The panel’s topic was provocative: will computer vision exponentially increase clothing commerce? It required the two panelists, including our CEO and founder George Borshukov, to peer into the hard-to-predict future of how technology will reshape the way people shop for apparel and footwear.
Held earlier this year in New York City, the panel was part of a two-day summit organized by LDV Capital to explore how digital imaging and video technologies will empower or disrupt businesses and society.
Joining Borshukov on the panel was Alper Aydemir, co-founder of Volumental, the Swedish company whose technology powers 3D body scanning in stores. The editor of TechCrunch, Jonathan Shieber, moderated.
Here’s an edited transcript of their lively and insightful discussion:
Shieber: When you think about the visual learning space or...visualization, it's a technology that's been around for awhile. So as a new startup, how do you differentiate yourselves? How do you stand out from the crowd when there's been so much research and deep technology that's been developed around this stuff.
Aydemir: I think one way to answer that is that, sure there has been a lot of technological advancements and stuff, but that doesn't really translate into successful products. Today's hot new tech is tomorrow's GitHub repo story. That's a given and that's especially true in computer vision and software. Whatever that is today, clothes—we can't talk about it—but we had a bajillion amount of investment and so on. Once it's released, someone will reverse engineer it, put it on GitHub, and then that's it. The real value, I think, in computer vision is counter intuitive. It's like all good problems in life are…and all good things are counter intuitive, but the good way of building a computer vision company is not to call yourself a computer vision company.
Shieber: It's about the application, then. It's not about the technology itself. You can have the whizziest, bangiest, newest technology, but if you're not solving a real use case, then what the hell's the point?
Borshukov: You heard nobody believes computer vision works, right? The moment it works they try something else.
Aydemir: That's true. That's true.
Shieber: How has that impacted this tension between the technology and the application? How has it impacted the way that you've thought about your own business, George, and the way you're transitioning?
Borshukov: Absolutely. I have great stories. My background is in visual effects. Actually, Steve [Sullivan] and I go a long way back. What he showed today that's done at Microsoft was kind of a dream back in 1996 when we worked on the first Matrix movie. There were these scenes called Bullet Time that we were challenged to make, and we had no idea how we were going to do them. We looked at different technology, and John Gaeta, the visual effects supervisor, challenged us to capture the scenes exactly like Steve showed us. We were supposed to film with multiple film cameras, capture Keanu Reeves dodging bullets, whatnot, and then be able to replay those scenes in the movie, in the slow motion that someone showed earlier. We looked at this for months and the technology just wasn't there, so we ended up on the first movie...with still cameras. Then for the backgrounds we used photographs. But long story short, it all goes full circle. I like change. I am in technology. I have applied it in movies and video games, and then six years ago I decided to apply it to fashion e-commerce. I got involved with a start-up [Embodee]. I was a CTO for six years. We wanted to do virtual fit technology.
Shieber: Let's telescope down into that a little bit. We've gone from the general, what's going on in visualization, broadly how you need to think about it in terms of an application rather than a technology, to specifically the ways in which this can be applied for retail and fashion for consumers. If you project out the next five years, how does your visualization technology get integrated into the retail experience? What does that store look like?
Borshukov: For me two things are big in e-commerce right now. There's the trend of mass customization, which is still a niche, but it's increasing and some people believe it's huge. The problem there is, before you buy the product, it has actually not been made. So you need good visualization. The second problem in e-commerce for clothing is that most of the things you buy are not going to fit you when you get home. Therefore you'll be disappointed. There'll be high returns. Both of these things require really good visualization, really good computer graphics. They require computer vision so we can capture the clothing because CAD doesn't exist in the apparel industry.
Shieber: Right. Alper, how does that relate to what you all are doing over at Volumental?
Aydemir: Right. I think when we think about the retail experience, it's really related to these sensors that are not yet quite in your pockets. So that's one place to reach these consumers and you would rather reach them, I think, at home if you can. Of course I think the retail stores will never go away. You will always have a need for those. But then again, even at the store environment, you need the right kind of experience. Most people, for example, are not comfortable with stripping down and going through something that looks like an airport scanner, and so on and so forth. I already feel awful about trying on three pairs of pants. I don't want to go through this extra step again. I think a lot of that depends on nailing that user experience where the technology is invisible. It's not even part of the conversation. You're there to buy a product and this happens to be the best way of doing it. You don't care if it's computer vision or X-ray.
Borshukov: Also it's fun.
Shieber: ...when you think about the entire supply chain, or how this gets integrated, you all have talked a lot on your blogs, respectively, about the notion of fashion and retail moving from mass production to mass customization. When you think about the enabling technologies that are going to help move that along, obviously the visualization piece and the scanning piece is one part of it. How do you see that integrating with a broader supply chain to create a new kind of retail market?
Borshukov: Right. All the apparel companies will love that model. It will be much more efficient. It will be much more efficient for the environment. It will be much more efficient for all of us, but it's a very slow process. Right now the way the apparel industry works is this 30-30-30 rule—30 percent of the clothing is sold at full price, 30 percent is sold at the deep discount, and 30 percent ends up in the landfill. That's the business model. If you bring in mass customization, all of a sudden you're talking about having stations that only produce clothing on demand. Therefore you have no waste, essentially, especially if you have good visualization. So it could be a revolution.
Shieber: You talk about a revolution and yet the stores that have been able to implement this most effectively that I've seen—and we can talk a little bit more about who you all think have been really good at this—have been bespoke stores that are doing tailoring anyway but are now doing really high-end [stuff]. Frankly, I'm a reporter; I'm too poor to afford that. I'd like to be able to afford it at some point, so what do I do? How does that happen for me? When does it come to my level where I can spend 20 bucks and get a nice shirt?
Aydemir: I think we are still quite some time away from that when it comes to getting something customized, or entirely customized. Part of our team spends a lot of their time in manufacturing ateliers and factories and so on, trying to understand the craft of it. A lot of it is historical, and you're trying to digitize something, an industry, that is maybe a thousand years old—like shoe making, let's say. The thing is that we can't just come up with computer vision algorithms and hope that it will work immediately out of the box and they will be able to integrate that into their existing system. They won't be able to.
Shieber But if you look at something like Soles, or you look at something like, what's that earbud company? Normal, here in New York. They're both here in New York, and both amazing.
Aydemir: But then again it's coming back to the thing that you said: it's quite expensive. It's bespoke, it's not like a 50-buck insole that you can buy. You need to actually go somewhere, scan yourself.
Borshukov: There's a special machine on the side. That's the one thing we know. There's the mass production area in the factory, and there's a couple of actual machines that are set on the side for the custom stuff. Those require more work, but they're more profitable, so how do you get the volume up on that stuff?
Shieber: At some point do you all envision a world where there are 3D printers, in-home 3D printers that will be able to take these scans, someone will customize the clothes, and they'll just print it out? Is that a future that you can envision, or is that a pipe dream? Is that science fiction?
Aydemir: It's not a pipe dream, but it's quite Jetsons I would say.
Shieber: Are people too lazy for that?
Aydemir: It just doesn't exist. You can't really print a shoe at home. But then you can still deliver a ton of value. You don't have to go all the way Jetsons.
Shieber: Who are the companies that you all are working with or that you all are looking to who are doing this successfully and doing it well? I mentioned Soles, I mentioned Normal, mainly because those are the two companies I can think of right now.
Borshukov: I see it from a different angle. You're talking about startups doing the ninja stuff, really cutting-edge stuff that's very small still. Maybe it'll make it big. My experience comes more from the fact that we have approached all the biggest brands out there. We've approached them first with the virtual fit idea, and trying to digitize their existing lines of clothing, and offer a virtual fitting room online. That idea is very hard for them to understand and fund and take. So we ended up pivoting to the customization piece. It's especially hard for the more fashion-oriented brands. Where we've found our best business ends up being with companies...on the sports apparel side. They're very big companies, very technologically savvy. It turns out that they really do have a technological edge to them, so we've found them to be the best adopters of technology.
Shieber: That was your previous business model, right? That's not the business that you're in currently.
Borshukov: Right, so I think on the next one you go full on, you go direct-to-consumer. You don't go B2B with the brand. You just create a destination where, if you want to buy clothes online, you go to there, if you want to make sure they fit. Then maybe you get redirected to Amazon or wherever to make the purchase. But in order to make sure that it's going to fit, you go through that intermediary. Integration into each of the brands on an e-commerce operation is not going to work.
Shieber: Direct to consumer is, in your opinion, is the future?
Borshukov: It's the only way for the fit side of things.
Shieber: How do you feel about that?
Aydemir: Okay, for the fit side of things? Do you also mean the making the product side of things?
Borshukov: No, no, no. I mean only for the fit, the product then is made-to-wear, basically.
Aydemir: Right, exactly. I think similarly, and another strategy that we can capture is really the long tail and that's all these bespoke or smaller companies. There are a lot of them that you can have really quick rollouts, get some really quick early-adopter user base, and also revenues from it, and learn from that to catapult yourself to the really, really mass customized future while all these big giant companies are busy moving their various parts.
Aydemir: That's one thing we are doing. We are working with bespoke manufacturers, for example in Italy, in Germany, in Denmark, all these places. Also, as of this week, in headwear, also in eyewear. 3D printing is much more doable when it comes to eyewear, for example, than shoes.
Borshukov: Because it’s hard surfaces.
Aydemir: Right. When it comes to making the actual product, that's always something that is hard and you will be stuck with one kind of product, whereas the fit solution offers a lot of different opportunities.
Shieber: Let's talk some hard numbers, now. How much does it cost for a company to implement y’all’s solutions, and what's their ROI on stuff like that? How many have you done? You say you're working with a number of small companies. What's the time, what’s the payback for these guys?
Aydemir: I can give one example that is public.
Shieber: Give us some private ones. Just a few.
Aydemir: I would but I think I would not be able to afford to travel here next year, or be allowed to.
Aydemir: One example is a company that has about 20 stores in Europe and they have their own shoe masters. These are people who are the shoe nerds of this world. There are a few of them. What happens today is actually quite interesting. This person goes to 20 stores, one by one, and measures people’s feet. They spend their life on a plane, basically. Obviously they can't scale. They are doing pretty well, but they can't scale. And so what if you could replace this person with a scanner and algorithms and stuff. We just did a man versus machine competition for that. We had the same 150 people, 150 shoes.
Shieber: Did John Henry win?
Shieber: John Henry. The guy with the hammer and the railroad and the locomotive. Did the guy beat the machine?
Aydemir: No the machine beats them.
Shieber: Damn it!
Aydemir: They are quite comparable in performance and so on and when it comes to fit, but obviously it's a lot more scalable, it's faster, and so on and so forth. That's one thing that we are doing. To install that into the store is in the low or mid hundreds of dollars.
Shieber: Per device?
*Aydemir: * Per store.
Shieber: Okay, per store.
Aydemir: Per device.
Shieber: All right, one device per store?
Aydemir Yeah. The capture time is about a second. So that's the thing. People shouldn't be even aware that they are being scanned by this Star Trek-like device. You just walk in and say, "Yeah, I want to buy some shoes," and then the salesperson there assists you. "What kind of shoes are you interested in, please sit there or whatever," and then in that instant you are done. The tech part is done.
Aydemir: I think that's what we want to build to make this more like a utility. It's not visible, it's just like, you turn on electricity and it starts.
Shieber: Does this become a software program? Do you see this as something that people have the option to download for shopping at home? Is this just in store? Is this a physical retail product?
Aydemir: Obviously once that is done you can take that data and use it to see if e-commerce shoes on Amazon will fit you well or not. So you can take that data anywhere with you. Our system, we've built the system to allow for that.
Shieber: George, how does that jibe with what your thesis is around Intervisual, now that you've launched out on this new venture and you've got this new thing going on?
Borshukov: Intervisual is to really restart the virtual fit idea and customization remains with Embodee. Intervisual is more about restarting the original idea of the virtual fit, in a better environment. When we started Embodee it was 2009, it was a very bad year [for the economy]. Now a lot has changed. Now we have 3D cameras coming around the corner that will be on every tablet, on every mobile phone. That means we're going to be able to capture our bodies…Basically I think that it's a moment where it's probably still worth trying the idea. We're seeing what's happening with augmented reality and virtual reality. Something like the HoloLens could be an amazing device to implement the virtual mirror. The virtual mirror will require some of the same ideas. To digitize the clothing, you're going to need real-time cloth simulation. You're going to need to superimpose it over a real image feed of the user. I think it's an exciting time. There's so much buzz around these new technologies and this could be one of the killer apps.
Shieber: I've seen a company actually. I was at South by Southwest—and I apologize to that company if this is a live feed and they're watching me, I forgot their name—but they were doing this virtual mirror. I guess my question is, what's the friction around that? You run into that same problem at the store where you're dealing with taking off your clothes, getting scanned, and it seems like there's a lot of friction there. Is there a way to reduce that process? Do you have the virtual mirror in the fitting room and you say, "Just take your kit off once and then we'll try on all these clothes?”
Borshukov: We'll have to figure it out. I think that's what's exciting about it, but I think now there's all these new devices and new technologies coming on the market that, six years ago, there was nothing. It was the Dark Ages.
Shieber Do you see a supply chain for that stuff now when you think about the companies that you would partner with to make this happen?
Borshukov: Totally. All the big technology companies I think love the concept of this idea. We all understand it. There is a problem with shopping online. I don't shop online. There's a reason why I don't shop online. It's because everything I order doesn't fit me when it comes home, because there's no universal sizing. That problem has not been solved. There's been a hundred startups that have crashed and burned in the last 15 years. We tried it. We had a trial for two years with Hurley. It was super successful from as far as the data, but the business model didn't work. This was four years ago. I think now there's technology companies out there, the big Top 10 names that are starting to realize that what it's going to take to put this to the consumer. Because the consumer wants it. The apparel companies don't care, so we have to get a technology company to fund and seed that to really make it happen.
Shieber: Who's pushing the hardest on that front, from a hardware and enabling technology perspective?
Borshukov: I can't say.
Shieber: If you were to mention some companies that don't talk about what they're doing and are pretty impressive when it comes to this stuff.
Borshukov: Magic Leap will be one.
Shieber: HoloLens looks pretty impressive, too.
Aydemir: Yeah, that's pretty cool stuff.
Shieber: Do you see this integrating with AR or VR?
*Borshukov: * I think it's more AR. For me it's more AR because it's more flexible. VR is really about experiencing a presence. I think AR is about mixing things in the real world, obviously. We're talking about seeing something on your body and trying to decide whether you want to purchase, and when experiencing a product virtually in a way that sales will bridge the Sensorial Chasm. One of the reasons you don't shop online is because you're not quite sure. All you’re facing is a couple of rollover pictures and that's not always enough to make a decision. But if you have an experience...
Shieber: How does that play into Volumental's role in this?
*Aydemir: * The thing about this is that it's really about building trust and if it doesn't work I think consumers are really quick to say "Oh, this is some gimmicky thing. I tried it once. It didn't really help, so I will never try this again--maybe in the next 10 years or 5 years." I think we are trying to be super careful in bringing this to the public. It shouldn't be jittery or it shouldn't be laggy. It shouldn't be all these things. If your dog runs in front of you, it shouldn't just crap out. All these things. We turn down a lot of offers from handset manufacturers who are building these 3D cameras into the tablets for fit. They came to us and they said, "We want you to build a fit app for this.” You can, and you will probably make some money from that, but it would—
Borshukov: It would be horrible.
*Aydemir: * Exactly.
Borshukov: Because that's just the first step, the scanning of the body is one of the three important steps. You need the content, which is the garment, and then you need the visualization, which is the garment on someone's body. Those are difficult problems. The body scanning is the easiest.
Aydemir: Exactly and the same goes for algorithms and what we heard today previously. It's almost like that, I said this before, it's almost like that scene from that movie Full Metal Jacket. It's like "This is my algorithm."
Shieber: Did you just refer to Full Metal Jacket at a visualization conference?
*Aydemir: *Yeah, it’s like, there are many others like it but this one is mine...There are many others like it but this one is mine.
*Shieber: *That's an impressive reference there, an impressive reference. We're about five minutes out from the end of the talk. There’s one [audience question] up front...
Audience member: My question is, when you were replicating that shoemaker, did you actually replicate it independently, or did you have input for this talent, and then what happened to him...
*Shieber: *Are robots taking over? Did the shoemaker get fired because he wasn't as good as the visualization tech?
Aydemir: No, he wasn't getting fired. Because simply he can't be in two places at the same time. We probably prevented there being more shoemakers, if that makes sense...What we did is that we sent a couple of people down there and spent some quality time with these people and understand how they are...There is no institution in the world that says, “This is now .01 millimeter accuracy fit. Get your stamp.” There is no such thing. It's also about the psychology of it, that you are getting this premium experience. You have this trust that it's going to work and you don't really care if it's computer vision or you are doing some freaky X-ray thing. It's just that it works.
Second audience member: Psychology is the perfect word for my question, which has to do with the virtual mirror. When I've talked to online retailers about, "Wouldn't you like to take a 3D model of the customer and show them in the clothing?" they're all very, very clear: "No."
*Borshukov: *They don't like it.
Second audience member: They don't like it and, furthermore, some of the smarter ones I've talked to, they've done A/B testing. They don't necessarily have the 3D model, but they say, "Look, we show a model in the best possible conditions, with beautiful lighting. They look gorgeous. Most of this is women's clothing. The psychology is you want to give them the sense that they're going to look really great.”
*Aydemir: *This is great, yes. Most people don't want to see themselves in 3D. It's disgusting to see yourself in 3D if you don't have the body of Cristiano Ronaldo.
Shieber: We're all just wasting our time.
Second audience member: I want to see myself in clothing before I buy it, but I don't think my wife does.
Aydemir: In some, not in all cases. I want to verify that, but you get my point.
Borshukov: No, but this is my goal exactly. I know that, because I've lived this for six years, and you're absolutely right about it. The brands don't want that. They want to glamorize. They want the glamour shots. That's what generates the sales for them. But we have done enough research with consumers, and consumers ultimately want it. It's going to be a utility. This is why I said it's never going to work with a brand. They're never going to like it. But I think ultimately we as consumers have to take charge, and it's going take the technology companies, the Microsofts, the Intels, the Googles of this world, to enable this technology to reach critical mass. Then it's all going to be good. But the brands don't want it. First of all they don't have the funds to pay for it, and they're too conservative.
Shieber: What is the retailer then, or the brand company, that will come in and disaggregate from the bigger brands and say, "Look, we want to do this"? If it's hardware on one side, what's the...
Borshukov: I think it's about aggregation. You basically drive a truck into any warehouse facility and scan everything en masse. You need a scalable scanning solution. You don't ask anybody. Just go and scan the stuff and build a destination that's brand agnostic, if you wish. To enable that you need a Google or an Intel to put up the money just to get it going.
Shieber: With that we are actually out of time. Thank you all so much for paying attention. The panelists were great. I was hopefully more than mediocre. Thank you so much. You all have a good one.
Borshukov: Thank you.