Actuaries in Non-traditional Roles: Robotics and Cognitive Computing

By Celeste Bremen posted 11-03-2017 03:58


By Celeste Bremen, ACAS, Candidate Representative on the Candidate Liaison Committee

Catherine Cao photo.JPG

It’s an exciting time to be an actuary.  The use of big data and advanced analytics are on the rise, and this gives actuaries a chance to change the way they approach actuarial analysis.  I spoke with Catherine Cao, a Fellow of the CAS who works in Zurich North America’s newly formed Robotics and Cognitive Computing team, to learn more about new developments in reserving and how actuaries are using automation to streamline processes in other areas of the business. 


Cognitive computing and reserving – A new way to approach reserves

Prior to working in Robotics and Cognitive Computing, Catherine worked in Predictive Modeling for three years where she primarily built pricing models.  Catherine still creates models in her current job, but rather than predicting prices, these models are designed to predict ultimate losses.  She says that one of the main differences between creating models for pricing and for reserving is the ability to test reserving models more easily.  In reserving, “you have new losses coming in so you can backtest very easily” to see how accurate your model is.  Pricing models can be more difficult to backtest because the ultimate price charged is affected by more than just the model results - for example, rating and underwriting actions also affect the price the insured pays.  In contrast, backtesting ultimate reserve only requires the historical loss triangles, which makes the comparison between actual results and model results easier. 


When creating predictive models, a common tradeoff is model simplicity and clarity versus complexity.  For example, when using Generalized Linear Models in pricing, the results of the model must be communicated to underwriters and other stakeholders.   The factors and variables selected should be clear so they can be explained to these end users so they understand why a risk is being priced in a certain way.  In reserving, on the other hand, the models can be more complex, albeit more difficult to interpret, because the results do not need to be communicated in the same way – instead, the emphasis is on the final result of the model.


While many companies already use predictive models to price insurance, few use these models for reserving.  Catherine enjoys this part of her work because it gives her the opportunity to use new models and techniques that may not have been used before.  However, she also notes that these new techniques can be used in harmony with more traditional actuarial methods.  They can provide another way of looking at reserves and validating ultimate loss selections.


Robotics – Streamlining processes and reducing manual work

Catherine’s job responsibilities are not limited to reserving; she and her teammates also work on automating processes to make them more efficient, and they use a variety of software to do so.  While spreadsheet macros can be a useful way to eliminate manual work, other types of software can process larger amounts of data, and robots have the capability to work across programs.  For example, a robot can be programmed to copy data from a webpage into a spreadsheet, process it in SAS, and send you an email to let you know that the process has been run.


There are many reasons to automate a process, and the Robotics team considers these when determining what to automate.  For example, automation might lead to accuracy improvement – that is, a reduction in errors due to human error.  It can also lead to less hours spent on tedious or repetitive processes and thus more time for employees to spend on deeper analysis or customer interaction.  However, the benefits to automation must also be weighed with the risks.  There are IT costs and the chance of errors in the automation process – if a process requires a lot of human interaction and decision, it may not be a good candidate for automation.


Catherine says that “the biggest challenge of robotics in actuarial is finding the right process to automate.”  Before beginning the automation process, Catherine and her teammates take a step back and see if there are process improvements that should be made first.  While some processes need to be automated, there might be process improvements that should be made first to make the project more “automation friendly.”  Catherine enjoys this untraditional actuarial work because it requires both programming skills and an understanding of the process that needs to be automated.  The Robotics team often automates processes that their actuarial colleagues in other areas of the business work on, and their actuarial background helps them better understand these processes.


The future of actuarial analysis

Actuaries can build upon traditional actuarial methods by using the abundance of big data and advanced analytics that are available today.  This gives actuaries the opportunity to approach problems in a new way and create new solutions to streamline processes.  While the technological landscape may be changing, one thing that will not change is the need for the analytical rigor that actuaries can provide.