How Warby Parker Supercharged Its Data Science Tools

For e-commerce companies, customer behavior is a treasure trove of helpful information. Here’s how Warby Parker uses it.

How Warby Parker Supercharged Its Data Science Tools
[Image: Flickr user fronx]

When it comes to data science, a lot of companies talk the talk. But a lot of this “science” comes down to interesting–but not particularly actionable–correlations. Warby Parker has a different approach. For the eyewear retailer, data are essential feedback that change the way they run their business. But how does a company implement data science in a way that actually drives growth?


“It’s kind of unusual to have a data science team that’s our size very early on in the company growth,” says Carl Anderson, Warby Parker’s director of Data Science. Anderson was hired by the company a little over a year ago–just three years into the company’s existence. He has spent much of the past year revamping the company’s data strategy. This is where he started.

Buckets Full Of Data

Anderson’s approach to data science is a holistic one. In a business, it’s not always clear what use a given dataset might have when one gets down to the brass tacks of actually making business decisions. As such, an effective data science team is one that’s involved on every level and works to make sure that findings are easily found and interpreted for everyone in an organization.

Like a lot of things involving math and statistics, that’s really easy to summarize in a paragraph and difficult to discuss in detail. Anderson does it by breaking down what he and his team does into four discrete “buckets.”

1. Data Engineering

This is the rough stuff, the trailblazing. According to Anderson, data engineering means “getting the data from our different internal systems or from our different vendors, into databases in a form that people can actually use it.” That means normalizing the information and joining together datasets that are otherwise separate.

“All the data was pretty much siloed, and it wasn’t in a form that people could join,” says Anderson of Warby Parker’s earliest data efforts. It’s a common scenario for that see exponential growth. “Everyone was working with Excel spreadsheets, which are great, to some extent–but they don’t scale.”


2. Enabling Your Analysts

Anderson describes the second part of his team’s role as becoming “an enabler” for the analysts. That means doing things like developing business intelligence tools and automating processes that don’t warrant any manpower. It’s cleaning up and anonymizing data, training employees in statistics and SQL databases, and helping them learn how to make useful queries.

If the team’s first role is to make sense of the data, its second is to make sure it’s accessible and not a time-consuming drag to sort through. It’s that necessary middle step toward turning your data into a resource that can be used toward real business decisions.

“That’s one of the things that drives me as a data scientist,” says Anderson. “These insights that start as abstract math, and then you build a tool that answers business questions, and make the world a lot different. “

3. Get Everyone Thinking About Performance Metrics

A big part of Anderson’s strategy is to get the whole organization–that includes employees outside the data team–to start noticing what metrics matter. Once those heuristics are established, the data science team can start to build reports and analysis that help shed light where it’s needed.


“Carl [Anderson] has often come in and asked a question… Are these the right metrics? Are they tightly coupled to the work? Are we measuring them correctly? What’s going on?” says Lon Binder, VP of Technology for Warby Parker. “What’s happened over the last nine months is that we’ve really seen that become a part of the Warby Parker culture across the board, so more and more people are asking those questions, even when Carl’s not in the room.”

4. Actually Do The Data Science

Finally, the last bucket is the sort of highly specialized work that you’d imagine data scientists would do.

“We complement what the data analysts do, and then further it,” says Anderson, “moving on to advanced predictive modeling with classifiers and natural language processing and recommendations engines. Things you might call ‘data products.’ It gives you a sense of [what happens] when data itself becomes an asset in its own right.”

Over the past year, Anderson and his team have equipped Warby Parker with the proper tools to interact with its data. Now, he says, they can have fun “building things that learn from data and build insights.”

Because ultimately, to Anderson data science isn’t just another tool for a company to make use of–it’s an entirely different perspective that a business can operate from.


“It’s making them understand that there’s this whole world out here,” says Anderson of his approach at Warby Parker. “[You can] ask questions of data that answer back in rapid succession. You really talk to the data… [remedying] a lack of tools that were keeping the analysts from reaching full potential, from reaching the best use for the data.”