Kevin Mercurio is a Data Scientist at Spotify and the Teaching Assistant for our upcoming Live Online Introduction to Data Science part-time course - sign up for a free sample class here. The upcoming course will be taught by Sergey Fogelson, course designer and VP of Analytics and Measurement Sciences at Viacom, from October 3rd - November 9th on Tuesday and Thursday evenings. View full course details here.
A common thread weaves through Kevin Mercurio's career. No matter the role, he's always had a hand in helping other people find their way to data science. As a former academic and current Data Scientist at Spotify, he's been a mentor to many over the years, giving sound advice and guidance on both the hard and soft skills it takes to find success in the industry.
We're excited to have Kevin on the Metis team as a Teaching Assistant for the upcoming Live Online Introduction to Data Science part-time course. We caught up with him recently to discuss his daily responsibilities at Spotify, what he looks forward to about the Intro course, his fondness for mentorship, and more.
Describe your role as Data Scientist at Spotify. What a typical day-in-the-life like?
At Spotify, I’m working as a data scientist on our product insights team. We embed into product areas across the company to act as advocates for the user’s perspective and to help make data-driven decisions. Our work can include exploratory analysis and deep-dives on how users interact with our products, experimentation and hypothesis testing to understand how changes could affect our key metrics, and predictive modeling to understand user behavior, advertising performance, or content consumption on the platform.
Personally, I’m currently working with a team focused on understanding and optimizing our advertising platform and advertising products. It’s an incredibly interesting area to work in as it’s an important revenue source for the company and also an area in which data-driven personalization aligns the interests of artists, users, advertisers, and Spotify as a business, so the data-related work is both fun and valuable.
As many would say, no day is typical! Depending on the current priorities, my day could be filled with any of the above types of projects. If I’m lucky, we might also have a band drop by the office in the afternoon for a quick set or interview.
What attracted you to a job at Spotify?
If you’ve ever shared a playlist or a mixtape with someone, you know how great it feels to have that connection. Imagine being able to work for a company that helps people get that feeling every day!
I grew up during the transition from buying albums to downloading MP3s and burning CDs, and then to using services like Morpheus or Napster, which did not align the interests of artists and fans. With Spotify, we have a service that gives millions of people around the world access to music, but finally, and more importantly, we have a service that enables artists to earn a living off their work, too. I love our mission to help make meaningful connections between artists and fans while helping the music industry to grow.
Additionally, I knew Spotify had a great engineering culture, offering a combination of autonomy and flexibility that helps us work on high-priority projects efficiently. I was really attracted to that culture and the opportunity to work in small teams with peers who turned out to be some of the sharpest, friendliest, and most helpful bunch I’ve had a chance to work with. We’re also great with GIFs on Slack.
In your former roles, you worked with a number of Ph.D.s as they transitioned from academia into the data science industry. You also made that transition. What was it like?
My own experience was transitioning into data science from a physics background. I was lucky to have a physics role where I analyzed large datasets, fit models, tested hypotheses, and wrote code in Python and C++. Moving to data science meant that I could continue using those skills that I enjoyed, but then I could also deliver results in the “real world” much, much faster than I was moving through research projects in physics. That’s exciting!
Many people coming from academic backgrounds already have most of the skills they need to be successful in data-related roles. For example, working on a Ph.D. project often presents a time when someone has to make sense out of a very vague question. One needs to learn how to frame a question in a way that can be measured, decide what to measure, how to measure it, and then to infer the results and significance of those measurements. This is exactly what many data scientists have to do in industry, except the problems pertain to business decisions and optimization rather than pure science problems.
Despite the conceptual similarity in problem-solving between industry and academic roles, there are also some gaps in the skills that make the transition difficult. First, there can be a difference in tools. Many academics are exposed to some programming languages but often have not worked with the industry standard tools before. For example, Matlab or Mathematica might be more common than Python or R, and most academic projects don’t have a strong need for DevOps skills or SQL as part of a daily workflow. Fortunately, Ph.D.s spend most of their careers learning, so picking up a new tool often just takes a bit of practice.
Next, there’s a big shift in prioritization between the academic environment and industry. Often an academic project seeks to get the most accurate result or yields a very complex result, where all caveats have been carefully considered. As a result, projects are usually done in a “waterfall” fashion and the timelines are quite long. On the other hand, in industry, the most important objective for a data scientist is to continually deliver value to the business. Quicker, dirtier solutions that deliver value are often favored over more precise solutions that take a long time to generate results. That doesn’t mean the work in industry is less sophisticated – actually, it’s often even stronger than academic work. The difference is that there’s an expectation that value will be delivered continuously and increasingly over time, rather than having a long period of low value with a spike (or maybe no spike) at the end. For these reasons, unlearning the ways of working that made you a great academic and learning those that make you effective in data science can be tough.
As an academic, or really as anyone trying to break into data science, the best advice I’ve heard is to build evidence that you’ve sufficiently closed the skills gaps between your current and desired field. Rather than saying “Oh, I’m sure I could build a model to do that, I’ll apply to that job," say “Cool! I’ll build a model that does that, put it on GitHub, and write a blog post about it!” Creating evidence that you’ve taken concrete steps to build your skills and start your transition is key.
Why do you think so many academics transition into data-related roles? Do you think it's a trend that will continue?
Why? It’s really fun! More sincerely, many factors are at play, and I’ll stick to three for brevity.
- - First, many academics enjoy the challenge of tackling vague, difficult problems that don’t have pre-existing solutions, and they also enjoy the lifelong learning that’s needed to work in quantitative environments where tools and methods may change rapidly. Hard quantitative problems, inspiring peers, and rigorous techniques are just as common in industry as they are in the academic world.
- - Secondly, some academics transition because they’re pushing back against a feeling of being in an ivory tower – that their research work may take too long to have a visible impact on people or society. Many who move to data science roles in healthcare, education, and government feel that they’re making a real impact on people’s lives much faster and more directly than they did in their academic careers.
- - Lastly, let’s combine the first two points with the job market. It’s clear that the number and geography of academic positions are limited, while the number of research and data-related roles in industry has been growing tremendously in recent years. For an academic with the skills to succeed in both, there may now be more opportunities to do impactful work in industry, and the demand for their skills presents a great opportunity.
I absolutely think this trend will continue. The roles played by a “data scientist” will change over time, but the broad skill set of a quantitative academic will be malleable to many future business needs.
As you join Metis to TA the upcoming Live Online Introduction to Data Science course, what do you look forward to? Why did you want to take this on?
I’ve always loved teaching because it’s so fulfilling to have an impact on students or professionals seeking to grow in their careers, or who just have a passion for learning new topics and having those “aha” moments. I owe all my success to teachers and peers before me who were patient enough to help me get up to speed, so I like to give back to others when I can.
Since starting my own career, I’ve been involved in mentorship and career guidance for people seeking roles in data science and tech, but I was particularly excited about the Metis course because it’s an opportunity to get back to the more structured, course-oriented approach to learning that I personally enjoyed as a student. The Metis Intro to DS course is particularly strong because it combines material that is both highly relevant and deep in content with a live-instruction model that mimics a real classroom, but it’s even better because people can attend from anywhere. I’m also really excited to see how the community and social aspects of being in live lecture help the students learn together and form a network that they can participate in even after the course ends.
Want to learn from Kevin and Sergey? Sign up for a sample class or enroll here.