Big Idea: Data & Analytics Interview
September 24, 2013
By: Tyler Spalding (StyleSeek), interviewed by Renee Boucher Ferguson
A new startup uses data and highly tuned algorithms to make both business and fashion decisions — upending conventional wisdom about the need for intuition, creativity and an eye for style.
StyleSeek co-founder and CEO, Tyler Spalding.
StyleSeek, a new recommendation website, uses data and constantly refined algorithms to help both fashionistas and the fashion-challenged alike to discover and choose exactly the right thing to wear, at exactly the right time (think work, weekend, night out).
But using data to identify fashion choices isn’t to everyone’s taste. Take GQ’s response to the idea of applying analytics to an essentially creative process — one filled with intuition, experience and, of course, an eye for style:
It has to be said that style is and always has been about individuality. This is the main problem that faces the developers of the algorithm — one they are trying to sidestep with … StyleGame, a nine-step profiling quiz that attempts to work out whether you like semi-cutaway collars by flashing up pictures and asking you whether you prefer Daft Punk or Outkast, beach houses or generously appointed New York lofts… Style isn’t really that cut and dry.
With about 50,000 actively registered users — and close to 200 retailers on board, including the likes of Nordstrom, Macy’s and Anthropologie — StyleSeek co-founder and CEO Tyler Spalding (@trspalding) doesn’t quite agree with GQ’s perspective. That’s because conquering the fashion industry is just StyleSeek’s first step in its real mission — to transform e-commerce. The company, which started as a project while Spalding was a student at MIT’s Sloan School of Management, employs 10 people, half of whom are technologists and the other half content curators — or, in Spalding’s words, people who actually understand style “better than I ever will.”
Spalding spoke with MIT Sloan Management Review contributing editor Renee Boucher Ferguson about his company’s data-driven mission and its early successes.
By: Tyler Spalding (StyleSeek), interviewed by Renee Boucher Ferguson
A new startup uses data and highly tuned algorithms to make both business and fashion decisions — upending conventional wisdom about the need for intuition, creativity and an eye for style.
StyleSeek co-founder and CEO, Tyler Spalding.
StyleSeek, a new recommendation website, uses data and constantly refined algorithms to help both fashionistas and the fashion-challenged alike to discover and choose exactly the right thing to wear, at exactly the right time (think work, weekend, night out).
But using data to identify fashion choices isn’t to everyone’s taste. Take GQ’s response to the idea of applying analytics to an essentially creative process — one filled with intuition, experience and, of course, an eye for style:
It has to be said that style is and always has been about individuality. This is the main problem that faces the developers of the algorithm — one they are trying to sidestep with … StyleGame, a nine-step profiling quiz that attempts to work out whether you like semi-cutaway collars by flashing up pictures and asking you whether you prefer Daft Punk or Outkast, beach houses or generously appointed New York lofts… Style isn’t really that cut and dry.
With about 50,000 actively registered users — and close to 200 retailers on board, including the likes of Nordstrom, Macy’s and Anthropologie — StyleSeek co-founder and CEO Tyler Spalding (@trspalding) doesn’t quite agree with GQ’s perspective. That’s because conquering the fashion industry is just StyleSeek’s first step in its real mission — to transform e-commerce. The company, which started as a project while Spalding was a student at MIT’s Sloan School of Management, employs 10 people, half of whom are technologists and the other half content curators — or, in Spalding’s words, people who actually understand style “better than I ever will.”
Spalding spoke with MIT Sloan Management Review contributing editor Renee Boucher Ferguson about his company’s data-driven mission and its early successes.
What does StyleSeek do?
We’re a new kind of e-commerce personalization platform. Ninety-nine percent of existing e-commerce recommendations are done very inefficiently on a variety of levels. They generally take really, really big data sets and extract patterns and match keywords. What we’ve created is analogous to a Pandora-like system. Each product is mapped by human style experts across a variety of universal characteristics. So we can literally compare a T-shirt and a couch and actually understand the differences in style and what each person likes.
Right now we have a consumer app that focuses on clothing, but from the very beginning, it has been designed for all types of e-commerce, from magazines to furniture. What makes our platform really unique is that it’s not based on collaborative filtering, like many other e-commerce recommendations software.
Everything that we’ve put together, all of the qualitative aspects, is purely derived from analytics. Every piece of our algorithms is constantly driven and refined by everything that we’re measuring on the site about what people are doing. I’m an engineer by trade, as are other people behind all this, so every single thing that we’ve done is purely rooted in analytics and what the data is telling us to do, from the very beginning.
Why fashion? There doesn’t seem to be a direct correlation to, well, engineering.
Oh, great question. So, our goal is to do this [interest-based recommendations] for all lifestyle products, but we started in fashion primarily for two reasons. One is that everything is data-based, so we needed that really great data of what people liked, what they didn’t like, what they bought, what they saved, and how they used that information to create new profiles for themselves. And we saw that fashion by far had the most visceral reaction — that when we try to personalize something for you and say, “Here are the results we think that you’re going to like,” people immediately react to fashion.
They’re like, “oh, well, I own those clothes,” or “I would never wear that,” or “I would wear that.” And it was just a really easy one-to-one connection. When we did other things, like furniture, it was less strong. Clothing gave us this really, really great data.
And the second piece is, from a data perspective and also from just a software architecture and branding perspective, if we started right out of the gate doing everything, we think we just would have confused people. It would have been really challenging to say, “oh, we’ve got this start-up, but we’re going to compete with Amazon.” It just really didn’t make much sense. So, we really wanted to zero in on one vertical and start and really massage the technology and build out our platform right. We felt that fashion by far was the best opportunity for that.
Oh, great question. So, our goal is to do this [interest-based recommendations] for all lifestyle products, but we started in fashion primarily for two reasons. One is that everything is data-based, so we needed that really great data of what people liked, what they didn’t like, what they bought, what they saved, and how they used that information to create new profiles for themselves. And we saw that fashion by far had the most visceral reaction — that when we try to personalize something for you and say, “Here are the results we think that you’re going to like,” people immediately react to fashion.
They’re like, “oh, well, I own those clothes,” or “I would never wear that,” or “I would wear that.” And it was just a really easy one-to-one connection. When we did other things, like furniture, it was less strong. Clothing gave us this really, really great data.
And the second piece is, from a data perspective and also from just a software architecture and branding perspective, if we started right out of the gate doing everything, we think we just would have confused people. It would have been really challenging to say, “oh, we’ve got this start-up, but we’re going to compete with Amazon.” It just really didn’t make much sense. So, we really wanted to zero in on one vertical and start and really massage the technology and build out our platform right. We felt that fashion by far was the best opportunity for that.
How long have you been in business?
A total of around a year. We just came out of a private beta in April of this year [2013]. We’ve actually only been completely live, where people can now sign up without having to go through a more rigorous process, for only a few months. So, that’s where a lot of the growth has, obviously, come from. And we also just recently launched the site for women as well. So, we were proving the concept out with just menswear.
It seems somewhat counterintuitive to me, just based on what I know about fashion, to open StyleSeek up to men first.
Oh, yeah. Smaller market from a business perspective for sure, but we were aware of that. Men hate shopping online. They just hate it. We all know many men aren’t as fashion inclined and often don’t want to provide information. They don’t want to browse online. They’re just not really interested, which actually worked great for us because we viewed them as our hardest customer to ever please.
So, we tried to build our entire site around the concept of, “what’s the bare minimum information we can get from someone and the easiest way in order to provide value add to people.” And it has actually worked really well in that, for the game that people play on our site to determine their Style DNA, we have greater than a 95% completion rate. So, we found this to just be immensely sticky. Even guys that don’t like shopping are filling this out and giving us really, really valuable information. And so that was a home run for us.
And then the second part of it was, because we were focused on menswear, that’s one of the biggest pain points for retailers. They just want as many men as they can get. So, a Nordstrom.com or a Macys.com [is] really struggling bringing men to their sites. And so when we originally came up and said, “hey, we’ve got this menswear site,” they were all ears. They actually contacted us, and they emailed us and said, how can we help be a part of this? What relationships can we grow? How are you doing this?
It ended up being a really good acceleration for us to really validate ourselves in the market and work with retailers and get data from them. It was the shiniest new tool that was out there. And it really scratched an itch that they had.
Did you know that going in, or was that an ah-ha moment?
No, we didn’t know it going in. We discovered it during our research at Sloan, in figuring out what we wanted to build and why. In talking to customers, they were saying one of the biggest problems in personalization was menswear: “If you guys are even doing something there, we’d be really, really excited.” So, we sort of had a little inclination, but that was definitely the market research and talking to the actual customers that validated that for us.
Why did you decide to build a company that is based on analytics and where, as you mentioned, analytics tell you exactly what to do?
A lot of it is probably just more a personal preference. So, I’m an engineer by trade and have always been math and science guy, one of the biggest nerds you’ll probably meet. I was literally a rocket scientist for the NASA space shuttle program, so I’ve just got this in my blood to where it’s really what I love to do. I’ve probably been writing algorithms that predict behavior, whether it was something for video games when I was 16 all the way up now to e-commerce in my 30s and everywhere in between. It’s just really been my personal fascination. It’s where my expertise is.
So, from the beginning we really had that as a part of our DNA as people: the data behind what’s happening will tell us what to do. And the data basically validates everything. So we’re really, really strict on making sure we measure everything the right way and that we have the right metrics in place and everything else to really understand what’s happening.
One quick example of that is, when we first launched the site, we actually had editorial from across the entire Internet. So, we would have articles from GQ, for instance, all of which stayed within our site. And we had images there, and we’d be able to curate style from around the Internet for you. And we delivered that in its own stream, as well. And people unanimously told us that was the greatest thing. And they did interact with it. They’d read the editorial, and they liked it. They liked the recommendations. We even matched it to clothing items. We thought it was the greatest piece of technology we had built. But sure enough, we saw that people were gearing toward the products and not the content. And then we started asking more people anonymously, “What drove you here?” they said, “I just want to buy things.”
So, despite people telling us in the very beginning that they wanted all this editorial and that it was going to provide an inspirational experience for them, it really boiled down to, people just wanted a tool to allow them to buy things more quickly and easily. And so we ultimately just removed all the editorial; it ultimately was a distraction to people, which made them less confident and less excited by the site in general. The engagement that we’ve had has been even greater since. That was a surprising revelation for us.
How much will you continue to rely on data alone, as your business grows? And what if data proves you wrong?
I guess having it prove us wrong is a risk we’re willing to take. I’d say at this current stage we’re probably still at a 100% reliance on data mainly because we still have this consumer-facing application. So, in building the technology, we obviously use our own experience and coding of the back end. The actual code, obviously, comes from our experience. But what we show to consumers, and why, is almost entirely driven by data and likely always will be, just because it’s been our experience that it’s impossible to prove what a mass of consumers will actually want or even predict what they’ll want.
So, I envision us having literally hundreds of tests potentially going every single day, whether it’s copy messaging, whether it’s call-to-action messages placement, button placement, where menus are located and things like that, and we’ve built the system to AB test and expose new features to specific users.
We’re just constantly watching what people are doing.
I guess having it prove us wrong is a risk we’re willing to take. I’d say at this current stage we’re probably still at a 100% reliance on data mainly because we still have this consumer-facing application. So, in building the technology, we obviously use our own experience and coding of the back end. The actual code, obviously, comes from our experience. But what we show to consumers, and why, is almost entirely driven by data and likely always will be, just because it’s been our experience that it’s impossible to prove what a mass of consumers will actually want or even predict what they’ll want.
So, I envision us having literally hundreds of tests potentially going every single day, whether it’s copy messaging, whether it’s call-to-action messages placement, button placement, where menus are located and things like that, and we’ve built the system to AB test and expose new features to specific users.
We’re just constantly watching what people are doing.
Not to belabor the point, but when you’ve got in-house expertise — I know you’ve got a couple of people on staff from the Fashion Institute of Design — and their recommendation differs from what the data tells you, is there a point where intuition or experience overshadows data?
Yeah. What would happen is, if something like that occurred, we’d probably just run a test. Anything where someone would say, “All my experience says that we do X,” we would surely entertain doing X, but then we’d probably also build the Y, the complement to it, and we’d literally run the test and just prove it. And if you do that intelligently, it’s very, very easy to prove very quickly that, “oh, if I do this and I put this banner here, this call to action, the button at the top will make all the difference in the world.”
We have built this system where we can release something to the next 20,000 people that come to the site. Let’s see if the conversion rate changes. We can do all that. And we can see where all the traffic comes from, too. We basically have data at every single touch point to where we have a real robust analysis to say, “okay, well, how did that compare to the last month or the month before that?” And it’s really easy to run the correlation against that.
You’ve mentioned that retailers were very happy to hear that you were attacking the menswear market first. But was that, or is that, your value proposition when you approach partners? I mean, is it menswear or analytics that they are happy about?
In the current format, it’s all about traffic, in that we’re able to drive potentially tens of thousands of new visitors every month. This is brand new traffic that they might not have otherwise had. That’s the easiest way to get someone talking to you — is to say, “we’re going to help send traffic to you, and people are going to buy things.” That is exciting.
And then the thing that they’re seeing, even if the analytics hasn’t really been a factor, they’re also seeing that the traffic we deliver is, as of right now, up to three to five times more valuable than other traffic. Because we’re literally getting three to five times the conversion rates. The industry average of any click-through to an e-commerce site is between 0.5% to 1.5%, or something in that range. It’s relatively low. But we’re often able to deliver anywhere from 5% to 10%. So retailers see that, and they’re saying, “Something good is happening here. We’re getting 100 clicks from StyleSeek (or whatever the number is) and we get five sales, when we were only usually able to have gotten zero all along.” So, this is more highly qualified.
With historically low conversion rates in e-commerce, that’s a really big jump. How is it that you’re able to get there?
What we’d like to say, as now we’ve proven, is that it’s all about relevance. We take the steps required to show you things that are relevant to you. And we have personalized our site for you. You’re only seeing things that you should have an interest in, and then we have the tools to allow you to do the “complete the look” function — see other things, browse other things based on your style, or the retailer’s price points. So by the time you’re actually clicking that button out to a retailer’s site, it’s just that much more likely that you’re willing to buy. That’s basically what we’ve proven.
That was our goal, or question, from the very beginning — if you personalize a website, will it ultimately lead to higher conversion? And now we’ve proven, time and time again, that the answer is absolutely yes. And that’s actually the impetus behind the company itself in that the Internet is all becoming personalized now: Facebook, Twitter, Pinterest, all these sites.
If I go to Nordstrom.com right now in Chicago, and you go to Nordstrom.com out in Cambridge, we get exactly the same results, which is just mind-blowing to me — how that can really be possible, with all the technology that’s out there and what they’re selling, [that] you see literally exactly the same page that I do? And that’s what we really wanted to change.
One last question: As far as other verticals that you will pursue, have you decided which ones? Or will that decision be made after testing?
After testing for sure. We have inclinations [toward] things that are popular on Pinterest and Fancy, and we have a bias towards thinking they’ll be interesting, but really it’s all going to be data-driven. Every single choice will be data-driven.
No comments:
Post a Comment