Data Science for Risk Management

At the 2016 RIMS Annual Conference, Michael Elliott, Senior Director of Knowledge Resources with The Institutes presented a session on how risk managers can use the large amount of data for analysis.

The Internet of Things (IoT) includes things like robotics such as exoskeletons which can enhance the capabilities of workers and help prevent injuries. Wearables can monitor vital signs from workers looking for signs of overexertion and other warning signs. Also, in transportation things like lane assist and other warning devices can help prevent injuries.

Internal data sources include structured data such as claims history and safety data, and unstructured data such as adjuster notes. Harnessing all this data can be a powerful tool.

An example is the information that telematics can generate for commercial vehicles. You can track seat belt use, speed, braking, acceleration, route, and other elements to ensure the drivers are operating the vehicle safety.

Data science involves three types of skills; Math and Statistic Skills, Hacking Skills, and Domain Knowledge. Math and Statistic skills is people like actuaries who analyze data. Hacking skills involves those who find ways to extract the data into something useful.

Data Science techniques for risk management include:

• Association rules – Looking for association between data elements.
• Clustering – Looking for outliers in the data groups.
• Classification – Classic statistical analysis.
• Regression – Used by actuaries in conjunction with classification.
• Text Mining – Newer area of analysis taking text and turning it into numbers for analysis.
• Social Network Analysis – Looks at connection between things to identify potential fraud.

Machine learning can also assist with worker safety. This takes into account multiple elements that can contribute to a claim including experience, weather, job duties, location, shift, and use of safety devices. Data can look for interaction between these data elements that can lead to trends.

Predictive modeling can assist in identifying fraud and finding claims with potential for adverse development on the claims side. Carriers are also using this to forecast ultimate losses on the underwriting side. There is a large number of potential variables or attributes that can be analyzed. One goal is finding those attributes that have the biggest impact on outcomes. You use those findings to come up with sets of rules of classifications which trigger differences in how the claim will be handled.

Text mining of adjuster notes and emails can look for information about us co-morbidity, lifestyle issues, and other elements that could impact the claim. When key words are found in the notes, that creates a corresponding number that is used in the analytics.

Social network analysis is not just social media but any type of network analysis, in other words anything that can be accessed through a computer network.

Finally, all this analysis should eventually assist in reducing loss costs. Your data is a valuable asset that can assist you in managing risk.