A range as diverse as the requirements

The core of structured reinsurance is that it is tailored to the specific needs of each primary insurer. “We always start from a customer’s specific problem and try to find a solution using reinsurance contracts,” says Dr Eike Meerbach, expert for structured reinsurance in Deutsche Rück’s Central Underwriting Management. To find tailored solutions, it is therefore crucial for the reinsurer to understand its clients’ needs precisely and to work with them very closely.

The range of structured reinsurance solutions on the market is diverse – reflecting the individual requirements of primary insurers. Typical elements include cover for low return periods (5 to 20 years), the limitation of profit and loss potential, limited but sufficient risk transfer, multi-year contracts, and the aggregation of risks and self-financing components. These covers are generally placed with one or a few reinsurers, rather than on the general reinsurance market. For the reinsurer, they are a means of building customer loyalty and differentiating themselves in an increasingly competitive market – while maintaining low overall earnings volatility. For the primary insurer, they offer tailored and cost-effective solutions.

Optimising capital requirements and reducing earnings volatility

In recent years, standard forms of structured reinsurance have emerged on the market that enable primary insurers to address key challenges. A classic example is the structured solvency quota contract, which makes capital requirements more predictable – being tailored to the client and across multiple lines of business and/or years (where applicable). Accordingly, structured solutions can support targeted growth plans and the development of new business areas.

Another established form is multi-year stop-loss/AXL arrangements (annual aggregate excess of loss). They reduce earnings volatility and safeguard plan figures – likewise across multiple lines of business and also as calendar-year solutions, where required. So-called retention sub-layer programmes are examples of this approach and have seen increased demand in recent years. These programmes provide protection against increased frequency below the priorities of the traditional reinsurance programme, within a single line of business but also across multiple lines – thereby securing the primary insurer in cases where priorities in the traditional reinsurance programme are increased. Occasionally, retrospective covers are also used to manage solvency capital requirements. These are covers of technical reserves, known as adverse development covers or loss portfolio transfers, on a structured basis.

24/7 availability, efficient handling of complex claims, highly tailored policies – insurers meet these customer demands, in part, by leveraging artificial intelligence (AI). Machine learning, a category of AI, has been supporting insurers for years, primarily for processing structured data – in classifying, identifying and evaluating it. The use of generative AI (GenAI) is more recent. GenAI generates entirely new content from unstructured data – autonomously and based on what it has analysed before. “GenAI doesn’t just refer to language models,” explains Ruben Wienigk, AI & Automation Engineer at Deutsche Rück. “The technology can be applied in many ways, even if it doesn’t perform equally well everywhere. The challenge is identifying use cases that work reliably and truly add value.”

Knowledge databases

A typical use case is knowledge management. Insurers store their internal know-how in databases and link them to a GenAI solution. Similarly to interactions with ChatGPT, employees can then ask questions on specialist topics and receive answers based on internal documents. “GenAI builds a bridge between the organisation’s accumulated knowledge and the people who need it in their day-to-day work, and must access it quickly,” explains Robert Schnoeckel, Head of New Technologies at Versicherungsforen Leipzig, a German knowledge service provider for the insurance industry.

Document extraction

GenAI can contextually process unstructured data from emails, PDFs or scans as part of process automation. It identifies document types, such as invoices and contracts, and extracts the relevant information. This reduces errors and speeds up processes – a major advantage at a time of staff shortages and cost pressure. “GenAI can help employees tackle their daily challenges more efficiently,” says Wienigk.

Risk models

Insurers have for a long time used software to assess risks. However, GenAI delivers faster and more precise insights thanks to the large volumes of data it can process in a very short time. It can, for instance, draw on external sources such as weather data, relate them to customer data and historical claims, and identify patterns or anomalies. Schnoeckel explains: “This is a classic interplay between humans and GenAI. The systems process enormous amounts of data and generate suggestions. Risk analysts assess, prioritise and create tailored proposals.” What matters is that the employee, as an integral part of the process, actively reviews and takes responsibility for the final decision.

Dialogue systems

Chatbots respond to enquiries around the clock, react in context and enable fast, personalised support. The systems handle routine queries independently, allowing employees to focus on more complex customer issues that require human judgement. Customer communication is therefore significantly more automated with GenAI. However, Wienigk stresses: “In customer relationships, closeness and trust are equally important. Some customers don’t want to speak to a chatbot, but to a real person. Especially in claims situations, empathy is essential – and only human contact has been able to provide that so far.”

Image recognition

GenAI image-recognition tools analyse, for example, photos of damaged cars and instantly assess the severity of the damage. More advanced tools go even further: They identify the necessary repairs, calculate the costs and automatically transfer smaller payouts. To do this, the tools draw on vast amounts of previously processed claims and claims photos and derive meaningful conclusions from them. Claims handlers can process cases much more quickly as a result. “Claims handling is one of the biggest cost drivers for insurers. The potential for savings is enormous,” Schnoeckel explains.

Personalisation

AI algorithms are already used today in insurance products, such as telematics. AI tools continuously analyse data – for example, from connected cars, smart homes or wearables – and greatly simplify handling these data volumes. “This allows an insurer to price the actual risk instead of relying on blanket assumptions – for example, that inexperienced drivers have more accidents than experienced ones,” Wienigk explains. “That makes products fairer and more attractive.”

The AI itself is not liable, because it has no legal personality. So people must bear the responsibility. Anja Käfer-Rohrbach, Deputy Chief Executive at the German Insurance Association (GDV), explains: “The same principles apply to AI-controlled drones as to motor vehicles or other aircraft: There is strict operator liability, regardless of fault.” However, insurers could pursue recourse claims, and lawyers examine whether the operator is at fault or whether the drone was defective – the hardware, the software, or the training data. If it is not possible to trace how the AI arrived at that decision, providing such proof becomes difficult. Early stage legislation is taking a closer look at the new technologies – but do they also change liability constellations?

What is changing legally

The EU AI Act came into force on 1 August 2024. The regulation classifies AI systems by risk category, sets out how companies may use them and threatens fines for breaches. “The regulation governs the handling of AI – but not who is liable for damage,” says Dr Daniel Kassing, specialist insurance lawyer and partner at Clyde & Co Europe LLP.

The new EU Product Liability Directive takes a different approach. It has been in force since 8 December 2024. The directive governs compensation claims by natural persons against manufacturers or other economic operators. And for the first time, it creates clarity: Software and AI systems are explicitly defined as products in terms of product liability.

The directive also expands the group of those who may be held liable. In future, liability will also extend to platform and fulfilment service providers – for example, when a Chinese manufacturer proves unreachable. Companies that previously saw no liability risk must now consider taking out product liability insurance, the GDV explains. The directive also makes it easier for claimants to present evidence. In court proceedings, they can request that the manufacturer disclose relevant information. The manufacturer must then show where the fault lies. This makes legal proceedings more complex and more costly. “More and more complex compensation claims against manufacturers are to be expected, especially in the early years,” predicts Kassing.

Originally, the EU intended to introduce a separate law to establish liability rules for AI applications outside product liability law. “There was also discussion about whether operators of certain AI systems should be required to take out compulsory insurance – similar to what is required for road traffic,” says Kassing. In February 2025, however, the EU Commission halted the initiative due to a lack of agreement. Whether the idea will be revived remains to be seen. “If not, I can imagine discussions about supplementing existing laws with AI liability rules,” he adds.

Why existing rules still work

The GDV assumes that little will change outside of product liability law for the time being: “Even though AI is perceived as something entirely new – software has been supporting people in their work for a long time,” says Käfer-Rohrbach. “AI cases can be managed using the existing liability rules.”

As an example, dermatologists have long used AI tools for diagnosis, to detect skin cancer and other conditions. If the software overlooks a tumour and the doctor fails to notice it, the doctor's professional liability insurance will usually be the first to pay – because the final decision lies with the doctor, who bears responsibility. “AI is a tool that one cannot rely on blindly,” Kassing emphasises.

If the software was defective, the insurer can seek recourse against the manufacturer of the medical device. The decisive factor here is this: Was it a treatment error or did the device fail? “In principle, anyone who is otherwise liable for damages, regardless of AI, can be held liable,” the GDV summarises. “As a rule, it initially affects the party whom the injured person can most easily pursue.”