Capital One Labs - Data Science at a Bank: Randy Carnevale Interview
We recently caught up with Randy Carnevale, Director of Data Science at Capital One Labs. We were keen to learn more about his background, his move to data science from medical informatics, his choice of going to a financial firm to do data science, what he thinks of data science education, and why he has chosen to work with the Metis Data Science Bootcamp to find up-and-coming data scientists...
Hi Randy, firstly thank you for the interview. Let's start with your background and how you became interested in working with data...
Q - What is your 30 second bio?
A - Having done a bachelor’s in computer science, I’m a bit of a hack of a programmer. With a master’s in public health, I also added “so-so statistician” to my resume. Coupling those with a PhD in medical informatics, I’ve somehow managed to turn it all into an interesting career as a data scientist in banking
Q - Was there a specific "aha" moment when you realized the power of data?
A - Two things come to mind:
I still have them every time I work on a model fitting problem in an unfamiliar domain. Sharing results and interpretations of predictive models with a subject matter expert is always a fun experience. Maybe 10% of the insights you come up with from the interpretation of the model are things they hadn’t heard or thought of before. That tends to be the interesting part of the exercise for them, and is pretty rewarding for me too.
However, the 90% of what I can tell them that they know already is often even more interesting to me. By combining good data with a good model, you can often instantly back into ideas and lessons people have had to spend large parts of their careers learning. Pretty amazing stuff.
Q - How did you come to work at Capital One Labs?
A - It was a bit of a roundabout process. My PhD program at Vanderbilt had close ties to the academic medical center there, and I was lucky enough to be able to implement programs I developed directly into hospital systems from time to time. It was pretty amazing to see things I’d created directly improving patient care. However, I was pretty certain I didn’t want to stay in academia after graduate school since living grant to grant didn’t appeal to me.
After I finished my degree, I went to work as an analyst in health insurance. Though I had the pleasure of working with an outstanding team there, I found that I missed the tight ties to patient outcomes since the levers you can pull as an insurer are much larger in scale but much farther removed from the end results. However, I found that I really enjoyed working with the larger scale data, and started playing around with Kaggle competitions to scratch that itch.
In the process, I came across Capital One Labs through Kaggle’s forums, and decided to take a look. Through the interview process, I found that they were working on a lot of problems at that large scope, and though it’s a very different industry, the Labs have a very large focus on Capital One’s customer experience, giving me back a nice part of that closeness to the end user of the data products I’m building.
Thanks for sharing all that background. Let's switch gears and talk about Capital One Labs, where you work as Director of Data Science ...
Q - What is the 30 second bio of Capital One Labs?
A - Capital One Labs is essentially a tech startup within Capital One. We’re a mix of product managers, designers, data scientists, and engineers who are focused on designing and building the next generation of financial products and services. We’ve been given a lot of room to explore new ideas and new ways of doing things, and we’ve got strong partnerships with people in the core of Capital One that allow us to have an impact on 60 million customers.
Q - What does being Director of Data Science at Capital One Labs entail?A - I lead a team of data scientists focused on internal consulting projects. This gives us a lot of opportunities to apply the general purpose data skills we’ve all picked up along the way to a wide range of banking problems, as well as to learn about what’s interesting and unique about each segment of customers we serve as a company. Managing a team also means that I get a bit less hands-on-keyboard time than I used to have, but I do still get to write my fair share of code when we’re working with Capital One’s business teams.
Q - What is the Data Science team composition at Capital One Labs?
A - We’re a pretty wide mix of backgrounds. Unsurprisingly, many are from technical backgrounds: some from engineering, others from pure and applied math, and others from physical sciences. We’ve also got a few social scientists mixed in, and our best data visualization guy studied theology before seeing the true light of D3.js. One of my favorite things about data science is that it’s a field where it doesn’t matter what your credentials or certifications are; it’s just about what you know and can do, and our team really embodies that.
Got it - very interesting dive into Capital One Labs and what your role as Director of Data Science entails! Now, we'd also love to dig deeper into the type of Data Science Work at Capital One Labs that your group does...
Q - What have you been working on this year, and why/how is it interesting to you?
A - I’ve gotten to do some work with our Small Business team, who is trying to explore what sorts of untapped data sources might help us better serve our business customers. Because of their size, small businesses represent a really interesting mix of “individual-like” behaviors and “business-like” behaviors, and while our team focused on small business understands them very well, for a newcomer like me, it’s been a fun challenge to find those sorts of patterns in data.
I’ve also done a lot of work on refining the tools we use in our interview process to try to identify good data scientists. Because official data science credentials are pretty new, we need to be able to find quacks-like-a-duck data scientists rather than just people holding data science degrees. As a result, figuring out how to test for the skills we’re looking for while avoiding puzzles or “guess what I’m thinking” types of questions has been an important part of our process, and we’ve been refining it quite a lot over time.
Q - How is the work viewed by the organization and how do you “spread the gospel” of data science?
A - We’ve been very fortunate to have senior leaders who recognize the value of data science without much convincing, so that’s made the job much easier than it might have been given that Capital One is a pretty large company. As a result, we’ve been able to get exposure for the work our group drives and to get plugged into highly visible projects within the lines of business. Producing strong results on both fronts has helped us further spread the data science love through the rest of the organization.
Q - What real challenges have you faced at Capital One Labs?
A - Though my shift in industry from healthcare to banking has been pretty smooth overall, there’s still a lot to learn when you start working with a specific part of the business. There’s a lot of shared knowledge and history that are really helpful to have when working with a specific segment of our customer base, and it takes time to develop that familiarity.
In fairness, this pattern happens fairly often by design. Since one of the goals of Labs is to draw lessons from other industries that can be applied in the rest of the company, many people who join the team have no professional experience in finance. When they start learning about it by working closely with the business teams, it ends up being a nice opportunity for the Labs to share some of their past experiences as well.
Q - What tools do you use at work?
A - We do a lot of Python and Hadoop (and I’m a particular fan of IPython Notebook), but we tend to have a very best-tool-for-the-job mentality overall, especially when it comes to the research-heavy projects. We’ve experimented with a lot of programming platforms and processing frameworks, and when there’s a better option than the one we might default to, we’re not afraid to switch.
Thanks for sharing your thought and experiences on what Data Scientists at Capital One Labs work on - very interesting! Let's talk a bit about the hiring process for your group...
Q - What types of candidates are you looking to hire for?
A - We’re always on the lookout for good data science and data engineering talent. Because we focus mostly on the skills needed for the job rather than the background or pedigree, we don’t necessarily have a specific type of candidate in mind.
Q - How do you sell the work & labs to potential candidates – what will they be working on 3 months, 6 months, 12 months, 2 years, and 5 years from now?
A - The 12 month timeframe is the easiest one to project and sell, I think. Our group is constantly getting new project ideas from the business, and so within the first year, a new data scientist will get to work on at least a few very major projects that directly impact a large group of our customers along with a few smaller gigs here and there.
For shorter time horizons, it mostly depends on what skills and knowledge new team members come in with since we’ll want to find projects that play to their existing strengths. For longer times, it’s difficult to project given the rate of change we see in data science in general.
Very interesting - hopefully some of our readers will reach out to you! Next, let's talk a bit the data science educational process...
Q - What do you think of the current education offerings (e.g., MOOCs, Masters Programs, etc.) teaching Data Science?
A - I’m a huge fan of MOOCs as a way to dip your toes in the water with areas like data science in a very low-cost, low-risk way. It’s a fascinating field, but it’s certainly not for everyone, and so they give you a convenient way to try on data science pants without having to buy them.
It’s still too early to tell about the master’s programs out there since they’re so new. I suspect like any technical master’s in a hot area, they’ll produce both graduates who are really dedicated to improving their craft and graduates who heard that data science is the next Big Thing and want to cash in. Either way, I do look forward to seeing them develop further.
Q - What skills do you think are not actively being taught in most classes that should be?
A - More open source all around. Any program being taught using proprietary languages or tools is doing their students a really large disservice since, while their students almost certainly learn the technical concepts, they don’t develop familiarity with a universally portable platform. Though smart and technical people can pick up new languages and platforms quickly, x+1 years of working with a tool will always be strictly better than x years, and so getting started on those sooner is better. Plus, learning the sort of source control workflows ubiquitous in the open source world makes it much easier to step into a team development environment more easily.
Got it - more open source and more people trying on what data science is will be a great thing! Now, we'd also love to talk a bit about your choice of choosing to work with the Metis Data Science Bootcamp...
Q - You’ve chosen to work with Metis Data Science Bootcamp, why?
A - From first principles, Metis is a really easy program to like: they’re a really nice mashup of practicing data scientists and a company that knows education deeply. They’ve carried that through to practice, being very well aligned with my core view that data science is about the knowledge and skills rather than some certification. I also really like the cementing of the ample teaching they do with the project-based work. It’s one thing to hear someone talk about a concept or to read about it, but it never really hits home until you do it yourself. Having that baked into the structure of the program ensures that it happens that way.
Q - From a hiring manager’s point of view, why should someone consider Metis Data Science Bootcamp?
A - Honestly, there’s no real reason not to as someone trying to hire data science candidates. Because the structure of Metis’s bootcamp focuses more heavily on the student education aspects than other similar programs, their terms are extremely employer-friendly, asking for nothing but a little bit of time interacting with their students through their events, which are fun in their own rights for data science nerds like me. It’s a bit of a no-brainer in my view.
Q - What do you think Metis Data Science Bootcamp offers potential data scientists?
A - The Metis instructors are well-versed in data science, and they pack a lot of very practical teaching into a bounded amount of time, so it’s a great way to get any missing skills you might have in place before you start looking for a data science role. When you do start to look, they’re also very tapped into the broader data science community, making it easier to open doors. It’s not easy to find both of those features in one program.
Thanks for sharing your thought process on choosing to work with Metis Data Science Bootcamp - very interesting! Finally, let's talk about where you see the future of this industry and field...
Q - What is something a smallish number of people know about that you think will be huge in the future?
A - I think data science will have a massive impact on hiring and HR in general in the future. Most of my work on candidate evaluation has been driven by the fact that I don’t trust my ability to separate good data scientists from duds or fakers on the basis of a resume or an unstructured conversation. Those things are more likely to find you mediocre programmers who happen to be good writers or good conversationalists, and you’ll miss out on the stereotypical awkward-but-brilliant types that way.
I’m sure there are people out there who are better at picking out the signal manually than I am. At the same time, I’ve seen good models + good data beat intuition enough times to know that when we’re able to capture the right information, we’ll be able to make better decisions. Since I also believe that the single most important thing a company can do is hire the right people, I think it’s just a matter of time.
Q - Any words of wisdom for Data Science/Machine Learning students or practitioners starting out?
A - Don’t feel like you need to know about everything there is to know in data science. It’s easy to fall into the trap of impostor syndrome where you feel like everyone knows way more about the field than you do because you can always point to something someone else knows that you don’t.
That particular illusion sets in easily because all the good data scientists are T-shaped in their skills with a broad set of knowledge and a lot of depth in one specific area. As a result, you can always point to others’ depth as something that you’re lacking without recognizing the things you have that they don’t. Just keep in mind though that it’s impossible to know it all and that the best data science teams are diverse by design so that you can rely on the strengths of your colleagues.
Randy - Thank you ever so much for your time! Really enjoyed learning more about your background, your move to data science, your choice of doing data science in a financial firm. Good luck with all your ongoing projects!
Readers, thanks for joining us! If you want to find out more about Randy Carnevale, he can be found online at LinkedIn here.
If you want to find out more about Metis Data Science Bootcamp, Metis can be found online here and on twitter @thisismetis. If you are interested in attending the bootcamp, check them out here - Metis Data Science Bootcamp. On April 7th, they are holding their “Metis Data Science Career Day”, and you can get your free tickets here.
P.S. To keep up with our interview series as well as the latest curated news and articles in Data Science, sign-up for our free weekly newsletter -> Data Science Weekly ... subscribe button below :)