Presented By O'Reilly and Cloudera
Make Data Work
March 13–14, 2017: Training
March 14–16, 2017: Tutorials & Conference
San Jose, CA

Organizing for data science: Some unintuitive lessons learned for unlocking value

Eric Colson (Stitch Fix)
2:40pm3:20pm Thursday, March 16, 2017
Secondary topics:  ecommerce, Retail
Average rating: ****.
(4.36, 14 ratings)

Who is this presentation for?

  • Chief algorithms officers, VPs and directors of data science, and leaders that can influence organizations

Prerequisite knowledge

  • General experience with data science

What you'll learn

  • Understand how data science is different with respect to ethos, efficiency of work, and culture
  • Learn how to successfully organize to get value from data science


Building a large data science team is hard. It’s still a nascent practice, and there are no models to follow. Much of what works for established departments like engineering, BI/analytics, and other functional areas does not apply to data science. Don’t force this decidedly square peg into a round hole. Instead, take time to understand the nuances of how a data scientist thinks, works, and thrives and design something completely appropriate for them.

Through his unique experiences of building several large data science teams (80+ people) at companies like Stitch Fix and Netflix, Eric Colson has learned a few things. These lessons aren’t obvious and could have only come through experience, trial and error, and iteration. Join Eric as he recounts some valuable lessons learned, including:

  • Where the data science team should live within the organization: when is it appropriate for data science to be its own department reporting to the CEO?
  • The talent distribution curve for data scientists is not Gaussian, and “average” is actually bad.
  • Develop your own data science unicorns. These mythical creatures with quant skills, software engineering skills, and business acumen do exist—you just need the right environment grow them.
  • Hire early. If it’s obvious that you need to hire a data scientist, then you are too late (even when this rule is applied recursively).
  • Foster innovation without structured “20% time.” Instead, let curiosity (almost) kill your cats.
  • Prefer generalists to specialists and less to more (and, yes, this is consistent with “hire early”).
  • Enable your data scientists through a kick-ass data platform. But for the love of everything sacred and holy in the profession, make them write their own ETL!
  • Compensate them handsomely. . .but not necessarily monetarily.

Some of these may seem unintuitive, and they may not apply to all cultures or organizations. But understanding the nuances may inspire new thinking that allows you to unlock the potential of your data science team.

Photo of Eric Colson

Eric Colson

Stitch Fix

Eric Colson is chief algorithms officer at Stitch Fix, where he leads a team of 100+ data scientists and is responsible for the multitude of algorithms that are pervasive to nearly every function of the company, from merchandise, inventory, and marketing to forecasting and demand, operations, and the styling recommender system. He’s also an advisor to several big data startups. Previously, Eric was vice president of data science and engineering at Netflix. He holds a BA in economics from SFSU, an MS in information systems from GGU, and an MS in management science and engineering from Stanford.