Machine Learning Times
Machine Learning Times
EXCLUSIVE HIGHLIGHTS
Effective Machine Learning Needs Leadership — Not AI Hype
 Originally published in BigThink, Feb 12, 2024.  Excerpted from The...
Today’s AI Won’t Radically Transform Society, But It’s Already Reshaping Business
 Originally published in Fast Company, Jan 5, 2024. Eric...
Calculating Customer Potential with Share of Wallet
 No question about it: We, as consumers have our...
A University Curriculum Supplement to Teach a Business Framework for ML Deployment
    In 2023, as a visiting analytics professor...
SHARE THIS:

CONTINUE READING: Access the complete article at multithreaded.stitchfix, where it was originally published.  

8 years ago
Data Science at Stitch Fix

 

Over the last couple of years, Stitch Fix has amassed one of the more impressive data science teams around. The team has grown to 65 people, collaborates with all areas of the business, and has a well-respected data science blog plus several open source contributions.As a member of this team since late 2014, and someone who has spent 15 years in the analytics space prior to that, I’ve often reflected on how the data science team at Stitch Fix got to this point. Is it attributable to our business model? Or, is Stitch Fix doing something differently when it comes to growing and managing its data science team?

The short answer is that the business model does provide a fertile environment for data science. However, it goes deeper than that: the approach to managing and building data science teams at Stitch Fix is unique in many ways. In fact, it has debunked many of the beliefs I held about management and growth prior to joining the team.

  xkcd

Busted Belief #1: Project Management Software to the Rescue

My belief was always that project management tools are an effective way to promote efficiency. By entering requests into a system and reporting on project status, nothing would supposedly slip through the cracks and stakeholders would have full visibility into the status of any given initiative.

I was mostly wrong.

Project management tools are great for projects with defined milestones that need to be executed in a certain order, whether the projects involve data science or not.

However, for the more typical, unstructured data science projects, project management tools do more harm than good. Here are some examples of adverse effects:

  • Lack of context. People feel less compelled to provide context and reasoning when requests can be made through a tool. Less context makes it harder for data scientists to get to the root cause.
  • Lower morale. Receiving requests through a tool with little context is a soul-gutting experience for any data scientist. It also creates an uneven partnership where business partners become customers and data scientists become order-takers.
  • Waste of time. It is fairly cheap to submit requests through a tool. The tool is not going to push back or ask questions. And as a result, you can end up working on stale problems or someone’s pet project.

In an environment without project management tools, humans will communicate naturally. This creates a “survival of the fittest” environment for projects where only the important projects will make it into the queue to begin with. This sets the bar higher: business partners have to be sufficiently motivated to explain the problem that needs to be solved. It also ensures a subsequent dialogue between business partners and data scientists.

Busted Belief #2: Turn Life into School

There’s a famous quote that sums up the dilemma of training in the workplace: “In school, you’re taught a lesson and then given a test. In life, you’re given a test that teaches you a lesson.”

In the corporate world, we often try to turn life into school with extensive training and development plans. Telling people how to do things before they do them seems like a logical approach to avoiding future mistakes. I certainly leaned towards this theory before I joined Stitch Fix. And, to be fair, there are many areas where training is an absolute necessity.

But, at Stitch Fix I’ve realized that there is an even better way when it comes to promoting growth for data scientists: Create an environment where people are encouraged to seek out new challenges and gain experience from the “battlefield.” Let people learn the lessons from the “test” through instant feedback – not months later during a scheduled review process. In this environment, data scientists expand their scope by going after opportunities and owning them, which they won’t if they’re just following a scripted development plan.

This approach, which builds on trust and experiences, is also the best way to foster team bonding. True team bonding – at least between data scientists – is not created by playing a scheduled game of frisbee during a company off-site or role-games (although that can be fun). Bonding happens when you work intensely together as a team in a room with a whiteboard. I’ve found that this setting occurs naturally in the consulting industry where small groups of people travel and work together for an extended period of time. However, it’s less likely to take place in an office environment where a myriad of meetings tend to pull people in different directions. The solution? Send people off-site to work through problems when they’re hitting a roadblock, starting a project, or caught in a fire-drill. Clear their calendars for those days – most meetings can easily wait. Not only will this promote team bonding, it will also yield creative solutions that may not have been invented in the office.

Busted Belief #3: Hiring Success is Dictated by Your Circumstances

I used to approach the hiring process as follows: contact people I know and trust, post job descriptions and wait for recruiters to provide good leads.

During this process I’d typically gripe about the quality of the leads I was getting from recruiters, as well as the challenge of competing against more recognizable brand names.

There was no need to gripe.

First, if your company’s brand is not well-known across the globe, it doesn’t mean that you cannot have a strong tech brand. You can be known in the community for your tech talent and innovation and that can mean even more than a brand name to a data scientist. In fact, I would argue that, even for the most well-known companies, having a strong tech brand is key in order attract top-notch data science talent. Companies with appealing tech brands write blog posts, host meetups, and provide open source contributions. Is it hard to find the time to do this kind of stuff? Absolutely. But the payoff is huge.

Second, I never fully utilized one of the better recruiters that I know: myself. I know the role I’m hiring for better than anyone else, and I know exactly the type of background I’m looking for in a new recruit. Surely, the short-term pain of spending days scouring LinkedIn and “cold-calling” people is significant, but it pales in comparison to being understaffed or hiring the wrong people for the team.

Confirmed Belief: Cool Work Attracts Talented People and Talented People Create Cool Work

There’s a symbiotic relationship between a data-driven business model and a management style that does away with the long-held beliefs about data science teams. Stitch Fix is working on a wide array of interesting problems that makes it an enticing place to work for data scientists. But these same data scientists are using their skills and innovative mindsets to spot opportunities and initiate projects. This goes back to hiring the right people and giving the them freedom to innovate and grow.

Here is a partial list of areas that we’re actively working on today:

  • Recommendation systems – empowering our stylists to pick the best items.
  • Churn modeling and lifetime-value estimation – personalizing the client experience.
  • Inventory management – what to buy, how much, and when.
  • Human computation – combining the strengths of humans and computers.
  • Quantitative psychology – asking the right questions to collect the best data.
  • Transportation problems – where to ship from and which carrier to use.
  • Demand forecasting and long-term planning – balancing demand and supply.
  • Marketing attribution – assigning the appropriate credit to each marketing touch point.
  • Natural language processing – leveraging cutting edge techniques to listen.
  • Platforms – building robust and flexible platforms for data science.
  • Computer vision and deep learning – extracting properties from images.
  • Designing new styles from data – enough said.

Kim Larsen, Director of Client Algorithms, Stitch Fix
Originally published at http://multithreaded.stitchfix.com

Author Bio:

Kim LarsenKim Larsen is the Director of Client Algorithms at Stitch Fix in San Francisco. Kim has experience working in the software and consulting industries, as well as e-commerce and financial services. Throughout his career, he has managed a wide array of data mining and analytical problems including price optimization, media mix optimization, demand forecasting, customer segmentation, and predictive modeling. Kim frequently speaks at data mining conferences around the world in the areas of segmentation and predictive modeling. His main areas of research include additive non-linear modeling and net lift models (incremental lift models).

Leave a Reply