Lizzie Mabe is a data scientist at Well, a health tech company that uses AI to help users track their health and build healthy habits. She is also an IAA graduate (class of 2024) whom I’ve had the opportunity to connect with last December during mock interviews.
Olivia: What do you do at Well, and what was your journey to that position?
Lizzie: I’ll start by talking about what Well is. Well is a small company. It is a healthcare app. And it is directed at individuals using the app to improve their health. And so, we do that through your employer, so you would get access to this through your benefits package. And it would allow you to engage with content that is directed personally to you, that’s directed at improving your health in some way. And it’s sort of a gamified system that allows you to unlock points and get gift cards and stuff. So it’s all directed at a gamified system that ultimately helps your health improve and lower your healthcare costs. Within that, I am a data scientist on the BI and Health Outcomes team. [I’m] really focused on how the app is working. How well is it working at our goal, which is to improve people’s health and reduce their healthcare costs.
Olivia: That’s really neat and really cool. Are there any projects that you’ve worked on at Well that you are particularly proud of or excited about?
Lizzie: Yeah, sort of the one that I’ve been working on the whole time I’ve been here. I started in July. It’s been 7 months. The one that I’ve been working on pretty much the whole time has to do with intervention effectiveness.
Within the app, all of the different things that you see, we call them interventions. It might be an article about lowering your blood pressure, or it might be a journey, we call it, of several days, of here’s how you can improve your diet. If you’re focused on XYZ, these particular conditions [are] related to you.
All those things, we really want to get an understanding of how well they’re working. A big part of my job is figuring out this individual intervention, how well did it do at achieving our goal, which was to reduce stress by this specific amount, or then we can do that on larger and larger scales.
We could say what channel was most effective for that, sending out an email versus sending something within the app, versus having somebody reach out to you and ask about something and give you information that way. Which of those are most effective for which outcomes we want to look at?
And then finally, we can do it on a larger scale of how many total interventions, or points of contact, do we need to have with somebody before we can expect their stress, or their blood pressure, to actually eventually improve? It’s been really fun to work on that project.
It’s also really fun to work for a smaller company, because I get to sort of create the system for doing this, or be heavily involved in that. Where, if you were at a bigger company, this would have already existed and already been in place, and you’d be working on very small parts of it. For me, it’s like we’re inventing and coming up with ideas of how best to do this, so it’s been really fun.
Olivia: That sounds really fun. So you’re creating systems, what methods or tools do you use?
Lizzie: It’s probably [a] 50-50 split between SQL and Python for me. So that’ll be two weeks at a time of doing the SQL to get everything in place to have the data that I need to then do the modeling process, which is in Python. And then eventually it goes into Looker, so that’s dashboarding. I’m not actually involved in any of the creating of the dashboards, but I’m aware of it because some of the SQL work then leads into dashboarding.
Olivia: That’s really cool, just sort of a follow-up question, what type of data are y’all using? And also, what type of modeling are you doing?
Lizzie: Yeah, good question. Because we’re offered through your employer, that allows us to have access to your healthcare claims data. In addition to the demographic information about you and how you behave in the app specifically, we can also get your healthcare claims, so that makes it really interesting. We have a lot of rich data to work with, and it makes it extra complicated and fun.
And then the types of modeling, it really depends.
I’m generally of the mind that you should use the simplest model possible to achieve what you want to achieve. But that said, that really runs the gamut of different options. We do some logistic regression. We have done some XGBoost. I’m getting into a lot of Bayesian right now because we want to have some sort of understanding of distributions and simulations. So a lot of different types of modeling. It’s really fun.
Olivia: That’s really cool. I think right now, we’re learning Bayesian.
Lizzie: Good luck with that. It’s so hard because you learn to think in one way, and then Bayesians just, actually, think in a completely different way.
Olivia: You mentioned you work with claims data and insurance data, and right now at the IAA, we’re spending a lot of time discussing data ethics. Are there any ethical considerations or regulations that you have to consider when working with such data?
Lizzie: Definitely. [It’s] a big-time HIPAA situation. There’s a lot of painstaking attempts to de-identify all of our data. The data sets that we work with don’t contain anybody’s names or any personal information, and that’s really important to maintain. And generally just having an understanding of these are real people that you’re working with, this is their actual health, which is so personal. Every single data point is a real person, and trying to keep that in mind and make sure everything stays private.
Olivia: That’s really good. Shifting gears a little bit, do you have any advice for current MSA students or aspiring data scientists?
Lizzie: I think that the main thing that is different from what I was expecting versus now, and I wish I’d known a little bit more is that my job, in comparison to being a student, is a lot more big-picture focused.
I think the skills that have ended up being really important are the ability to large-scale understand a problem, understand the big chunk steps that need to be taken to accomplish the problem, and then putting that into action. I think, especially with AI becoming more and more prevalent, we’re going to be doing less and less of writing our own code. It’s great to be able to have the technical and specific skill of writing code, but I think the ability to zoom out and think about [it] in the broader scope of things, in the big picture, what needs to happen in order to actually accomplish this problem.
I think technical people have a tendency to want to fit a problem to their solution, so they’re like, “I have these fun models that I want to work with, and I know how to code really well. I want to be able to do that.” And then find the problem to fit what they already want to do, and I think it’s much more important to have [an] understanding of the problem, and then fit a solution to it.
Olivia: That makes sense, and that’s actually really good advice. I’ll definitely take that with me into my future roles, wherever those will be. So my last question is the fun question. I saw on your LinkedIn profile that you consider yourself to be the Elle Woods of data science. Do you have any advice about how to make business casual or business formal attire more fun while still keeping it professional?
Lizzie: Oh my gosh, yes. I mean, for me, it’s wear pink, because I am the Elle Woods of data science. But I think generally you don’t have to lose your personality to be professional. I really appreciate, and I think a lot of people really appreciate, when people are fully themselves, showing their full personality, whether that’s through their clothes or through decorating their offices. My office is completely covered with pink decorations.
I think when you show up to work fully as yourself, you’re the best version of yourself, and that shows in your work as well.
Columnist: Olivia Sturges