How MassMutual's Data Science Lab Pushes the Envelope
Insurance Networking News traveled to MassMutual's data science lab in Amherst, Mass. The lab is led by Sears Merritt, VP of data science for the life insurer, who talks here about the lab's mission, its relationship with traditional IT and how it works to push the boundaries of what's possible in insurance.
INN: Can you tell us a bit about your background?
SM: I spent a lot of time doing network infrastructure builds for telecom carriers and consulting work for small firms out in Colorado. While I was doing that I was pursuing education, getting my Ph. D. in computer science.
INN: How did you end up in insurance?
SM: I wanted to transition away from traditional technology work and transfer into stuff that was more data-driven. My consulting work involved predictive modeling for sports and the job market, and I came out of that and was really looking for a place that had a really rich set of problems to work on. A lot of the more mature firms on the tech side have problems that they’ve been working on for 10 years. The gains that you can make there are much less than a more traditional industry like life insurance.
INN: What do you mean by that?
SM: There are incredibly hard problems to be solved and we work on those every day, and there are some more basic things that need to be done. We’re set up in such a way in that when we work on a basic thing, we use it as a foundation for the next two things.
INN: Can you give an example?
SM: Well, we recently started a middle-market retirement business. First, we had to build out a basic data infra that pulls in all our data and all the key performance indicators they need. Now we can do things that are more sophisticated, like forecast how long it will take a particular consumer to transit our sales process.
INN: MassMutual has a robust data science staff development program to populate this center. Can you tell us a little about it?
SM: Our program is a mix between this center and tight partnerships with five area colleges. Every year we take in a cohort of individuals that are graduation with some sort of STEM background. They take additional courses, and in parallel with that, we pair them up with senior data scientists here and work on projects. Over the course of three years, we set up a sequence of milestones that they have to perform to become an independent data scientist.
INN: How does the data science lab work with MassMutual corporate over in Springfield?
SM: Obviously we work with the business to specify a program, identify the data, the computational and statistic methods that are associated with that problem, and how to take those answers to turn into business process change. We have relationships with the data warehousing team and the core infrastructure team -- the big data tools and systems runs on MassMutual infrastructure, so they support that for us. When we build tools and systems, we work with them to set up environments to set up infrastructure that they ultimately support with us. Generally, we push apps and software into production much like a traditional IT organization.
INN: What are some of the major projects you’re working on?
SM: Anything focused on customer lifecycle is critical. The environment where insurance has been sold is changing. We have to realize that those changes are coming and generate knowledge and insights. Then there’s post-sale customer lifecycle events: How do we do an even better job at customer service, what can we predict or estimate will happen in the future, what’s the best time to call a customer in order to minimize the possibility of an event like a lapse? That can be a very broad set of problems. There’s also things like fraud that have a lot of promise from analytics.
INN: Why is it so important for insurance companies to be analytics leaders?
SM: If we just focus on the underwriting process that’s been around forever, it’s always been a data intensive type of process and we want to maximize our customer experience. Now that the data is available more passively, we can get access to it much faster and do that.
INN: What are some of the new kinds of data sources you are looking to tap into?
We use data from all different places to understand who our policyholders are. We have a substantial retirement services business and there is some fraction that are also mass mutual policyholders. So we can build profiles of those individuals and look at the broader RS pop and find how many of those profiles data that’s being generated from wearable devices to estimate risk. From an LI POV that’s hard because wearables haven’t been around for a long time to estimate that risk if you will. So we have experiments and research partnerships set up where we’re trying to incorporate that space.
INN: What are the challenges to innovation in this way?
SM: For life insurance, it’s complicated because you don’t get to observe outcomes like you do in P&C. P&C claims come in on the order of months. The cycle by which you can run an experiment and get results is much faster. And also, in our world the regulatory framework really does put guardrails what you can and cannot do. You have to come up with novel ways to balance traditional with emerging.