What is Data Science at Lyft?
By Simran Mirchandani and Thibault Martin
Throughout this post, we will discuss our various (and newly rebranded!) Lyft Data Science roles, explain how Data Science fits into the rest of the company, and describe some unique challenges that our Data Scientists tackle. If you are interested in learning about our work and how we approach it, read on!
Data Science Roles
Lyft has assembled a team of 200+ Data Scientists with a variety of backgrounds, interests, and expertise in order to make the best possible decisions, and thus build the best possible product. We organize the team based on the typical output they produce: Decisions and Algorithms.
Data Scientist, Decisions: “Data Science for Humans”
Decisions Data Scientists influence decisions made by executives, product managers, engineers, ops/business teams, and other stakeholders. They utilize a deep understanding of the business to develop decision frameworks that drive alignment on the most impactful problems and solutions.
Data Scientist, Algorithms: “Data Science for Machines”
Algorithms Data Scientists develop models that power internal and external production systems. They typically apply a combination of optimization, machine learning, and inference methods to design, improve, and monitor models and systems.
Many components of day-to-day work are the same across all Data Scientists, such as identifying product or business opportunities, surfacing data insights, designing experiments, and interpreting their results. The strength of the team comes both from these common expectations as well as from a diverse spectrum of complementary skills.
The distinction between Decisions and Algorithms helps us hire and staff based on business need: influencing human decision-making or writing high-quality production models. However, the archetypes are not intended to be restrictive. Depending on the need and people’s ability to execute or desire to learn new skills, Data Scientists are encouraged to explore both spaces.
Focus Areas
Teams at Lyft are organized by focus area, each of which includes members from multiple functional areas of expertise (Product, Engineering, Data Science, Design, Business/Operations). The various functions sit together and work hand-in-hand towards common roadmaps and targets. The goal of this structure is to clarify ownership, streamline communication, and enhance cross-functional collaboration. Here’s a sneak peek at what each focus area works on and how Data Science contributes.
Marketplace
This area is responsible for the underlying systems that decide how to dispatch and price each ride. Marketplace is all about trade-offs and efficiency: making the best use of our resources to keep drivers, riders, and the company happy. Some example projects are:
- Designing a model to assign drivers to riders across a city in the most time- and cost-efficient manner.
- Determining which riders should be matched in a Shared ride such that all riders have positive experiences, while also making the ride profitable.
- Leveraging pricing and incentives to balance supply and demand.
Rider
This group focuses on acquiring and retaining riders by building high-quality end-to-end products and experiences. We do this by:
- Measuring the quality of riders’ experiences, and developing and testing frictionless products and features to improve their experiences.
- Analyzing past behavior to understand riders’ needs and preferences, and designing memberships and other programs to better retain riders.
- Optimizing rider acquisition and engagement funnels through targeted models and marketing campaigns.
Driver
This area focuses on ensuring that we grow, retain, and engage our driver base by providing a best-in-class product experience. We do this by:
- Developing new pay structures that ensure drivers receive a compelling value proposition of consistent and predictable earnings.
- Designing and testing new product features to help deliver the most intuitive, distraction-free, and safe driving experience.
- Optimizing the driver acquisition and onboarding funnels, building models that identify drivers at risk of churning, and creating user segmentation that enables customized product and marketing features.
Rideshare Planning & Operations
This group owns the growth/profitability goals for Rideshare, and accordingly builds plans and sets market-based strategies. Some focuses are:
- Building the diagnostic tools that improve our understanding of the business, for example through anomaly detection, attribution, competitive intelligence, and reporting.
- Building investment plans under various scenarios that evolve with changing conditions, for example through forecasting and optimization.
- Creating decision making frameworks to optimize our investments in our riders, drivers, and our marketplace, for example through causal attribution and valuation modeling.
Mapping
This area helps to build and measure the quality of Lyft’s map, map data, and the numerous services built on top of it. A few of the problems that Data Science tackles are:
- Providing the most accurate inputs (travel times, travel distances, driver and rider locations, pickup and drop-off spots) to downstream algorithms.
- Establishing a framework for measuring map accuracy, which includes bug detection for new updates, inferring missing map elements, and establishing application-specific quality criteria.
- Contributing to Lyft’s routing technology in order to generate probable and optimal routes between two points in real-time.
Customer Platforms
The group’s focus is on applying cutting edge data science models and frameworks to our billions of dollars of variable costs such as payments, support, fraud, and insurance. Example projects include:
- Building a telematics platform to convert raw phone sensor data into signals like aggressive braking or phone use so that we can help drivers improve their driving abilities and reduce accident rates.
- Optimizing our routing of transactions across different payment providers to balance costs with long-term value.
- Identifying and classifying the root causes of negative experiences so that we can predict and prevent them for our users.
Lyft Business
Lyft Business focuses on building scalable commute, courtesy, and travel management solutions to help organizations reduce costs and streamline their ground transportation needs. Data Science work includes:
- Driving product decisions by developing frameworks and tools to understand gaps in our current product experience and identifying opportunities for improvement.
- Developing techniques and ROI frameworks to identify opportunities to partner with organizations (like in the airline or hospitality spaces) to provide high-quality travel experiences and loyalty programs.
Research (Marketplace Labs & Economics)
These areas bring scientific and research-based expertise to help solve some of the most complex problems at Lyft and in the broader transportation-as-a-service industry. We collaborate with partners to develop transformative solutions to high-risk yet high-reward problems. Recent efforts include:
- Delivering novel solutions for real-time driver supply, driver earnings, and Shared rides efficiency.
- Leveraging state-of-the-art statistical, machine learning and econometric techniques to improve Lyft’s ability to measure the causal impact on the marketplace of various levers like pricing and new product launches.
- Pairing existing economics literature with econometrics and statistics tools to explore research questions such as: How should we price a new mode? What impact do we think new regulations will have on the business? How do new roads affect travel speeds?
Fleet
Fleet includes Express Drive (Car Rentals for Drivers), Lyft Rentals (Car Rentals for Consumers), and Driver Service Centers. Data Science helps improve Fleet’s growth and profitability. Some examples are:
- Building models to determine optimal pricing to balance demand, supply, and driver risk.
- Developing frameworks that minimize operational costs by identifying risky drivers, improving fleet forecasting, and optimizing retail locations.
- Improving renter retention and engagement by identifying opportunities to improve the funnel and product experience.
TBS (Transit, Bikes & Scooters)
This area develops and supports all of the non-car products at Lyft. Some of the unique focuses are:
- Improving operations by establishing metrics of operational performance and developing mathematical models that optimize on-street product rebalancing, battery-swapping, and maintenance processes.
- Building multimodal routing algorithms and modeling user preferences by integrating transit and micromobility to ensure Lyft users get the right type of ride at the right price at the right time.
- Analyzing data sent by the best-in-class vehicles (Lyft- designed and manufactured bikes and scooters!), and developing fusion algorithms with sensor data to ensure the vehicles are ridden and parked safely.
Level 5 (Autonomous)
This group helps accelerate the rollout of self-driving vehicles onto the Lyft platform. Some sample projects are:
- Measuring the progress of our self-driving technology on the road and in our simulated environment.
- Determining which autonomy features to prioritize for rideshare use cases, and growing self-driving pilots in scale by determining optimal locations.
- Building models to determine the availability of self-driving vehicles for our riders to maximize the utilization of the fleet.
Conclusion
We hope this gives you insight into the Data Science team at Lyft, and the myriad of problems we get to solve day-to-day. The varied expertise of our team paired with the company structure that allows us to work so closely with other functions ensures that we approach every decision with rigor and logic.
If you are interested in joining our incredible team of Data Scientists and improving people’s lives with the world’s best transportation, check out our careers page!
A huge thanks to the following people for their great contributions to this post (in alphabetical order): Akshat Jain, Alexis Weill, Alya Abbott, Ankit Syal, Elizabeth Stone, Eric Smith, Josh Cherry, Mark Grover, Mike Frumin, Neil Chainani, Nicholas Chamandy, Quang Nguyen, Sarah Morse, Varun Madduri, Varun Pattabhiraman, Venkat Devireddy
