Q&A with senior technical director David MooreFebruary/March 2018
Photograph: Joel Maisonet
Q: How are you using predictive analytics in your job?
A: I lead the Life Insurance Analytics and Predictive Modeling team at Nationwide Financial, which is charged with using predictive analytics to enhance our life insurance business processes. Currently, our focus is using predictive models for life insurance underwriting and accelerating the new business process. However, as we have gone down this path we have realized there is much more to value in predictive analytics than just making a faster underwriting decision. Analytics can help throughout the life cycle, from identifying the customers to whom you want to market, to understanding what drives customers to buy, to finding inefficiencies in your existing business process, to the way you engage with inforce policyholders to encourage them to live a healthy, low-risk lifestyle. Since our accelerated underwriting models went online in 2015, we have been continually trying to enhance them by looking at new data that can provide us with some valuable insight. In turn, we use the data and analytical capabilities to enhance the customer experience.
Q: How did you learn the tools and techniques of modeling?
A: First, predictive modeling is a growing and evolving application, so everyone working in data science today needs to be continually learning. The technology and the tools available to apply to data keep improving, and this requires you to continually evaluate what you are doing and how you are doing it.
I was lucky to get involved in predictive modeling projects from the business side first rather than actually building the models myself. I helped define the business case, put together the cost benefit analysis, analyzed model output and tested predictive models built by data scientists. With an actuarial background, it was easy to catch on to the fundamentals of those models—which at that time were generalized linear models—and see the potential in using predictive analytics in many other applications across insurance companies as well.
Once you get started in predictive modeling, you discover there are tons of resources available to help you learn. There are free data science courses available online at websites such as Coursera. Many data scientists are using R or Python, which are open source languages. This makes it very easy to get started, and allows you to quickly search for help online or post questions to user groups.
Now that the Society of Actuaries (SOA) has increased its predictive modeling content, any actuary who is interested should look through the website and explore the podcasts, newsletters and slides from the recent Predictive Analytics Symposium to explore the topic as it applies to the actuarial profession. Actuaries have the right background in math and statistics and already have experience with a self-driven education model from completing the SOA exams, so it should be an easy step for someone looking to get involved in predictive modeling.
Q: What are the main skills actuaries need for work in predictive analytics?
A: The baseline skills for a role in predictive analytics are part of every actuary’s education, as the basis for everything that we are doing is in statistics. However, the tools that one needs to use for analytics—to handle data, develop models and analyze results—are rapidly changing as more data becomes available and the software to handle the data get better and better.
There are a couple of areas where actuaries may be challenged while developing their skills. First, the massive amounts of data and the increasing complexity of tools often require projects to be collaborative in nature. Actuaries often will want to take over all aspects of a problem when building a model or solution; however, there is considerable value to be gained by problem-solving using a team approach. In the past, it was more typical for one or two people to set the strategy for how to accomplish something, and that model would get built. Now, especially in predictive analytics applications, we can allow more people to try something innovative to see what approaches work best. A decentralized approach to problem solving can be brought together by using ensembles of predictive models, in which models developed separately are combined to produce a solution that works better than any individual model does alone.
I often hear that we are looking for unicorns (i.e., something that doesn’t exist) to build predictive models in actuarial departments, but I think that is a bad analogy. As long as you can build a team with the right skill sets in aggregate, you can accomplish the task. So, if your actuaries are still learning predictive analytics but know the business, you can add a data scientist or a data engineer to bring those skills to the team.
Q: What are some of your best professional experiences as an actuary?
A: I met my wife, Jen, at the Fellowship Admission Course, so that has to top the list. It also means I can’t forget the date we met since it is on our Fellowship certificates.
One of the more unique experiences was being mentioned in a cover story in the Wall Street Journal in 2010 about the use of predictive analytics in life insurance. In the story—based on some of the sessions at the SOA Life & Annuity Symposium—I was simply referred to as “another consultant” as it detailed a presentation I gave on third-party data sources available for predictive modeling. We have come a long way from the initial predictive models, and it’s been a great experience being a part of something new and innovative in the industry.
In my current role, getting out the first model in production was a great experience, as it was the result of countless hours of work for the many team members across the organization who contributed. It also is a building block for future success, confirming that we can apply innovative solutions to life insurance.
Q: What skills positioned you for work in predictive analytics?
A: I somewhat came upon my first project by chance. I had worked in a life insurance inforce management role at my first employer and had experience developing mortality tables and working with the underwriting department. I moved to a consulting firm after completing my FSA and was working with Alice Kroll and Chris Stehno, who were looking to leverage the predictive modeling work that was being done in property and casualty underwriting but for life insurance. They needed a subject-matter expert to work with the data scientists to tie the models together with the risk selection and mortality aspect of what they were trying to build. I was fortunate to be able to work closely with the team and learn from them. Looking back, the data available to companies for model development in 2008 is nowhere near what is available today, but Alice and Chris had the foresight to see the potential before it grew to where it is now.
Q: What is the most challenging aspect of your work?
A: Even though the underlying models are technically very complex, the most challenging aspect of applying analytics in the life insurance business is often change management. Life insurance has a well-earned reputation of being an old, slow-moving industry, but at the same time we are also under threat of disruption from outside the industry. There is currently an incredible opportunity to improve how our industry operates, but now there is also a sense of urgency about it. I think the best approach is to try and find the common ground and look at the problem from the eyes of those impacted by the change. The insurance customer’s expectations are changing rapidly because of technology, as is the data the customer can provide us. We need to meet those expectations for a faster, easier and more personalized experience. To do this, we need to make significant changes to the way our organizations operate. This is a challenge, but it’s also very exciting to be working on something that will be a turning point in the life insurance industry.
Q: Can you tell our readership audience some things they may not know about predictive analytics?
A: The prevalence of predictive analytics in everyday life in incredible. In almost every action or interaction we make during the course of the day, past observations are used to push and pull us in ways so subtle we are often not consciously aware. The marketing you are subjected to and the level of customer service you receive are all determined by algorithms. Even job offers in some industries can be determined by a model, if it has deemed you are more likely to succeed.
Because of this growth of using data-based decisions, especially in insurance, it is important to emphasize ethical use of predictive models to ensure transparency from data to decision. We also need to be diligent when testing and implementing models to review the output, and make sure that the output follows as expected from the model.
Q: If you could turn back the clock, knowing all that you know now, would you choose the actuarial profession again? If yes, why?
A: I chose to pursue an actuarial career because I thought it was the best way to combine a love of math and interest in business, and I would not change that route. In fact, the insurance field is more exciting than ever as a result of the new technology and the disruption being caused by InsurTech. However, if I were starting now, rather than turning back the clock, I would question whether the exams are the best way to prepare someone for the current environment. As the pace of technological change accelerates, it puts increased strain on the SOA to keep the syllabus up-to-date. It also challenges actuaries to be able to continually stay on top of the tools used for managing risk in the age of big data. Fortunately, the SOA is making investments with updates to the actuarial education system, as well sponsoring events like the Predictive Analytics Symposium, to bring practitioners together and help us learn from each other.