Building and deploying effective data science teams Nikita Lytkin, Ph.D.
Introduction Ph.D. in Computer Science, Machine Learning (Rutgers University) Postdoc in Machine Learning for Genomics (NYU School of Medicine) Lead Data Scientist in Online Advertising (Quantcast) Senior Data Scientist in Data Products (LinkedIn)
Motivations Intense competition for talent Lack of established guidelines on how to interview and build data science teams Personal qualities are crucial to success of a data scientist, yet are often underemphasized in talent selection process
Overview What do data scientists do? Building data science teams Leading data science teams
What do data scientists do?
Creating new data products
Creating new data products
Creating new data products
Creating new data products
Winning presidential elections Source: h*p://en.wikipedia.org/wiki/obama_logo
Insights and analytics Understanding the customer o Identifying drivers of customer retention and growth Understanding performance of existing products o Recommendations on how to improve performance Business forecasting and creation of actionable metrics that correlate with business objectives
Building data science teams
Where do data scientists come from? Strong quantitative backgrounds Experimenter s mindset of forming and testing hypotheses Advanced degrees from a broad range of fields o Computer Science, Statistics, Mathematics, Physics, Theoretical Chemistry, Operations Research, Neuroscience, Engineering, Economics,
Core technical competencies Strong analytical ability Reasoning with data: asking questions and obtaining answers Statistical inference and Machine Learning Mathematical optimization Principles of software engineering
Personality characteristics Mindset over the dataset o Attitude and character are as important as technical skills, but are much harder to develop and are often overlooked Creativity Initiative Thirst for learning
Creativity Loves asking meaningful questions and generates ideas Persistently explores space of possible solutions Effectively manages ambiguity
Initiative Strives for impact o Clearly articulates motivation for the work Takes ownership and responsibility o Makes recommendations independently Hungry for challenge and growth
Selecting talent Complementarity of interests and strengths, and variety of backgrounds help drive innovation Who are the future leaders in your data science organization? o They can help take charge when the team grows and act as catalysts continuously motivating the group
Leading data science teams
Lesson #1: Encouraging autonomy Matching projects with interests Encouraging team members to take on lead roles on projects Providing space for exploration o Soliciting project proposals o 20% projects o Blue sky sprints/hack-a-thons
Lesson #2: Managing uncertainty Portfolios of projects with a mix of uncertainty profiles increase likelihood of team members obtaining positive results and helps maintain morale Evaluating member s performance based on quality of execution vs. experimental results
Lesson #3: Team development Forming project groups o Fosters exchange of ideas o Mitigates isolation Study groups and brainstorming sessions Conference attendances o Keep team informed of most recent developments o Networking opportunities Collaborations with researchers in academia
Feedback welcome: www.linkedin.com/in/nikitalytkin