Name: Inside The Algorithm Powering Stitch Fix
Uploaded: 2019-10-29
Duration: 5 min 41 s
Description: And why humans are still a key ingredient.

[ MUSIC ] I think a lot of people think of Stitch Fix as subscription service for clothes, but that's not Not the entire picture. So just quickly in your words, what is Stitch Fix? Yeah, so what's really special about Stitch Fix is that we're about personalization. We get to know our clients, we get to know your size and style preferences and then through a combination of, Data Science, understanding our products and the feedback that you share with us and human stylists, then we are able to send you product to try and in your home for you to be able to find things that you love easily send things back that you don't love. It's a very high revenue engagement model, but really what drives that is this personalization and the ability to really get to know you and your family. Preferences and what you're gonna love. So our clients come to Stitch Fix. com, fill out a profile, which tells us a lot about them. They answer questions from their fit preferences, what colours they like to wear, or maybe prints that they want to avoid. They get assigned to a stylist. Like myself, who read through all their profile information to understand who they are as a person and then from that information I as a stylist and partnership with the algorithm, Select five items that will help them feel like their best self and that really match who they are as a person. So as a stylist so I sell fixes myself, I actually can see all the product recommended for you. It already will be only things that are relevant to your size based on your preferences, your likes and dislikes. And then it actually lets me know a recommendation score and so it will say these jeans have a 76 Percent chance of being kept, this cardigan, has a 69 % chance of being kept. And that's a really helpful tool for me as a stylist because now, I can say, okay, with confidence, I'm gonna these jeans, I'm gonna do this cardigan, I'm gonna take a little bit of a risk with this top. And so those scores really helped to inform the stylist to be able to Efficiently but also really accurately get you products that you're gonna love in your fix. [ MUSIC ] If the system tells a stylist that I'm 76 % likely to keep a certain pair of jeans, how does it know that? The underlying data is really, really special in our business. And so we have this differentiated data that we really think nobody else has. We have this amazing data about you. When you're signing up you're letting us know these are the things I like, these are the types of things I'm looking for, these are my dimensions. And then on the product side all the product we have in house so we can take the measurements that we know are relevant and important so that we can know that this is actually fitting more like a smaller extra small versus a larger extra small and that's something that we can actually learn about the product And then the most valuable part is actually what happens at the intersection of those two things. 85 % of the time that somebody is getting a fix, they are sharing with us some level of data about what they thought, I like this. I didn't like this. This was too cheap. This is too expensive. This didn't fit right. And the fact that we get all of that data is what really helps us to be able to have this super actionable super accurate recommendation that then. We can learn from so that the next time we have you as a client or the next time we have someone who even looks a little bit like you as a client in the future, we're able to have really predictive information around what your preferences mean in terms of the product. So our stylists have an amazing platform that I use to style our clients. Some of our clients have been with us for 50- plus fixes. So say down the line somewhere they mention that they didn't really like a cream sweater, cuz cream washes them out. Well, that is a heavy lift for our stylist to try to uncover. So we have a tool that helps surface that information for our stylist to make the best informed decision that will best serve our client. Our clients also have tools available to them. One of those is style shuffle where they really get to interact and engage with the pieces that we have in our inventory, by telling us if they like them, or they don't like. Kind of started out as a little bit of an experiment, of just like let's figure out ways to understand people's style in a more dynamic way. And it's just a really rapid response way to just get your gut on, do you like this or not? We found that the style shuffle data I mean this point we have 3 billion readings. That data has actually with statistical significance helped us to improve outcomes. Of course for people who play Style Shuffle. So if you Style Shuffle we are more likely getting you think that you're gonna love in your fixes. But we actually have found that it even helps people who don't play Style Shuffle, because we can learn from your experiences and be able to extract from that. And what the preference of others might be. Can a stylist override the system and give you something, send you something in your fix that the algorithms aren't recommending? For sure. [ LAUGH ] And I think that's part of like the real humanness of Stitch Fix is that there's gonna be times You may be asking for something really specific in your fix, that the algorithm may not be able to interpret but I could. We've had requests notes like, you know, my husband is coming back from a deployment, or I'm going to be hosting a conference. These use cases where I think the human knowledge and the human context can really be helpful. And so those might be examples where you might go beyond the bounds of Jessica. Kind of what a normal recommendation is. So it truly is a balance of art and science with the algorithm of supporting us in their decision making and also reiterating the choices that we're making, through the recommendations that it brings up to us. So it really allows me the space to create those human connections deliver a really personalized fix. And build those relationships with my client. Because searching through unless inventory that we have in our warehouses is not really topping into my creative side which is what brought me to Stitch Fix in the first place. [ MUSIC ]

Inside The Algorithm Powering Stitch Fix

Tech

And why humans are still a key ingredient.