How Stitch Fix Relies on Data Science to Power Its Business

Fast-growing apparel subscription service Stitch Fix (SFIX -4.44%) believes that its data and algorithms give it a massive advantage over traditional retailers. Customers provide the company with loads of information about what they're looking for, allowing the company's stylists to provide the right recommendations.

In this episode of Industry Focus: Consumer Goods, the team takes a deeper look at how Stitch Fix uses data to offer a better service and products. But is it possible management is overhyping the data side of its business in order to get investors to view it as a tech company rather than a retailer?

A full transcript follows the video.

This video was recorded on Dec. 12, 2017.

Vincent Shen: The thing that really impressed and fascinated me about Stitch Fix in particular is they really try to combine apparel and retail, which is an industry that's been around for hundreds and hundreds of years, and the company has a huge focus on using data to best serve customers, to win their loyalty. When I was reading the company overview and description in the prospectus, it struck me almost immediately how often the company mentions data or data science. It feels like these words come up in almost every other sentence. I don't know if you noticed that, too, Adam.

Adam Levine-Weinberg: I did. We can talk about this maybe a little bit more later, but it's definitely one of the things that's really interesting about the company. It's also a little worrisome. It makes me wonder if they're trying too hard to seem like a tech company when they're really just another retailer. You've seen this problem before, because tech companies are getting such high valuations in the market these days, everyone wants to seem like they're really just a tech company and everything else they do is incidental to that. So you definitely see Stitch Fix talking a lot about their data, but everything still goes to a human. The data is providing recommendations to the stylists for fitting and product choices, but the stylists are still interacting and deciding which five items go in the box to the customer. So the question is really: How much of this is the data science informing, and how much is the stylists using their own intuition?

Shen: Yeah. When you look at the breakdown you mentioned, they have 75 data scientists who help work on these algorithms, but they also have their team of over 3,400 human stylists who help finalize these orders. You have to wonder how much of it is, in terms of the promotional side, is actually driving their business. But right now, let's cover some of the stuff on their data before we get into the results that's actually generating in terms of financials.

We know a lot of, for example, brick-and-mortar retailers right now, they are rolling out things like royalty programs, they have mobile apps, they want to collect a lot of data on their customers. And I think Stitch Fix is definitely in an enviable position where their customers give them 100% voluntarily everything the company needs to optimize and curate their offerings for each person. I have a quote here from the company prospectus that gives you an idea of what tools and information they can leverage when assembling an order or a Fix, as they call it, for a customer. It's long, but it tells you a lot about what they have at their disposal:

On average, each client directly provides us with over 85 meaningful data points through his or her style profile, including detailed size, style, fit and price preferences, as well as unique input, such as how often he or she dresses for certain occasions, or which parts of his or her body the client likes to flaunt or cover up. Over time, through their feedback on Fixes they receive, clients share additional information about the preferences as well as detailed data about both the merchandise they keep and return. Historically, over 85% of our shipments have resulted in direct client feedback. This feedback loop drives important network effects, as our client-provided data informs not only our personalization capabilities for that specific client, but also helps us better serve other clients.

So you take all the information that company has on your sizing, your other preferences, and they also apply a similar mindset to all the product they have in inventory. They'll track a lot of things about each piece of clothing, for example, like the brand, the size, the color, the material, but they'll also supplement that with the item measurements, with the more qualitative descriptions, and some of the client feedback that they have received on that item when sending it out in previous Fixes. I guess some of the magic there is that Stitch Fix can plug all of that into their algorithms, the information they have on their customers and their products, and that algorithm will make suggestions on what to offer in each order, and it comes out with the probability that the item will match a specific customer. Again, as you mentioned, you have that probability number now, but that still gets fed through a human stylist, one of the team of 3,400 or so. They provide the finishing touches, and they will further curate or finalize that order for each person.

So if we try and quantify how well the algorithms and stylists are picking products for customers, something I thought was pretty telling was, the number of items purchased per Fix was up 22% in 2017 over 2014. So definitely good progress to be seen there. The company overall takes the same kind of data-driven approach to forecast demand, plan its inventory. They use it to optimize the distribution to their five U.S. fulfillment centers.