If Interested in a Career in Data Science at Uber, Invest in Becoming a Full-Stack Data Scientist with Production-Level Coding Capabilities
Franziska Bell is a Senior Data Science Manager on the Platform Team at Uber, the multinational transportation network company. She strongly believes data science needs to partner with business and be involved in defining business strategy and new projects. Her team comes from a wide range of backgrounds; she sees diversity of thought as critical to solving challenging problems. She advises students to get hands-on experience and work toward full stack competency.
At Uber, Dr. Bell founded the Anomaly Detection, Forecasting Platform, and Natural Language Platform teams. In addition, she leads Applied Machine Learning, Behavioral Science, and Customer Support Data Science. Before Uber, Franziska was a postdoc at Caltech where she developed a novel, highly accurate approximate quantum molecular dynamics theory to calculate chemical reactions for large, complex systems, such as enzymes. Franziska earned her Ph.D. in theoretical chemistry from UC Berkeley focusing on developing highly accurate, yet computationally efficient approaches which helped unravel the mechanism of non-silicon-based solar cells and properties of organic conductors. She recently responded via email to interview questions from AI Trends Editor John P. Desmond
AI Trends: Could you describe your responsibilities at Uber and how AI fits in?
Franziska Bell: I head the Platform Data Science team at Uber. This is a group of full-stack data scientists and data analysts who build platforms to deliver faster innovation cycles and improved user experiences for internal teams at Uber. The vision for the organization is to transform anyone at Uber into a data scientist at the push of a button. One of our teams consists of world-class domain experts who build platforms and applications for computer vision, forecasting, anomaly detection, sensing and perception, conversational AI / natural language, experimentation, segmentation, and behavioral science in collaboration with their cross-functional counterparts in other parts of the company.
In addition, I also lead Science Identity at Uber, an initiative which focuses on building an internal Science community across our Data Science, Analytics, Research, and Economics teams.
We think of Uber as a company offering a ridesharing service, as well as food delivery. What keeps data scientists busy at Uber?
The core of Uber’s business is a real-time, multi-sided marketplace, which is a lot more complicated than most people realize. I believe that data science is key to Uber’s future due to a combination of (i) the global scale, (ii) need to make real-time decisions, while (iii) adapting to unique local conditions and product preferences of our users.
The only way to tackle this efficiently and effectively is to utilize data science and analytics at scale. As a result, Uber has been heavily investing in data science and analytics. Data scientists and analysts are working across all lines of businesses to enhance the user experience. Their work includes providing restaurant recommendations, identifying more convenient pickup spots for Uber Pool, improving ETA predictions, and providing response suggestions for our riders and driver-partners during the pickup experience.
How did your degrees prepare you for a career in data science?
My PhD (UC Berkeley) and post-doc (Caltech) are in theoretical chemistry and approximate quantum dynamics, which lie at the intersection of applied mathematics and high performance computing with applications towards the physical sciences. The underlying methodologies and skills are also core to Data Science and therefore highly transferable.
How well do data scientists understand business goals?
Be a partner to the business, not a service organization. I strongly believe that data scientists need to be partners to the business and be involved in every step, whether it is defining the business strategy or designing a new project. This requires us data scientists to deeply understand the business and identify opportunities where data science and analytics can have the largest impact. It requires us to think about the user experience first and foremost. My biggest concern with the data science industry currently is that people can get distracted by the newest, shiniest technology and invest in complexity for complexity’s sake, when sometimes simple solutions can get us 80% of the way.
How is natural language processing and conversational AI being used at Uber?
We have several applications of conversational AI and natural language at Uber to improve the user experience. Some examples include:
- One-click chat which is a smart in-app reply system that allows driver partners to respond to incoming rider messages at the push of a button.
- Customer obsession ticket assistant (COTA) assists our customer care representatives by making suggestions about the ticket type, potential actions to be taken, and selecting relevant response templates that the agent can use as a starting point.
Are you able to find the qualified people you need to work on the data science team at Uber?
At Uber we have world-class data scientists and analysts. Data scientists in my organization are full-stack, meaning that the data scientists write production level code and meet a high bar on machine learning, statistics, and problem solving.
Data scientists at Uber come from a variety of different academic backgrounds, including Computer Science, Statistics, Finance, Neuroscience, Chemistry, Physics, Civil Engineering, and many more. At Uber, we also provide mentorship and opportunities for members of other functions (e.g., engineers and data / business analysts) to transition into data scientists. This diversity of thought is critical when solving challenging problems.
You have had increasing responsibility in the management of data scientists at Uber. What are some tips for managing data scientists?
- Helping people grow. I believe one of the most important aspects of any leader is to help people grow. I meet with every person in my organization on a regular cadence to understand their pain points and identify areas where I can help. In addition, everyone in my organization has a career development plan (that extends beyond their work at Uber) and I carry out monthly career development conversations with my direct reports. It’s also important to mentor people outside your reporting chain.
- Be a partner to the business, not a service organization. As covered previously I believe that data science and analytics leaders should be thought partners to the business and be involved in every step, whether it is defining the business strategy or designing a new project. This requires business leaders to give data scientists a seat at the table and for data science leads to think about the user experience first and foremost.
- Invest in best practices. Data science and analytics is a fairly new domain and currently there are no standardized set of best practices for this function across industries. In order to ensure high quality, reproducibility, and consistency across large organizations, my teams and I have heavily invested in developing data science and analytics best practices, including peer reviews of technical documentation and introducing engineering coding best practices. This is a cultural shift that goes beyond data scientists, analysts, and their corresponding leads. It also involves shifting the mindset of product managers and business partners to allow data scientists and analysts sufficient time to incorporate best practices into their project timelines.
- Divide responsibilities by projects, not functions. I believe in data scientists and analysts owning projects end-to-end, versus dividing up a project into (i) analysis tasks that data analysts carry out and (ii) modeling parts that data scientists own. This means that data scientists who are in charge of a project are involved in data collection, quality assurance, data analytics, research of existing methodologies, iterative prototyping, productionization, rollout and monitoring. This approach empowers data analysts and ensures data scientists are closer to the business / product, which leads to more relevant data science solutions.
Do you have any advice for young people interested in pursuing a career in data science in AI in business?
- Hands-on experience. I recommend acquiring hands-on experience, which can be done through internships and similar opportunities. Data science and analytics problems as described in many textbooks and tutorials often do not reflect the challenges we face in industry (e.g., data cleaning, debugging, fast iterations, at scale constraints).
- Invest in becoming a full-stack data scientist. I see several advantages of becoming a full-stack data scientist, i.e., a data scientist who has production-level coding capabilities.
- Ability to design data science approaches with the software stack in mind. Often algorithms that work well during prototyping may not be suitable to run at scale in production given constraints of the (current) infrastructure. Having sufficient depth to understand the stack of your company can allow you to efficiently work around constraints or make suggestions on how to efficiently change the existing infrastructure.
- Self-sufficiency in programming. Being able to write production-level code reduces the dependency on software engineering resources. It is challenging to predict when the iterative prototyping of data science models yields an approach that exceeds the agreed upon success thresholds. As a result, software engineering resources may not be available and a significant lag may occur between finishing a viable prototype and starting the productionization effort.
- Reducing errors. Any handoff step is prone to errors. As a result we avoid transitioning white papers or scripts from the prototyping phase upon productionization between individuals and aim for the data scientist to productionize their algorithm.
Learn more at Franziska Bell’s Uber Engineering page.