From solving problems on the bleeding edge of machine learning to serving customers search results that anticipate their interests, CarGurus’ data science team covers a lot of ground in their daily work.
“We are a team of people who want to get our hands dirty solving technical problems that have tangible, real-world impact,” Senior Data Science Manager Jason Prentice told Built In.
WHAT CARGURUS DOES
CarGurus harnesses the power of technology and data to bring trust and transparency to the car buying and selling process. The online tools and features are designed to empower both dealers and consumers with the flexibility to choose their own journey through an all-in-one platform for car shopping, buying, selling, sourcing, marketing and more.
For Prentice, CarGurus has been the perfect place to dive into data science.
“There is a commitment to data-driven decision-making and respect for analytical rigor at all levels and functions of the company,” he said. “CarGurus has made significant investments in data infrastructure, which allows my team to focus on our core mission of researching, building and deploying machine-learning models.”
With members who hail from Big Tech, academia and beyond, the CarGurus data science team attracts those who seek the perpetually thrilling challenge of serving both consumers and dealers. Always poised for growth, the team welcomes potential colleagues with the markers of someone willing to roll up their sleeves.
“The online automotive marketplace is an excellent sandbox for a data scientist — rich enough that we’ll never run out of interesting data and questions, but focused enough that we can develop meaningful domain expertise and intuition,” Prentice said.
Built In sat down with Prentice to learn more about the opportunities his team has found to drive impact.
I’ve always been fascinated by tech, science and math. I’m most motivated when applying data and mathematical rigor to real-world questions. I started out in physics and earned a PhD researching the application of physics-inspired models to neuroscience experiments. My time in neuroscience exposed me to machine learning and sophisticated statistical modeling. These techniques are as interesting to me as their applications. Around the same time, data science was blowing up as a field. Data science offered the opportunity for impact on real-world problems, rapid iteration and exposure to a diverse problem set — so I made the jump.
I’ve since had the opportunity to work on a variety of fascinating problems as both an individual contributor and manager, from language models to predictive models and from global supply chains to the auto market. I’ve been grateful to have the chance to lead and grow highly effective teams.
What does the data science team at CarGurus do for the organization?
We have a centralized data science team at CarGurus, which means we work on projects that span the dealer and consumer products, marketing, sales and revenue. Data science and machine learning can be used anywhere the organization wants to discern and apply insights from data in a scalable way.
“Data science and machine learning can be used anywhere the organization wants to apply data insights in a scalable way.”
We work on projects like inventory recommendation — what vehicles to show to consumers based on the inventory they’ve already shown interest in — and dealer churn prediction, which predicts the subscription dealers most likely to cancel or downgrade their service, and why.
What are the most common challenges your team faces and how do you address them?
You need to have access to the right data to understand the problem. This means working closely with our data engineering, product data analytics, product engineering and sales stakeholder teams to ensure the relevant data is being accurately recorded and stored in our system.
Then, we need to understand exactly how the output we’re generating, usually a machine learning model that makes a prediction, is going to be used. An easy and well-known trap for data science to fall into is looking blindly to experts in an ivory tower — models from which companies cannot realize value.
“An easy trap for data science to fall into is working on models from which companies cannot realize value.”
We build machine-learning models that integrate directly into our systems via microservice. We are able to test the insights derived by our models directly and iterate on them over time to increase their value.
What key things do you consider when scoping and defining success for your team?
We want the projects we work on to have the broadest possible impact. We make sure that if we’re investing in something over multiple quarters, there are many possible opportunities for payoff.
We’re currently building a model that predicts the propensity to engage with our consumer products at the individual user level. It can be used in many areas of our business: to improve product personalization, enhance dealer-facing insights and improve our own internal reporting on traffic value and marketing efficiency.
When we undertook the effort to build this model, we knew that our risk wasn’t concentrated on one single stakeholder. We try to stay close to applications that are strategically crucial to the business — so we know that there will always be someone ready to use what we’ve built.
Second, we want to make sure that the project is a good fit for the data science team. When you want to build relationships it can be tempting to say yes to analytics projects. Having an open dialogue with our product data analytics, revenue analytics and sales ops has been essential to our success.
We always see our work as a complement to the work of those teams rather than something that could replace or supplant it — and we try to educate our stakeholders so they can appreciate that distinction.
THE TECH STACK THAT POWERS CARGURUS
- AWS SageMaker
- Snowflake SQL
- Apache Flink
What should potential hires know about working on the data science team at CarGurus?
It’s an exciting time for the team. We have grown from two people to eight in the last two and a half years. We’ve demonstrated our strategic value within CarGurus and have had the opportunity to build out a more sophisticated machine-learning model-serving outfit.
Given this momentum, we are looking for strong technical leaders with an appetite to tackle new challenges. The team is intrepid and enterprising, but we’re also incredibly supportive of each other. We spend a lot of time sharing what we’ve learned from our latest challenges for the good of all. Above all, we want to solve hard problems and lift each other up.