End to End Data Science in a Global Reinsurance Setting

Uwe Nagel, Head Data Analytics & Insights, Swiss Re Management AG

Uwe Nagel, Head Data Analytics & Insights, Swiss Re Management AG

Over the last few years, we've seen big data, artificial intelligence, cloud, and many other "game changers" that were announced to completely revolutionize the industry at large. In practice, things are a bit more complicated and there have been successful and unsuccessful small adjustments and medium revolutions moving toward a new kind of business behavior that integrates all the new possibilities of the digital world in a beneficial manner. While new technologies are widely available and can theoretically be integrated without much effort, the changes this entails require a certain mindset that is not prevailing in the insurance industry. We still see a low level of automation, a limited appetite to execute on innovation, and high concerns about the risks of failing projects. In consequence, a true transformation requires a change management process with management attention, providing space for learning and failure.

Among other initiatives, Swiss Re created a global data science team implementing an end-to-end approach. The setting is as simple to describe as it is hard to implement: anyone can realize a project idea. In a cooperative effort of the originator, consultants, and data scientists we develop the business case, do a feasibility assessment, and win a sponsor for the project. Sponsorship on the business side is crucial to ensure buy-in and commitment of resources such as the capacity of experts. Data scientists then develop a solution in collaboration with the business. Weekly or more frequent updates between data scientists and business experts are the minimum, embedded scenarios with data scientists sitting directly with the business can be beneficial. When a solution develops into a potential product, solution engineers and IT come into play.

Solution engineers are specialized in the implementation and integration of models with existing IT landscape and business processes – the backbone of successful data science projects. Ultimately, they also support improvements and curation of the product during its life cycle.

While this setup uses a centralized setup, one may argue that data scientists should be embedded within the business. Being close to the problems and having a good understanding of business needs is a clear advantage. There are good arguments for both sides – integrated and centralized – so let's have a look at them.

While data science projects are often very technical, an eye for the overall picture can often change a model with insufficient accuracy into a successful project, e.g. because it is "good enough" to provide input for experts or still better than nothing. Data scientists interacting with the business also tend to find their own work. They see processes and problems from their own perspective and can identify solutions that are not apparent for others. They also are "just there" being able to provide constant, barrier-free insights and support. All those little things where someone who knows how to handle data beyond standard spreadsheet manipulations easily adds up to a substantial contribution and over time will develop into projects with a greater individual contribution.

A centralized team of data scientists comes with a greater distance between data scientists and originators. Processes for project setup, prioritization, and other organizational means must be implemented to optimally distribute efforts and measure benefit. A project pipeline filled by the whole company comes with advantages: project prioritization leads to control through transparency and ensures full capacity usage. Project competition avoids gold-plating and ensures projects are finished in minimal possible time. Another aspect is scaling: technical specializations are not aligned with business areas but usually in demand in many places, as for example document intelligence. This is also true for technical solutions which can often be recycled in different contexts, ideally leading to the development of centralized platforms which increase efficiency in future projects.

In turn, this leads directly to the topic of standardization and quality control. Data science projects also involve non-technical aspects such as legal and compliance, dealing with external partners and clients. Introducing standards can help to reduce costs through increased efficiency while at the same time increase quality. Establishing and implementing such standards is a crucial advantage of a centralized team. Assuming data science will increasingly become a competitive advantage and therefore go beyond experimental stages, a centralized competence center with a strong cost control will be another benefit. Finally, a community with knowledge exchange and collaboration is a major source of motivation and an essential demand for data scientists. Constant education is obligatory in a field that is still in a very dynamic development with new techniques being developed almost monthly.

Ultimately, the structural setup of a data science team must be designed for each setting individually and there is no one-fits-all solution. I hope though, that the considerations summarized above can provide a contribution for a surely ongoing discussion in many companies.