Tatsiana is the former Director of Data Science at Stitch Fix, where she oversaw the Styling Algorithms Team. Prior to joining Stitch Fix as a Data Science Manager in 2015, she was a Senior Data Scientist at Silicon Valley Data Science, an R&D Developer at Accenture Technology Labs and a Data Scientist at Grokr. She received her B.S. in Pure Mathematics at California State University, and an M.A. in Mathematics at San Francisco State University. You can find her on LinkedIn here, and on Twitter here.
Stitch Fix is on a mission to create hyper-personalized shopping experiences through custom clothing assortments, called a “Fix,” based on a deep understanding of the customer’s style, habits and goals. Through a combination of data and expert insights from stylists, the company curates a custom collection of five items sent directly to clients. Customers can try everything at home, keep what they like and return what they don’t.
Stitch Fix’s algorithms garner a deep understanding of a customer’s taste, fit, color preferences and price point, and provide stylists with a refined, ranked set of items that align with the customer’s profile. There are nuances to personal style that an algorithm might not fully understand, like a Fix specifically designed for vacation, which is where the personal stylist comes in. Algorithms help stylists understand the client and make suggestions, while the stylist, based on their relationship with the client, shapes the final curation.
This combination of humans and algorithms is ultimately what helped Stitch Fix navigate shifts in consumer behavior during the pandemic. With the pivot to remote work, for example, the company saw athleisure sales spike 350%, and searches for the term “work from home” skyrocket. Often, AI has difficulty keeping pace with this kind of rapid change—but Stitch Fix adjusted to these new trends and preferences quickly, as human stylists were able to share actionable insights that an algorithm would otherwise overlook.
Stitch Fix has developed a platform that collects massive amounts of qualitative data to best understand customers. At the start of their journey, customers fill out a robust style profile, which provides the company with approximately 90 points of data to begin the curation process. More recently, Stitch Fix developed the Style Shuffle game, which allows customers to react to clothing suggestions as easily as swiping on a dating app, and share specific feedback on size, fit, color preferences, price and more. The game alone has generated more than 5 billion data points on customer preferences.
Stitch Fix features are designed specifically to engage customers in the process and create a continuous feedback loop. The algorithms then comb through mountains of data and analyze implicit and explicit client preferences, to inform stylists and match customers with the best possible selections. Customers also provide direct feedback on each Fix, informing both the algorithms and stylists for the next order.
In addition to new features that enhance the algorithms, like Style Shuffle, Stitch Fix continues to launch new features to cater to evolving customer needs. Recently, the company launched a direct-buy offering called “Shop,” where customers can browse and buy curated collections as outfits or by category. In the UK, Stitch Fix also launched “Fix Preview,” a form of window shopping which allows customers to collaborate with the stylists and algorithms in real time as they pick their five items, rather than browse thousands of pages for online inventory or wait until the items arrive at their home.
Stitch Fix has already seen incredible engagement from these new features. Why? Describing style is really hard, but a picture can speak louder than a thousand words.
In today’s world of retail, decision fatigue is pervasive, whether in considering brands, fits or styles. Stylists working with customers one-on-one are able to deliver personalized services that eliminate the anxiety of finding the right item. But doing so at scale – matching millions of consumers with five items out of thousands based on individual preferences – is far more challenging.
Tatsiana believes retail, and fashion more specifically, will continue to move towards an effortless experience where everything is hyper-personalized and curated specifically for individuals. Customers won’t have to ask: Will this fit me or look good on me? Will this fit an occasion? Instead, they will simply feel confident that they have the best possible item for their unique needs and preferences.
EPISODE 24: Tatsiana Maskalevich
Jim Freeze Hi! And welcome. I’m Jim Freeze, and this is The ConversAItion, a podcast airing viewpoints on the impact of artificial intelligence on business and society.
Stitch Fix, a popular online styling service that delivers curated clothing choices to its shoppers, has been a pioneer of AI-powered personalized experiences since its founding in 2011. Today, the company has 3.9 million clients that shop from 1,000 brands, assisted by over 5,000 stylists and 8,000 employees.
On today’s episode, we hear from Tatsiana Maskalevich, who up until recently was the company’s Director of Data Science. Before she embarked on her new venture, she joined us to chat about the plethora of data that informs each “Fix,” how the company keeps pace with evolving shopper preferences—like our new found reliance on athleisure and sweatpants—and the idea that stylists and algorithms are better together. Let’s dive right in.
First, let’s start off with a little bit of your background. How did you initially become interested in data science, and what ultimately drew you to Stitch Fix back in, I think, 2015?
Tatsiana Maskalevich Well, now thinking about it, data science wasn’t even around when I first started thinking about science itself. I think I remember around 2011, when people started writing, Wall Street Journal, and I still remember we would read their articles about machine learning, which was fascinating.
But back in the day, we still were doing statistics, we still were doing engineering. We were doing data engineering and analytics, we just didn’t call it data science. So long story short, I moved to the United States about 17 years ago. One of the first things I did, I took a class in personal finance. And I’m originally from Belarus, I studied economics and management in Belarus. So I was thinking, when I come to the United States, I will continue the same trajectory and career. I was kind of interested in finance. The professor and the college, this community college that I went to, he was actually quite great and got me really interested even further, to pursue finance through the personal finance class. I came to him asking for advice, like, what do I do? Which college do I apply to? What major to pick? He asked me this really profound question. He said, do you like math or computer science? I said, well, that’s a great question. I actually do really like math. I went to math high school before, my studies in Belarus were heavily directed towards math. So he said, why don’t you just inquire about going into mathematics?
And the rest is history. I did pursue mathematics and went to grad school. Around 2012, I was just so fascinated about this whole machine learning discipline. So started my career by first having an internship at a startup, then joining shortly after; doing a short stint in consulting about three years; and finally in 2015, meeting Eric Colson at a coffee shop. I vividly remember him bringing in his small notebook, drawing some graphs and telling me all about Stitch Fix, telling me why I have to join.
Jim Freeze He was very persuasive wasn’t he?
Tatsiana Maskalevich He was very persuasive and also very secretive. He wouldn’t tell me a lot about the company. That was fascinating to me because they had something amazing. They were also really, really quiet about these amazing algorithms they developed, and then just amazing traction that they had with their clients. But I guess Eric won me over with graphs and data. I was immediately interested and the rest is history.
Jim Freeze Well, seven years later, you’re still there. So he obviously was very effective. By the way, you’re in good company as a person who likes math. My undergraduate and my master’s degree are in mathematics. So I always love talking with people who love math. I do as well.
Tatsiana Maskalevich Yeah, that’s really fun. I mean, it’s interesting, people always ask, what are you going to do with a math degree? Especially, it was interesting during the undergrad. I don’t know if people ask you about that, but I never would have even imagined that I would be doing math and fashion, applying all of the things that I’ve learned, both in undergrad and the grad school in a day to day, helping people to be and stay fashionable. It’s just fascinating.
Jim Freeze It’s great that math is fashionable. So today you’re the Director of Data Science, overseeing the styling algorithms team. Can you talk a little bit about the role data science plays in the shopper experience and having Stitch Fix match clients with the right stylist, and ultimately deliver the right clothing recommendations?
Tatsiana Maskalevich Yeah, absolutely. I want to first start by maybe telling you a little bit about Stitch Fix, and the listeners as well. Stitch Fix started in 2011 and we initially started by pairing shoppers with their personal stylist, they did it by curating an assortment of five items. In a way, it works where our clients, when they come in, they fill out a style profile, which is a pretty robust set of questions, from which we collect about 90 points of data. Then after they’ve filled out the profile, that’s where the magic happens. Algorithms and stylists work together to curate an extremely personal collection of items, then they are sent to the clients, to their homes, which is really amazing because you can try everything at home. You can pair with the items you have in your closet. Then if you like something, you keep. If you don’t like it, you send it back.
So that’s where Stitch Fix started, and now we are just adding so many new features. For example, we launched direct buy offering which we call “Shop,” and our clients actually can shop directly curated collections that we present them with both in the form of outfits. Or our recent feature, we just launched, which is a shop by category.
All of those interactions, all of those moments of connections, create very valuable data for both algorithms and stylists to really connect with our clients. Jim, I was researching this podcast and I love the name of it. I was thinking about Stitch Fix and the journey that our clients take with our company as literally a conversation, both figuratively and literally, because clients constantly share a lot of information with us, both through feedback that they provide. Also, we have a lot of moments, like explicitly and implicitly, that they communicate with stylists and then stylists communicate back with them.
So over time, for example, I have a client who I’ve styled for six years, it’s totally a conversation that you’re having. We kind of, on the data science side, as well as on the stylist side, just looking and observing, enjoying the journey together with clients, their style journey, their personal journey, their career journey, you name it. So, that’s where Stitch Fix is at. As far as the data science team, and as far as my career, we have a pretty large team. My team specifically focuses on the styling part, which is what I call where the magic happens, where we bring our clients and our merchandise together. As I said, we do it directly, we do it through stylists and these curated fixes that we ship to their homes.
Jim Freeze Yeah. So one of the things I find so interesting about your business is that it’s something we relate to at Interactions, the company I work for, which is that we have a very strong point of view that AI needs people and people need AI. The two of them working together make better outcomes, which is, that as you talk about Stitch Fix, it’s pretty clear that algorithms and people are in fact working together and the algorithms need data. So could you talk a little bit more about how you garner more and more data to facilitate what you characterize, I think correctly, as a conversation between stylists and your customers, and how you can kind of optimize what you can deliver to them in terms of recommendations, as the conversation progresses.
Tatsiana Maskalevich Absolutely. Yeah. Well, again, data at Stitch Fix is pretty amazing. We say the algorithms need data, but also humans need data right? We are really drawn to patterns. We like to connect and understand. So what happens at scale, right? When you are, for example, working with one customer, perhaps human as a human, you can make a lot of decisions, but when you work with many customers and then you multiply it by thousands of items in our inventory, decisions become very challenging. Now, if you add a third dimension with all of the feedback we collect from clients, we also have the tool or feature on our app and the website called Style Shuffle. Clients can play this game, where they thumbs up and thumbs down merchandise that helps us, and them, to discover their personal style. There’s a lot of data points to sift through.
And so what happens is, our algorithm is designed in such a way to come through all of the possible data and the merchandise, and also understand both implicit and explicit client preferences, then match it all together and present stylists or clients in our direct buy offering with the best possible selection for a client at that time. Right? So that’s what the algorithms do, the classic crowd calculation, and we use a lot of different data points. To your question, we collect a lot of data continuously, right? The style shuffle, the clients give us a lot of feedback from size and fit, their color preference, as well as for each garment, whether the price was right for them. They also leave unstructured feedback.They leave unstructured requests, meaning just a flowing text of data. We use algorithms to process all of that.
But when stylists come in, they have much more refined and ranked sets of items to consider. So it’s no longer sifting through thousands of pieces of information that clients might leave to us and to a stylist to consider, it’s much more organized for stylists. Also, they are seeing smaller subsets of ranked merchandise that take in consideration everything from color preference, to fit, to price preference, and so forth. However, sometimes clients leave very, kind of, amazing gold nuggets in their requests or communication with stylists.
For example, I was just styling a fix the other day, one of my clients was super excited. She was sharing that she is going to Mexico for a girls’ trip. You can imagine in the times of COVID, that’s just probably very exciting for people to finally start traveling. So she was asking for some picks for her vacation, given where we were at and how we transition seasons, some of those picks might not be obvious to the algorithm because it’s just a change in context. So as a stylist, we also give them a lot of tools to refine, inject that new information, and find those specific and personalized items to kind of sometimes even override the algorithm because they just have this nuance and context.
Jim Freeze Yeah. What you’re talking about is very interesting coming out of the pandemic, and it’s pretty difficult to discuss any business and not talk about the pandemic and the impact it’s had on the business.
It’s pretty clear that the pandemic has had a big impact on how all of us dress. I read in Fortune, that in Q2 of last year when the pandemic was really hitting us and things were changing dramatically, Stitch Fix saw sales of athleisure items spike 350%. I heard another podcast you did, that searches for the term work from home, also skyrocketed on your platform. To the point you just made, AI sometimes has difficulty dealing with a big shift like that, because it doesn’t necessarily have the data sets to deal with that. So you talk a little bit about how you navigate that for customers, but you could drill down a little bit more on that, because I think that’s really interesting? You’re probably having to deal with the same thing right now, given the example you just gave about coming out of the pandemic too.
Tatsiana Maskalevich Yeah. This is such a great question. I guess to the point you’ve made earlier in our conversation, where algorithms and humans are better together, right? They can’t pinpoint on those changing trends quickly. We have the system, like the overall system, not just within styling, that is very robust, that can adjust quickly just as we reacted to the change in trend, during the pandemic with athleisure, more of a comfort wear and work from home trends. We are also very capable of reacting to changing trends as we are coming out of pandemic. So to the example with the client I just styled, I was able to find really cute picks for her that could be really great for a vacation.
It’s actually kind of fascinating that because at our volume, we can very quickly translate this kind of data and change in preference. The signal that is not, could be only utilized by stylists, but also our merchants. That data just flows very quickly and again, because we have the human in the loop, we can quickly translate it into very actionable insight, which sometimes for algorithms themselves might take a little longer because that nuance, the shift in trend, might not be picked up right away by the systems.
Jim Freeze That’s a great example of the point about humans and AI working together, or as you characterize it, humans and algorithms working together. I know Stitch Fix is passionate about ongoing innovation. I think you’ve recently rolled out new features, including Fix Preview and Live Styling. Can you talk about these updates and how it changes the shopper experience?
Tatsiana Maskalevich Yeah, absolutely. I’m glad you asked this question. I’ve personally been working on Fix Preview, which is for everyone. This is a really exciting new feature that we rolled out in the UK, we’re working on rolling out that for the U.S. geography. Basically how it works is that, rather than you waiting for five items to arrive to your home, you have an opportunity to actually collaborate with stylists and algorithms while we are picking five items that are personalized to you. So what we effectively did is introduce a kind of window, it’s almost like window shopping. So rather than browsing thousands of pages online for the inventory, you can browse picks that stylists are considering, that they’re planning to send to you and your fix.
We’ve seen really amazing engagement with that, I think it’s really fun. On the algorithm side, it’s really great because we are getting more data points from clients, their reaction, their feedback on the picks that both stylists and algorithms are picking for them. For clients, it’s also fun to engage and also provide additional information and something fun that this product ended up demonstrating is that people really love to engage with things. Sometimes when they see, they can describe things with words, but when they see a picture of an item, it’s much easier for them to say, oh yeah that’s what I was thinking about. Because what we found out is that describing style is actually really hard, but when you see a picture, it’s like a picture speaks louder than a thousand words. So we’re really excited about this feature, the team has been hard at work and we’re really excited for our clients to experience it.
Jim Freeze Yeah, it sounds really cool. I have just one kind of last question for you. I always like to kind of ask guests on The ConversAItion to be prognosticators about the future. Stitch Fix has, of course, been a very successful pioneer in curated shopping experiences. What do you think is next for the space? Where do you still see opportunity and room for improvement?
Tatsiana Maskalevich Well, wonderful question. I do see for my personal take, and things that really excite me and my team at Stitch Fix, is to continue personalizing the space. When online shopping started, it was very convenient for everyone. We actually demonstrated that last year, just to re-emphasize that online shopping is extremely convenient. However, if you try shopping for something, the choice anxiety is real. You have a lot of things to choose from. Both the fatigue of making the choice of considering the option and also knowing what’s best. You can go and read some reviews online about each garment that’s presented to you. Oftentimes they don’t even exist because fashion, fit, size are so subjective. What I see happening both in retail, but also specifically in fashion, is continuing kind of moving towards this personalized and effortless experience where you go in and everything is personalized for you.
You go into your specifically curated boutique shop and you no longer have anxiety, will this fit me? Will this look good on me? Will this fit an occasion? Then that opens up like a real door towards a whole family thing, right? We already see this at Stitch Fix, we serve men’s, women’s, and kids. Our clients, sometimes family, coordinate outfits and they might have family pictures, they might have some events. So I could see the whole industry kind of moving towards that extremely personalized shopping experience that is very much curated for each individual.
Jim Freeze I think that’s a very compelling vision of the future. Touchdown. It’s been great talking to you. It’s been fascinating hearing about what you’re doing at Stitch Fix, the role that artificial intelligence plays in understanding and catering to a shopper’s unique style. What you just added in terms of furthering it with personalization is very, very compelling. Thank you so much for being on the ConversAItion today.
Tatsiana Maskalevich Thanks, Jim. It was my pleasure and I’m glad we chatted.
Jim Freeze That’s all for this episode of The ConversAItion. On our next episode, we’ll be joined by Ella Alkalay Schreiber, Vice President of Fintech & Data at Hopper, a travel company that leverages AI to predict flight and hotel prices, and offer consumers personalized booking recommendations. We’ll discuss how AI addresses common pain points in the booking process, and how Hopper has navigated volatility in the travel industry over the last year.
This episode of The ConversAItion podcast was produced by Interactions, a Boston-area conversational AI company. I’m Jim Freeze, and we’ll see you next time.