health insurance or car insurance – where we work very specifically. And we will build AI agents that remove friction from the system: small agents that make work more efficient overall – for example, where bureaucracy creates friction. We are resolutely reducing effort and tedious tasks. So: build and disseminate vertical knowledge – and reduce horizontal friction.

As a final question – where do you personally use AI, professionally and privately?

I always try out products to get a feel for how AI works. This includes Gemini, of course, but also Perplexity and You.com. But I also use Eleven Labs to better understand how AI will occupy the topic of "voice" in the future.

(bsc)