Data science, AI, and ML | Pluralsight

Lauren Domnick has witnessed the power of machine learning and artificial intelligence firsthand. As Vice President of Data Science Innovation at Northwestern Mutual, she faced a challenge at the start of the pandemic: underwriters could no longer access the medical records and data that they needed to write insurance policies.

On the recent Perspectives in Leadership podcast episode, Lauren shares how her organization used data science, machine learning, and artificial intelligence to overcome this hurdle, improve the client experience, and enhance her team’s success.

*Answers have been edited for length and clarity*

How is your organization using artificial intelligence and machine learning to make the lives of your clients and underwriters easier? 

We’re using AI and ML models to support the risk decision process for insurance policies. In the past, a paramedic examiner would go to someone’s house, ask them questions, and perform any necessary examinations to assess their risk levels. That wasn’t possible with the pandemic. We had to figure out how we could get the information we needed with historical data that was relatively recent and available. 

With AI and ML models, we could reliably identify someone's risk class based on data without having the paramedic examiner go into their house. Clients no longer need to be poked and prodded with needles, get on a scale, or do any of those things that make the process painful. 

It used to take a month or two months to complete an application and payment. Now, we're doing it in days. Because the AI and ML models are fast and reliable, it’s a great experience for the customer, the underwriter, and the agent or field representative selling the policy. 

It’s an underwriter’s job to determine a product or insurance policy’s cost, risk, and other factors. Was there any initial pushback from underwriters about the concept of using machine learning or AI to do aspects of their job? Or was this something that they bought into and understood right away? 

Anytime you hear that machines or robots are going to be doing a job that a human has previously done, there's a little bit of anxiety. But when you boil it down, our underwriters are the foundational piece of our business. 

If the AI and ML models can take some of the easier, more mundane or less risky cases, that frees up the underwriters to do the more exciting, challenging cases. It makes their job even more valuable when they're able to focus on those cases and get the easier ones out of the way. That's part of it. 

The other part is that we experienced an influx of business during the pandemic. As a result, our underwriters had a backlog. They were able to help clear that backlog because AI and ML were supporting them along the way. So, I don’t think it's been the anxiety-driven experience that one would expect. It's actually been very positive for our company as we partner with our underwriters.

What type of data did you use to build out different policies that machine learning and artificial intelligence could use to determine price points?

For a company that's been in business for more than 165 years, we have a robust set of data that we can use for our models. Essentially, we're trying to get everything a medical underwriter uses in their decision, like lab data, known conditions, diagnostic history, and prescriptions.

We actually have a team dedicated to transforming our old paper records into digital records so that we can use those in our models, too. So, we're going back to the 1990s and pulling data from things like paper check boxes and structuring it in a way that the models can consume.

But as you can imagine, the life insurance industry is highly regulated, and there are certain things we can and can't use. So we certainly follow all of the regulations, but we also take it a step farther and think about the ethical or moral considerations of using certain data sets in our models.

If someone wants to investigate artificial intelligence and machine learning for their own organization, what are some practical steps they can take?

The key to any data science project is data. There's an 80/20 rule for everything. And in this case, it's 80% data understanding, data manipulation, and data preparation, and 20% modeling and validation on the backside. 

I'd say that the key to getting started is having some very clever and curious people around. We have a great team of data scientists, data analysts, and data engineers who work really closely with our program management department to build models that are sustainable and prioritize our values for our customers. 

Investigate what data you have available to use or what data you can make available. We purchase external data from some of our partner vendors that help us supplement our internal house data. We’re also constantly attending industry events and staying current with new modeling techniques that may be available. 

So it's really staying curious, staying current, having the right data, and just getting started. You have to do some trial and error and learn as you figure things out. Don't be afraid to give it a try.

Given the nature of the insurance industry, you’re connected to medical records and private information. How do you test new technologies securely? 

As you just alluded to, we have a treasure trove of incredibly personal and valuable data that needs to be protected at all times. To that end, we have an incredibly robust security and privacy process in place anytime we bring in a vendor. We also have checklists, policies, and procedures. 

That being said, it takes a little bit of time to bring in a new technology because we want to make sure we follow those policies to keep our customers' data safe along the way.

Do you have any tips for gaining buy-in or implementing machine learning and AI? How do you ensure that your technologists are comfortable using these new technologies?

We're really open to a customizable path depending on one's career interests. But like you said, things are always changing. We have a lot of training programs available, but we're also going out to conferences, connecting at industry events, and going to learning activities where we can keep current with what other companies are doing.

We also do monthly lunch and learns so that people can share different projects that they're working on. A data scientist in one area might be working on something and find a way to make the process more efficient. Let's share that out and make sure that other data scientists can utilize it, too. 

With the surge in open source technologies, the sharing has just improved. Someone doesn't have to sit and work on the same thing their neighbor across the country or across the office is coding. They can share it and make things more efficient for everyone.