Style is a very personal part of what makes someone who they are. The way you dress is a reflection of who you are or who you want to be, and what speaks to you may be totally foreign to the next person. Knowing all of that, it’s understandable if you believe that something as personal and experience-driven as style could never be boiled down to data points or plugged into an algorithm. But… you’d be wrong.

At Stitch Fix, a combination of human stylists, powerful A.I., and behind-the-scenes technology has created a winning model that delivers a personalized online shopping and styling experience straight to clients’ homes. A powerful data science team is one of the key reasons that Stitch Fix has been able to launch its valuation into the billions. Stephanie Yee is the VP of Data Science at Stitch Fix, and on this episode of Up Next in Commerce, she explains all the ways that data and technology are being put to use to create the best customer experience possible.

Stephanie describes how technology like GPT-3 is going to finally make seemingly unimportant data consumable to a consumer audience, and she explains how an event like COVID-19 can impact your backend models and what to do to adapt in that situation. Plus, she gives tips on how any ecommerce operation can go about building a data science team, and the soft skills to focus on when hiring talent.

Main Takeaways:

• Asking the Right Questions: The most important skill a data scientist can have has nothing to do with technical prowess. It’s about having the ability to frame a problem and then ask and answer the right questions. Encourage your team or new candidates to pump the brakes and reevaluate the “why” behind the question they are trying to answer or the problem they are trying to solve.
• Making The Indecipherable Easily Digestible: With the shifting demographics, and older generations now becoming more comfortable shopping online, tools need to be created to ingest and answer long-form questions in a way the consumer connects with. Technology like GPT-3, which is the most advanced language model to date, has the ability to do just this. Tune in to hear how!
• The Quick Change: Deploying algorithms and A.I. in conjunction with human resources/industry experts is critical for organizations to be able to adapt to big changes in a market. COVID-19 had a drastic impact on models that were trained on pre-COVID data. Should you scrap your current model and start over? Or build on what you have? Stephanie says a little bit of both.

For an in-depth look at this episode, check out the full transcript below. Quotes have been edited for clarity and length.

---

Up Next in Commerce is brought to you by Salesforce Commerce Cloud. Respond quickly to changing customer needs with flexible Ecommerce connected to marketing, sales, and service. Deliver intelligent commerce experiences your customers can trust, across every channel. Together, we’re ready for what’s next in commerce. Learn more at salesforce.com/commerce

---

Transcript:

Stephanie Postles:

Welcome back to Up Next in Commerce. This is your host, Stephanie Postles, co-founder of mission.org. Today on the show, we have Stephanie Yee, the VP of data science at Stitch Fix. Stephanie, welcome.

Stephanie Yee:

Thank you. I'm excited to be here.

Stephanie Postles:

Me, too. I know it's going to be a good interview when there's two Stephanies, but I'm slightly worried about how the transcript will look. Like who's saying what? Who sounds smart? I'll just take all your quotes and pretend they're mine.

Stephanie Yee:

Perfect.

Stephanie Postles:

So tell me a little bit how long have you been at Stitch Fix for?

Stephanie Yee:

I've been at Stitch Fix for almost four years. Yeah, four years in January.

Stephanie Postles:

Well, tell me a little bit what does the role of the VP of data science look like day-to-day?

Stephanie Yee:

Yeah. If I have to think about it, being the VP of data science, it really comes down to maximizing the value that the data science itself and the team can bring to the company, like how do we really get the full promise of an algorithm's approach to things? I think as you guys probably know, Stitch Fix is really thinking about how do we help people find what they love and how do we use data science and human expertise to do that? So the types of things that I think about in service of that are things like what are new opportunities that we haven't really discovered yet? And that's been pretty exciting over the last four years.

Stephanie Yee:

I think another area that I think about a lot is like what's the right almost interface between data science and data scientists and the business partners. So this is if we have data scientists working with the design team, or the product team, or the marketing team, or even executives, what's the place where data scientists can contribute the most? And also, just being really intellectually honest, like what's the place where it makes sense for others to take over? And then obviously, the last part of my job is to really create an environment where the team can be motivated and fulfilled in doing things that bring out the best to each of them.

Stephanie Postles:

That's great. So it would be great to dive a bit more into Stitch Fix. I know what it is because I'm a customer, but I think a lot of people may not know exactly what it is or all the things that go on behind the scenes to get the pretty box on your door. So could you explain what it looks like, what is Stitch Fix from a high level, for anyone who doesn't know, and then what goes on behind the scenes to create the company that it is?

Stephanie Yee:

Yeah, so Stitch Fix is a personal styling company. And at the core, we use both data science and real stylists and their expertise to help people find what they love. If you think about unpacking that, it's really about understanding... or from a data science perspective, it's really about understanding a client's needs, as well as being able to set the stylist up for success. The core of Stitch Fix, the way that it shows up is in a box of one or more items and clients are able to try it on, they're able to send back what they don't like and really just keep what they really love.

Stephanie Postles:

Tell me how do you go about making sure that you give the customer the exact outfits they would like or refine that process to where maybe the second or third time you've nailed it? Because for me, at least when I am getting the outfits, I'm like, "The first time, maybe like one thing was off or something," but then after that, it's like, "Okay, now, this stylist knows me, or this algorithm knows me." So how do you refine that behind the scenes?

Stephanie Yee:

Yeah. I think that that's a great question. I think a lot of it... I mean, as a data scientist, like I always think about the data that we collect and what's available, and this comes both from what clients tell us as well as what we're able to infer, so a really interesting example of this, and this is where you had mentioned like, "Okay, there might be one item off at first and the algorithm really learns over time," we really think about things in terms of the ability to say like, "Okay, what data do we have now?" And with the stylist, the stylist is incredibly important throughout the client's life cycle. With the stylist, like what's the right thing to be sending right now? And in response to feedback like, "Oh, that item that didn't really work out for whatever reason," we're able to respond to that.

Stephanie Yee:

I think a really interesting example of the approach that Stitch Fix takes, or rather one of the interesting things about Stitch Fix is that we're thinking about this and we're thinking about a purchase experience in terms of soft goods. So if you think about the way that ecommerce really started off, or at least as I recollect it, it was like comparison shopping sites where you were looking at like how many megapixels do you want in your digital camera. And a camera, those are very easy to compare because it's like, "Oh, it is three or it is four." Whereas with what I think of as soft goods, there's so many different variants on like a V-neck top that it's almost a little bit overwhelming.

Stephanie Yee:

And then on top of that, a lot of the typical searching and filtering is not really going to get people there, just because what might be a great top, even if it's the same aesthetic, what may might be a great top for you might be not as great for me or vice versa, just because it's like, "Oh, you know what? I really need things that are machine washable, or I have very narrow shoulders or something like that." So Stitch Fix is really trying to distill a lot of these things that are ultimately very difficult to categorize into what we would call a latent space, but really to say like, "Okay, we have something like style." Style is not what lunch table did you sit at in high school, it's really a form of self-expression. And because people are so different, we need to be able to use data science to quantify where people are on a spectrum versus what category they're in. To handle this like-

Stephanie Postles:

How do you even encourage people to get maybe the feedback that matters? Because I'm even thinking like if I were to get a shirt and I'd be like, "Well, it doesn't fit," I know you're probably behind the scenes like, "Well, why? What part doesn't fit? What don't you like?" How do you encourage someone to tell you what you need to know to then send them something better?

Stephanie Yee:

Yeah. That's a great question. So what we found, I think that the motivation to give us feedback is actually just an inherent part of the service. I think a lot of people they'll like... When I've styled people and maybe I've missed the mark, people will say, "Oh, you know what? You didn't get it right the first time, but here's more what I was looking for." If you think about it as a relationship, it's not a transaction where you walk into the store and you say like, "I'm happy or I'm sad." It's relationships and relationships are predicated on that back and forth. So it's really a phenomenal percentage of clients that leave feedback on a fix. It's something like 85%.

Stephanie Postles:

Wow, that's great.

Stephanie Yee:

It's just like an intrinsic part of the relationship just because we do frame it as a relationship.

Stephanie Postles:

Yeah. I think having that stylist there really is what forms a human connection, where you're like, "Well, this is..." Of course, there's a bunch of machine learning and algorithms behind the scene, but there's a face here, a human who's actually approving this style and making sure it's perfect for me. And you instantly feel that connection and you don't want to let your stylists down [crosstalk] get that feedback.

Stephanie Yee:

Exactly, exactly. And similarly, the stylist doesn't want to let the client down. So there's that level of trust that gets established. And from there, I think, a lot of the desire to say like, "Hey, this had a fit issue for me because it was too long or something like that." There's just something that's special there inherently. And then on top of that, we obviously do encourage clients to give us feedback, like we'll give them a nudge. But we're certainly not the type of company that has to like... They'll come to us rather than us having to really force the issue, will say.

Stephanie Postles:

Yep. What are some of these subtle nudges that you give that aren't annoying, but then encourage the person to give you the information you need to help them. I think a lot of brands struggle with that, where they either don't follow up at all sometimes if they want feedback, or they do it too much and you're like, "Whoa, chill." How do you guys get that right blend?

Stephanie Yee:

I think that there's two parts to that. One is saying like, "What's the right number of times to be asking or to be reminding really because it's less on asking?" It's just more like, "Hey, if you want, you can leave feedback and there's someone on the other end who's going to be really thinking about it and responding to it." I think it's figuring out like what's the right time to tell people, and it's really like when would this be relevant to someone? I think that there's some other aspects where it's like what's the right time of day to reach out to someone? And all of these can be distilled down into data science problem or data science opportunities. I really find that to be really interesting. I think that there's another aspect, which is that the clients do come back to the app and come back to the site, even without looking to transact. Once they're there, then it's possible to be like, "Oh, by the way, did you want to..." Just making it really easy and lowering friction to giving feedback. That's another way that we're able to implicitly encourage it.

Stephanie Postles:

Yep. So with all this feedback coming in, it's a lot of natural language that you're probably getting, or is there any tech that you're excited about right now to help you categorize it? Are you looking into GPT-3 or anything new this year that could help solve that problem when people are just giving you probably long paragraphs of like, "Here's the things that aren't working for me," and they're just putting in terms that you're like, "Okay, I can actually build any database at this."

Stephanie Yee:

Yeah. I think it's interesting because I think some of the unstructured texts or data generally that might be, I would say, overwhelming to someone like you or I. Computers are actually quite good at processing it. So I think GPT-3 is really incredible, sort of advanced in the way that we're thinking about the opportunities that come from natural language processing. So I think the team is really actively thinking about like what's the right way to bring that into the client experience? We certainly want our stylists to continue to be proactive and like a central part of that relationship. And we're actually trying to figure out like, "Okay, how can we actually bring the stylist forward even more?" But I think the way that I would look at it is I actually love it when there's a large corpus of data, will say, just because there's quite a bit of things that one can infer or pull out of that that would be otherwise a rather arduous task for a person like you or I.

Stephanie Postles:

That's great. Earlier, you were saying that the team is looking at how to maybe utilize GPT-3. And it can be for the overall industry, not just Stitch Fix too. Is there anything where you're like, "I could see this really impacting ecommerce in this way," because that's the one area that I've been trying to look into it? I can see all the things that you can do with it, from not having to code things and writing books and stuff. How could it actually impact ecommerce or data science or behind the scenes?

Stephanie Yee:

Yeah, I think that's a great question. I would say that if you think about... GPT-3 is a really great way to translate information into the format that people are used to absorbing information in, which is text. I think that it's especially important going back to the like you can't take a shirt. The specs of a shirt are not particularly helpful to a shopper. They can be helpful to a computer, but it's like, "Okay, the sleeve is 13 and a half inches, like who cares?" And GPT-3 is able to almost add in a way that would have been incredibly difficult before. It's able to translate some aspects of an item into what that actually means in someone's everyday life. So it's not like, "Hey, we could show you a table of information where it says, 'Here's the sleeve length.'" But it can be more like, "Oh, you know what? This shirt is going to hit your elbow and it's actually going to drape a little bit."

Stephanie Yee:

And because there's so much clothing out there and it's all slightly different in its own way, even if it's once again, the same aesthetic, same color, everything, we're able to bring that to the fore for a massive amount of inventory. So that, I think, gets me really excited. I think another thing that's really promising about something like GPT-3 is it'll let us... yeah, it'll really let us customize an experience to a client using a format that... and move beyond tables of data into information that might be more relevant or easier to absorb.

Stephanie Postles:

Oh, that's great. Yeah, that takes it to a whole new level. I think about right now when I'm shopping around and it shows, "Okay, there's this model and she's 5'9 and 135 or whatever it is," I'm like, "Okay, I could see maybe how something would fit if that's like a similar person to me." But that takes it to a whole new level and saying, "All right, Stephanie, this is going to be bad, you get your elbows and it's going to be very short on your waist," and just putting it in a contextual term where I'm like, "Oh, it's fixed. You all know me. Thanks for letting me know."

Stephanie Yee:

Yeah. And if you think about style and aesthetic, there's almost something... Style is a form of self-expression. And describing style in terms of only numbers is quite limiting. If you look at the way that people will describe clothing, and it's always really interesting to say like, "Okay, fashion week happened. What are they saying about what's being shown?" It almost becomes poetic in that level of abstraction. And I think that that's something that that language is much better at doing, or images even are much better at doing than just numbers and texts.

Stephanie Postles:

So the thing I was just thinking about... I mean, you guys have all these models running and algorithms behind the scenes and you have really large amount of data. How have your models changed? I'm thinking about like pre-COVID models, [crosstalk] probably around work and work clothes, and I want to look nice and heels. And then now, it seems like all those models probably had a big shake up because now it's, "I want athleisure and I want sweat pants and comfy hoodies." How have you guys models changed and what are you doing to adjust them, or what should brands be thinking about with adjusting their historical models that are probably wrong?

Stephanie Yee:

Yeah, I think that's a great question. It was funny actually. In April, one of the data scientists posted in Slack and he was like, "Oh my gosh, like all of the experiments that we're running, we're just going to have to start over." I think that there was a lot of stress behind that statement. And obviously, we're not starting over, but we're starting from a place where the data has changed. And the really wonderful thing about an algorithm and about being able to really take advantage of technology is that they can adapt much, much faster than a person. If we only had the styling team, it can take a little bit of time to figure out like how do we... If we're learning something about like COVID trends, how do you train a team of thousands of people to be on top of everything that is there, in addition to letting them style each client individually?

Stephanie Yee:

So what's really wonderful... and COVID was a fascinating situation because it's like, "Okay, all of the..." There was a tremendous amount of work that had to be done to say like, "Okay, given a pretty big step change in the way that both like the world writ large as well as the way that people are thinking about shopping and shopping online, how do we adapt things to that?" So there was quite a bit of work to do that across the board. And then on top of that, it was easier than it would have been if we hadn't taken a data science approach, just because so much of the models are designed to change. Some of our algorithms they'll be like, "Okay, this is just going to be updated every week just because it needs to be."

Stephanie Yee:

I would say that in terms of COVID specifically like... And a lot of it, we're sitting there and we're saying like, "What are people like?" We'll have conviction in where we think that the market is going. With COVID, it was like, "Okay, you know what? Everyone can anticipate. If people have to stay at home, then they'll have to work from home and maybe they won't feel a need for as formal closes as they normally would." But what was interesting, and this is just from how things unfolded from a data science perspective, we actually had... One of our data scientists was a former epidemiologist. So when we were trying to figure out like, "Oh gosh, the world has changed, like how much merchandise should we buy a year from now," she was actually able to contextualize a lot of the news.

Stephanie Yee:

As a company, we were able to come to what ended up being a pretty reasonable, I would say, assumption about the world and then to go forward and say like, "Okay, overall, how much should we buy?" And then within that, it's like, "Okay, how are consumer tastes going to change?" We can lock down that merchandise. I think the merchant team did a really great job responding to that. Within that, we can make sure that the clients who are looking for working from home clothing versus something else they can actually get it. I think in terms of general trends, I think it's like a 10X increase in requests for working from home clothing. Definitely, a shift out of formal work wear and into more like casual and everyday styles.

Stephanie Yee:

I think athleisure, those purchases have accelerated quite a bit. With Stitch Fix, because we sell actual items, the merchant team had to do a tremendous amount of work to really anticipate that. And then the styling team is able to make sure that those items get to the right people. Because if suddenly we started to only send out leggings, that's not really going to work for many of our clients who just need to make sure that the people who are looking for athleisure can get it.

Stephanie Postles:

Yeah. That's so smart having someone who understands that industry. I feel like there's more room for brands to partner with industry experts like that to help them build their models. Because oftentimes, it seems like everyone is so focused on just, "This is our company model. Only the executives of the company can figure out what the future looks like." But by tapping into someone who has very different experience, [crosstalk] maybe what's happening, it seems very smart.

Stephanie Yee:

Yeah. One of the things that I find to be really fascinating and amazing about Stitch Fix is the way that the executives... like for executive decisions are able to take advantage of the data science capabilities that we've built. And you almost get to this like the core question here, and this is almost... it gets existential, like is how do you handle uncertainty? For me, I'm like, "Okay, this is why I want an executive with like 20 or 30 years of prior experience because some of these questions are genuinely hard."

Stephanie Yee:

I want to arm them. Given the data available, the task of a statistician is to really squeeze out as much information as possible and to say like, "Okay, guys, here's what we can know, here's what we can't know. And the part that we can't know, to the extent that it's incredibly important to have a decision or a point of view on that, that is truly human judgment." So the executive version of that, I find to be really interesting and there's many versions of that throughout the company with the stylist, with the product team, with the marketing team, with the merchandising team, everyone.

Stephanie Postles:

That's great. So when thinking about updating the models and algorithms, would you suggest that a company rebuild from scratch, or should they update a current model to kind of pivot a bit? Because I guess when I think about updating a current model, I worry that there's so many things built into it after the fact and the algorithm just runs away on its own and people are like, "I don't really know what's driving it anymore," versus starting over again.

Stephanie Yee:

Yeah, I think that that's a great question. There's a couple of different aspects to it. Generally, we'll think, "Okay when you have a..." because a model is really expertise in how to use data. So if you find a model that seems to fit the world very well, then you will want to continue to improve it. If fundamentally the world responds quite well to a random forest, or we get very good predictions out of a random forest, then there's no need to change it just because, but there's opportunity to improve on that. Now, with that said, as research is continuing on different methods, people are going to try different methods. But I would say that you definitely want a mix of both because it's both the method and the tuning of that. It's both the type of model that people will think about as well as the tuning of that model or adding new variables to the model or something like that that we want to do.

Stephanie Yee:

So to give you a concrete example, like with COVID, we have a demand forecast. The demand forecast is really modeling client behavior and it's really being able to give the merchandise team and the executives and the operating partners visibility into like, "Okay, what's life going to be like a year from now and how should we plan?" When COVID happened, everyone's like, "Oh my gosh, the world is very different." But what was great was we were able to say, "Okay, here's some assumptions that we have. We can update those assumptions, but we've got several years of work into the capability itself. And the great news is that we don't need to start from scratch because things have been built in a way that can adapt."

Stephanie Postles:

Yeah, that's very smart. When thinking through your demand forecast, are you guys forecasting that the world will eventually return back to pre-COVID, or do you think it's a new normal and now people are going to continue working at home indefinitely and keeping it adjusted? How are you guys forecasting the future of apparel?

Stephanie Yee:

Yeah, I think that's a great question. I would say that there are certainly things that are very large shifts and there are other things that are just probably going to stay the same. I would say that it's a blend of the two. I certainly don't think and I certainly hope that we're not going to be working from home forever.

Stephanie Postles:

Yeah, I hope not.

Stephanie Yee:

Exactly, exactly. With the vaccine coming out and just how effective the vaccine seems to be, I think that we will be returning to... There's some things that are going to fall back into place. There's some things that frankly have already fallen back into place, and then there are other things that the company is really leaning in to take advantage of, so definitely a mix.

Stephanie Postles:

Yeah, yep, I agree. Have you seen any different types of consumer buying behaviors around what consumers are expecting now that more people are at home, they have more time to try things on? Have you had to adjust how you interact or work with your consumers during this time that was maybe different than COVID or pre-COVID?

Stephanie Yee:

Yeah. I think as I mentioned, there's definitely a difference in what it is that people are looking to buy. I think another thing that has been really exciting is that I know quite a few new shoppers, people who have never bought anything online before suddenly they're like, "Oh, shoot, all the stores are closed. I now have to try this new channel." So we're seeing people who aren't even used to a traditional way of shopping buying things. I think that's been really interesting because that behavior, as you can imagine, can be quite different. So it's great that the business is able to respond to that and-

Stephanie Postles:

Yep. It seems like there's a whole new demographic market that is opening up now that a lot of ecommerce companies are going to be able to have a lot of opportunities with. I'm thinking about Stitch Fix, my mother-in-law, who's almost 70, when they came back and told me she... I had never told her about you guys and I don't think she would actually ever do that. And she's like, "I ordered from this company, they picked things out for me, it fits perfectly." And I'm like, "Are you talking about Stitch Fix?" And I was genuinely surprised, but she found out about it on her own, went forward, bought it, worked with a stylist, and got her box. It just made me think about how many opportunities are opening up with this new group of people who never were probably comfortable with buying online before. But now, they're forced to it and it's now becoming normal for them.

Stephanie Yee:

Yeah. I love that story. That's wonderful. I think what's interesting too is that folks like your mom or my mom where they're not actually as used to buying online, they're more used to going into a store, so they're actually more used to being able to talk to someone. Whereas like my friends are like, "I don't want to talk to a human being."

Stephanie Postles:

Yeah, don't call me. Don't look at me.

Stephanie Yee:

Just text me. Right. Don't leave a voicemail, that doesn't work. Right. But the folks who are trying something online, they're used to a store. And Stitch Fix like the gap between some of these department stores where you do have a person and the department stores online presence is quite uncomfortable. So if you have Stitch Fix, obviously you're not in the store, but you get to try things. You get to work with a person, you get someone who's actually there to help you. I think in some sense, it's actually a more natural entry point, especially if folks aren't used to the current paradigm of shopping.

Stephanie Postles:

Yeah. How would you advise a company to be able to not only continue to focus on their traditional consumers that they're used to, but also lean into that new group of people because it seems like you would have to have very different messaging? Like you were just mentioning, some people like myself and you are like, "Just text me, do not leave me a voicemail. Don't try and call me, I'll decline it," whereas this group, you have to have a whole different mentality. Your customer service team probably needs to start calling people and doing things very different. How would you advise a company thinking about this, who wants to maybe connect with both of them, their current customers and the new ones who are now coming on the market?

Stephanie Yee:

Yeah, I think that's a great question. So the way that I would think about this is, first off, you have to come up... Let's take the messaging example. You want to think about what are the different messages that are going to be resonating with consumers? And then the second is how do you get the right message to that consumer? In terms of what will resonate, I firmly believe... There's a very interesting opportunity for interaction between design and data science and user research and things like that. Data science can contribute, but ultimately the messaging strategy is one that is the overall messaging strategy. You can try many different variants, but the overall strategy is one that is a judgment call.

Stephanie Yee:

And then machine learning is wonderful for being able to say like, "Given this message, or given this client and given a universe of messages, how do I make it so that the client can see the most compelling one and really understand on their terms what it is that we can offer?" I would say this is an area where you definitely need both art and science because messaging is so incredibly important and strategic. So it's working with the marketing team, it's working with the design team, and then the data scientists can really help figure out where should that message be delivered? How should it be delivered? What is the right way to make it land with the client?

Stephanie Postles:

Oh, that's great. So we're putting together this end of year commerce article about 2021 trends. And this is one thing that we're talking about is how much the over 55 demographics spend. And they spend twice as much as millennials. And I think I saw, let's see, 10,000 baby boomers are going to turn 65 every day until 2030.

Stephanie Yee:

Oh, wow, okay.

Stephanie Postles:

And then by 2050, the over 60s will account for 20% of all people globally. So when I started seeing these stats, I'm like, "Whoa, more people need to focus on this demographic." Oh, and then another one, the 50 and older crowd has a lot of spending power. And if you put it in terms of GDP, it would be the third largest in the world.

Stephanie Yee:

Oh, wow, okay.

Stephanie Postles:

US is 21 trillion, China's 14 trillion, and then Japan is 5 trillion. And this is where the people they spend 7.6 trillion in 2018. To me, I'm just seeing all these opportunities that are being missed right now. I'm like, "What the people be doing?"

Stephanie Yee:

Yeah, no, I think this is a wonderful group of folks. Within the tech industry, I would just say especially there does tend to be a focus more on millennials and things like that. I think the great thing about Stitch Fix is that we are... And oftentimes I think some brands they'll sit there and they'll say like, "Oh, our target demographic she is between 25 and 39. And after that, she's not us." I think with Stitch Fix, we're able to say, "You know what? We're not going to categorize you into one group or another, we're going to serve you where you are. And with personalization, we are able to..." I completely agree with the stats or the information that I have on how that generation of clients interact with Stitch Fix is very, very consistent with some of the numbers that you had described. So it's a really wonderful group of people who are thinking about their personal style, and I do agree it's folks who, I think, tend to be served a little bit differently, really at the retail industry's loss.

Stephanie Postles:

Yep. Yeah, I agree. How would you go about getting the right data to then be able to craft the personal message then? For Stitch Fix, it does feel a little bit easier because you can ask things like age and a bunch of other questions and they're like, "Well, they're styling me." But for a lot of other brands, if you were to ask age, they'd be like, "What?" How would you advise other companies to be able to get enough information to then be able to personalize a message like that?

Stephanie Yee:

Yeah. I think that there's a couple of different ways to do it. And a lot of it really is around the marketing and design toolkit. Because ultimately, when you're coming up with messaging, you don't want to say like, "Okay, this is the messaging for folks who are 50." I'm an old soul, so maybe I'll just really respond to that myself. A lot of this is just a strategic question. So data science can play some role where it's like, "Okay, based on what we know, people tend to respond to X, Y, and Z." But really, if you want to be looking forward, it's less like what've people responded to in the past?" You definitely want to take that into account, but it's more like where are things going in the future, especially at a time when things are changing so rapidly?

Stephanie Postles:

Yep, yeah. That's why I'm also excited about being able to ingest the sentences that people are asking the customer service reps or putting in the search bar because I think that alone could tell you who someone is just based on how they say [crosstalk 00:32:09].

Stephanie Yee:

Oh, absolutely. Absolutely. Yeah, I agree. I think that the notion of being able to have more conversations with people is something that I think is incredibly exciting and it does allow for a level of, I would say, flexibility of expression, especially once computers can really respond to that.

Stephanie Postles:

Yep. So when thinking about building up a data science team, what are your first steps? How would you tell a brand to think about it to be able to build it up in an efficient manner, where it's answering the right questions, you have the right goals in mind? Because when I think about data science from different companies I've worked at, some people are called data scientists when they're really a BI team [crosstalk] called data scientists. And then you have marketers who are also data scientists. So like how do you [crosstalk 00:33:26]?

Stephanie Yee:

It's certainly become a loaded term. It's funny because in recruiting, it can be incredibly frustrating like, "Well, this LinkedIn search is not very helpful." Yeah, that's the first thing that I would say is if people are thinking about like I need to build out a data science team, searching on the term data scientist is probably not going to be the most efficient way to get there. I think probably the step one that I would advise people to do is to really think about what role do you want data science to play and where are the areas that you see as high value? And this can be a little bit of a hard question because without... in the same way that I'm not a 100% familiar with a merchant's toolkit or a designer's toolkit.

Stephanie Yee:

If I, as a data scientist, look at a problem, I can be like, "Oh, this is something that can be very easily solved with the machine learning." It's hard for folks who don't have that background to know that, but really thinking about like what is the strategic problem that people are trying to solve? With data science, I'm very supportive of making it like a core... like