AI in facultative reinsurance - Automated text processing as a breakthrough in underwriting

Optional underwriting - what is it?

With facultative reinsurance, the risk carrier is free to decide whether to accept or reject a risk, the risks are examined and assessed individually and the policies are customised accordingly. Of course, such a complex procedure is not worthwhile for the family home next door, but large international corporations also need insurance coverage. The sums at risk in this area are so high and the expertise required is so specialised that most primary insurers avoid the business and leave it to either reinsurers or highly specialised industrial insurers.

For me, the teams that work there are among the most colourful in the insurance world. To be able to assess a wide variety of major industrial risks, industry experts are hired - it is not uncommon for a nuclear physicist to sit next to a cyber security expert and a mining engineer at the lunch table. These experts then carry out the underwriting, i.e. they assess the risk to which the company is exposed and negotiate the insurance contracts, including the premiums, on this basis.

Workflow - how can AI provide support?

The contracts are usually renegotiated annually, with each underwriter having fixed customers (accounts). Before the renegotiation, the underwriter receives risk-relevant data such as current property and business interruption values as well as technical reports on the individual business premises of the insured person(s) with the submission. The underwriters check the information and then have the opportunity to make various adjustments to make the risk situation acceptable for the reinsurer. For example, they can incorporate exclusions and sub-limits into the contract, increase premiums, assume only part of the risk and much more. They then negotiate these changes with the company or an intermediary brokerage firm and finally sign or decline the risk.

The problem is the large amount of information. Underwriters often look after 20 or more accounts, many of which have a double-digit number of business locations, and for each business location, there are one or more highly technical reports, often comprising more than 80 pages of text. In addition, most contracts are renewed at the end of the year, so all accounts have to be assessed at once. The result is that the information material can often only be roughly skimmed over and important information is overlooked. In extreme cases, risks are underwritten that the reinsurer has excluded.

This is precisely where the strength of AI lies. Risk reports can be processed automatically - the use cases are diverse. Risk ratings can be exported, risk matrices created and key figures such as probable maximum loss (average loss amount) and maximum foreseeable loss (maximum possible loss amount, usually scenario-based) extracted for further pricing, to name just a few.

Here is the complete list of use cases currently known to me:

• 1. extraction of the risk ratings
• 2. formation of risk matrices
• 3. formation of probable maximum loss and maximum foreseeable loss
• 4. extraction of geo-coordinates for the detection of accumulation risks
• 5. identification of partial risks that violate underwriting guidelines
• 6. comparison with previous year's reports, presentation of risk quality on a time axis
• 7. standardised extraction of valuable data into a database format for further processing

What has changed technically - why can AI now process such texts?

Automated text processing used to have certain limits; a very large number of technically similar texts (batch size) were required to train the AI. This is not the case in underwriting, as the risk reports in the various insured sectors differ greatly; a risk report for an oil drilling platform is fundamentally different from one for a rocket launch or a nuclear power plant. More modern solutions already bring a broad "world knowledge" with them. This means that the AI is already very well-trained to understand different texts. Now it only needs to be taught various specialisations. This reduces the batch sizes considerably and is in an area where considerable success can be achieved even with comparatively small quantities of training texts.

Sceptics like to argue that the new generation of AI also makes mistakes, but I think I have sufficiently explained that this also happens to underwriters under high pressure. After all, intelligence does not mean infallibility, but only the ability to think abstractly, logically and rationally. AI - like every other employee in the company - must be deployed according to the importance of the work result following the principle of dual control.

Conclusion

The advantages of using AI in underwriting are obvious. Not only can enormous gains in efficiency and quality be achieved, enabling underwriters to focus more on areas of their work that cannot be automated, such as conducting negotiations, acquiring new customers and consolidating existing customer relationships. More importantly, automated data extraction will unearth what is perhaps the last great treasure trove of data in the insurance industry. This enables targeted portfolio management and accumulation assessments, improves the possibilities of statistical risk assessment in the industrial sector and, for the first time, allows ad hoc checks of specific risk exposures across an entire portfolio - for example, the coverage of business interruptions due to mass illness of employees in the event of a pandemic - a use case that was frequently requested from adesso after Corona.

You can find more information on the digitalisation of underwriting on our website. There you will find various methods and tools for underwriting.

You can find more exciting topics from the world of adesso in our previous blog posts.