How CCR Re and Reacfin analyze reinsurance treaties with artificial neural networks
The context: AI techniques to analyze unstructured data
This November 16th, the Institut des Actuaires held its annual event 100% Actuaries/100% Data Science in Paris.
Reacfin’s Aurélien Couloumy and Loris Chiapparo seized this opportunity to present the recent work on natural language processing they made for CCR Re. With the special participation of Jérôme Isenbart, Chief Risk Officer & Chief Actuary at CCR Re, they gave a workshop on artificial neural networks and how they used them to facilitate the analysis of reinsurance treaties for managerial work and ultimately underwriting, pricing and risk management purposes.
Data Science can significantly help day-to-day activities of actuaries and risk managers, for example by improving performances but also risks assessments, or by facilitating market overview.
One very common situation is to use these methods to collect and enhance unstructured data that could be useful in actuarial tasks. A perfect example of this is the analysis of reinsurance treaties for underwriting and pricing.
The examination of those treaties is a heavy and repetitive workload. They are complex documents with diverse formats and structures that are often analysed by different eyes with their respective criteria. The analysis can therefore sometimes be incomplete, controls by hand not allowing exhaustivity and homogeneity.
That is why CCR Re approached Reacfin to develop data science methods that would support their reinsurance activities by automating their analysis of treaties, simplifying their understanding of these documents and improving quality control of the input.
How we reduced analysis time significantly
Focusing on two sets of strategies that mix deep learning and NLP with text mining, Reacfin developed a dedicated Python tool which:
- Thanks to recurrent neural networks, analyzes the structure and the topics of the different clauses inside the treaties;
- Thanks to text mining and regular expression, collects and assesses the context of relevant information that can be used by technical teams.
The collaboration with CCR Re allowed the creation of an app with notable practical and trade-oriented results:
- Reduction of process time
- Assessment of accurate information
- Support in all the processes that use the content of the treaties.
- Improvement of risk management: in the end, it can help reduce risks by defining KPIs, claims impacts, and increase compliance by facilitating the quality control of the collected info
But creating such a methodology is not as easy as it sounds, as Jérôme Isenbart explained: “Iteration and collaboration are the keys to success. First, we tested our approach on a few samples and then we extended our analysis to much bigger corpora. During the process, we also made sure to be in constant discussion with our teams so as to develop a tool that does not just repeat tasks, but performs efficiently. What we do is not artificial intelligence, but augmented intelligence. ”
For Aurélien Couloumy, the collaboration is a great success: “It is the willingness of companies like CCR Re to innovate that allows us to always push the boundaries of data science and to positively affect the (re)insurance market.”
The application of techniques such as artificial neural networks and text mining are not limited to the treatment of reinsurance treaties. Follow us to keep up to date with our latest business-centric uses of Artificial Intelligence.
More information on CCR Re
More information on Reacfin’s Datalab