Insuranceciooutlook

Using Machine Learning to Drive Business Outcomes

By Upendra Belhe, Global Chief Analytics Officer, Gen Re

Upendra Belhe, Global Chief Analytics Officer, Gen Re

Machine Learning (ML) has now gained respect in the insurance and reinsurance industries. Practitioners, as well as business owners, understand that it complements and supplements statistical techniques that are used for prediction purposes.

"It is not uncommon for us to have a decision tree picture on the screen and start a business discussion regarding “what is it telling us?” and “what can we do about it?"

In a statistical model, throwing tons of inputs into the prospective model is considered fishing for an answer and is frowned upon. In contrast, ML requires essentially no “a priori belief ” about the nature of the true underlying relationships. ML can be used more broadly to recognize patterns in data efficiently.

Gen Re’s success with ML is based on our choice of algorithms and how we carry the outcomes forward for business use. We downplay the sophisticated aspects of algorithms, while we focus on the interpretability of results. One technique that has made a huge difference in how our analytics team collaborates with our business units is Decision Trees.

We continue to use Decision Trees for classification purposes successfully. The subgroups it generates, and rules it uses, to make splits in the data bring out many business realities (including gaps in the quality of the data itself). It is not uncommon to have a decision tree picture on the screen and start a business discussion regarding “what is it telling us?” and “what can we do about it?”. This has begun a major cultural shift at Gen Re. Our underwriters and business owners exposed to using Decision Trees no longer think of analytics as something challenging to wrap their head around.

It is also imperative to understand how we achieved this. We have stayed away from the plug and play software platforms. Our data scientists are passionate about simplifying and demystifying analytics. We use open source languages such as Python and R to build the trees. We also try to put ourselves in the position of the business managers and ask, “so what?” as we develop and interpret the results. The code is maintained in a way where it can change hands when needed, and we continually look for opportunities to improve the accuracy of outcomes.

The subsets generated by classification Decision Trees are used to storyboard snippets of insights. Our Global Analytics team, along with their business counterparts use these storyboards for brainstorming purposes to propose actionable outcomes. As an organization, we are increasingly fond of collective intelligence using data visualization tools and acknowledging the power of ML in action.

Our Global Analytics Hadoop platform can ingest large amounts of data quickly and allows us to prepare different combinations and aggregations. Our ability to integrate open source decision Tree code with the Hadoop platform, and leverage the resulting efficiencies, has helped us promote the use of Decision Trees to “divide the data and conquer the insights” across the enterprise.

Our success with Decision Trees has encouraged us to investigate the use of other techniques, such as clustering. But we are committed to interpretability and business applicability more than any gain in the pride of our use of ML techniques.

Read Also

How to Make a Notoriously Reactive Industry Proactive

How to Make a Notoriously Reactive Industry Proactive

Mike Gulla, Senior Director of Underwriting, Hippo Insurance
Harnessing Power of Blockchain to Restore Trust

Harnessing Power of Blockchain to Restore Trust

Dorota Zimnoch, Founder and MD, ZING Business Consulting

Weekly Brief