Episode 97
March 6, 2019

Deep Fakes for Commerce: A New Era of Personalization for Retail

CVS launches "Beauty Mark", its truth-in-advertising campaign; while AI is generating faces that are plausibly real. Meanwhile companies like SuperPersonal are putting customers into model try-on videos. Have "deep fakes" - AI algorithms that map faces and micro-expressions onto stock footage - come to retail? How can they help? How can they hurt?

this episode sponsored by

No items found.

CVS launches "Beauty Mark," its truth-in-advertising campaign; while AI is generating plausibly "real faces." Meanwhile, companies like SuperPersonal are putting customers into model try-on videos. Have "deep fakes" - AI algorithms that map faces and micro-expressions onto stock footage - come to retail? How can they help? How can they hurt?

Show Notes:

Main Takeaways:

  • Brian and Phillip are podcasting live from #Shoptalk2019!
  • Deep fakes are getting a little too real to be comfortable.
  • Personal body mapping for try-on is becoming a reality.
  • Can companies figure out how to keep their data in-house?

Who's Waldo: Can Humans Even Spot Deep Fakes Anymore?

Personalization in 2020: Turning Regular People Into Models:

  • Personalization, especially in retail has become a theme of 2019, and the tech is finally catching up.
  • Phillip says that while most virtual try on applications are not very good, Warby Parker has changed the game.
  • Warby Parker's AR powered virtual try on is so good, it's almost like looking in a mirror, and they are using the same depth map as Apple's facial recognition software for iPhone.
  • Another company that's working to change the virtual try-on experience is SuperPersonal, an AI-powered virtual dressing room experience that would allow retailers to "multiply e-commerce photography to account for different ethnicities, skin-colors, and age-groups, without the need to shoot multiple models".
  • "Personalization in 2020 is the whole website is literally you".
  • Brian makes the point that because of the last 6-8 months of advancements in AI and machine vision, models will not be needed, and will only be required as "aspirational content."

Levi's New Story: From Finished Goods to Customizable Clothing:

How Can Companies Get to Know Their Omnichannel Customers?

  • So because 2019 is the year of clientelling, retailers and brands are having to build relationships with their customers, and they need the data to do it.
  • Phillip points out that the more companies aggregate the data in-house and operationalize it as a tech company, the more they will be able to figure out what works, and what doesn't.
  • During Brian's interview with Chris Homer from thredUP, Chris mentioned that thredUP has a policy of testing internally, and figuring out what works in-house, before bringing in tools to supplement those processes.
  • Companies need to figure out what works best for them and double down on that, and they also need to build real systems to house all the data that is collected, in order to utilize it effectively.

There's so much more to see and experience at Shoptalk2019! Stay tuned for more insights, and highlights from the show! Also, let us know, what was your favorite part of #Shoptalk2019 so far?

Go over to Futurecommerce.fm and give us your feedback! We love to hear from our listeners!

Retail Tech is moving fast and Future Commerce is moving faster.

Download MP3 (25.7 MB)


Phillip: [00:01:20] Hello and welcome to Future Commerce, the podcast about cutting edge and next generation commerce. I'm Phillip.


Brian: [00:01:24] I'm Brian.


Phillip: [00:01:25] And we are live at Shop Talk 2019.


Brian: [00:01:28] Live in the podcast booth.


Phillip: [00:01:30] Make some noise. Special thanks to the Shop Talk team for helping put this together. And this is the first of probably a few episodes that we'll be doing. And we're gonna start with some crazy news. So I have to get this out of my system because I've been saving it for a few weeks. We've had a lot of interview shows recently.


Brian: [00:01:50] We have. Yeah, I know. It's been hard because we want to talk about stuff.


Phillip: [00:01:53] Yeah.


Brian: [00:01:54] And then we're like, "Oh, well we've got to post the interview with ThredUp and...".


Phillip: [00:01:57] Which was awesome.


Brian: [00:01:59] Yeah. And Ann Inc.


Phillip: [00:01:59] Yeah, and Ann Inc. And that was... By the way, you did an amazing job at eTAIL West.


Brian: [00:02:04] Thanks, man. That was fun.


Phillip: [00:02:04] We had a great presence there. And man, I really love some of the content that came out, especially with Ingrid. I felt like that was really authentic interview. If you haven't listened to that interview, please go check that out, because that's just... It was was awesome. Actually, the ThredUp interview was awesome, too.


Brian: [00:02:20] It was. No, both interviews were really, really fun to do. Those were both Ingrid and Chris. Super smart people also. Yeah.


Phillip: [00:02:27] Of for sure. Yeah, I. OK. So I wanted to, I want to get deep into something real quick. Deep being the operative word because we don't really have a ton of time. We've got a couple interviews coming up here at Shop Talk. But I can't believe... Because very often we are, I want to say that we're like ahead of the curve, but we're kind of ahead of the curve.


Brian: [00:02:48] Yeah. You know.


Phillip: [00:02:49] No to toot our own horns or anything. But we were really far ahead of the curve when we started talking about deep fakes middle of last year.


Brian: [00:02:57] Yeah, I think we beat LinkedIn to sort of talking about them in a business context.


Phillip: [00:03:02] Right.


Brian: [00:03:03] And then like maybe a month after we posted that interview, LinkedIn did this whole story on it, and it was a really big deal. And now deep fakes are everywhere. Yeah. This is a big deal.


Phillip: [00:03:12] One of the things... So there's a couple stories that landed last week out of nowhere that were in this realm. One of them was an ecommerce, is a startup that's commerce focused, that is actually applying the technology to be a consumer centric technology, which we'll talk about in a second. But the first story I found on Fast Company is talking about an image authentication startup called Truepic. And it's talking about they've developed an ability to spot doctored images or fake social media accounts, and they can also use that technology to apply towards authenticating deep fakes. So Truepic is, you know, it's a startup. There's a lot of startups that are kind of launching into this space, but realizing that like there's broad application for this. Journalists need this. Lots of people need this. But it came back to mind recently because I walked into my local CVS drugstore, and CVS has rolled out a brand new program where any image in store that has not been retouched with Photoshop has a special icon on it so that you know that it's not a doctored image.


Brian: [00:04:21] Oh, my gosh, that's amazing.


Phillip: [00:04:22] Yeah. So there is transparency in retail around that, which is, by the way, it's coming out of probably the need, especially if you're a global brand. There was a law that was passed in France that is being widely accepted as part of the EU's, you know, sort of truth in advertising initiative. And the fact that it's a socially conscious retail presence of something that we've been talking about for two years now.


Brian: [00:04:51] Right. Totally. No, it's a really good point. And actually, I think we didn't even talk about that...


Phillip: [00:04:55] Yeah. Oh, for sure.


Brian: [00:04:56] ... law from France in that original episode we did. You know, it is just interesting to see how this is going to be applied even more and more, like Photoshop is one thing. But like deep fakes on such a different level that I feel like, you know, having that transparency about what's been doctored and what's not been doctored is going to become a bigger and bigger discussion.


Phillip: [00:05:22] And think about how that could be used by social media companies to be able to spot and automatically so... So think about how we're already using technology like this today. I don't know if you've ever seen a Facebook video where the audio is muted because there's copyrighted material in it.


Brian: [00:05:42] Right.


Phillip: [00:05:43] I think that this is the type of technology that will be broadly adopted and widely used in all kinds of contexts, whether it's for social media posts that have been doctored. Or social media posts that are deep fake social media posts. And by the way, for those who aren't familiar with the term, just to give you a little bit without explainifying it, deep fakes is a bit like a cottage term for the application of AI to map a video stream of a celebrity's face onto pornographic images. It started in still images. And then AI got so good it could do it to video and it was blurring the line of reality. It was kind of shocking.


Brian: [00:06:24] That's where it started. It's well, well beyond that now.


Phillip: [00:06:26] Oh well beyond that now.


Brian: [00:06:27] On Twitter... I just saw this, this was the craziest thing ever, Jennifer Lawrence was giving some talk and someone put Steve Buscemi's face, like morphed her face with his. It is the scariest thing I have ever watched, ever.


Phillip: [00:06:46] It's uncanny good. And in fact, it is uncanny. It's sort of in the uncanny valley. Right? Like it's semi plausible or semi believable. But there's a line there where it's not quite believable, but they're approaching believable... The believability factor is incredibly high.


Brian: [00:07:04] No, no, no, no, no. The only reason why I wouldn't believe it is because I know both of those people because they're celebrities.


Phillip: [00:07:13] Right. Well, it was her voice coming out of his head.


Brian: [00:07:15] Right. So there's definitely situations where I have no idea because I don't know anyone that's involved or any of the images that are involved. And if you combine with something you just sent me recently, which was the random face generator...


Phillip: [00:07:30] Yes. This is what we are going to come back to in a little bit.


Brian: [00:07:34] Totally.


Phillip: [00:07:36] Yeah. So you combine that with there's another technology. It was recently put out. So there's an in video developer that worked for Uber.


Brian: [00:07:43] Right.


Phillip: [00:07:44] This went viral a couple weeks ago, where given the wealth of training data, he was able to create a random face generator called ThisPersonDoesNotExist.com, and that face generator creates plausibly, plausible, like these are faces that look plausibly real. They look like they might be social media images of people that you might know. They're stunningly good. Now sometimes they're stunningly bad. And that's kind of the genius or the beauty of this is that you know that it's not just, they've farmed a bunch of social media images and they're showing them to you. You can tell on some of them. Not all of them. I would say nine out of ten could absolutely pass for real people.


Brian: [00:08:31] Absolutely. You know, occasionally you'd be like, "Ok yeah, that's a little bit off.


Phillip: [00:08:35] Like the glasses are sort of half there and missing a little bit on some of them. But they're shockingly good. And it doesn't just do adults and it doesn't just do Caucasians. It is pretty good at generating children's faces, people of color... It's freaky good.


Brian: [00:08:52] It's freaky good.


Phillip: [00:08:53] Yeah. And there was this quote that blew my mind, by the way, which was, if you think about the acceleration of technology five years ago, you know, we could barely recognize faces in images. And three years ago, we could recognize objects. And now we're generating faces. And I think, and this quote on Reddit said. And in five years time, we're gonna need AI to be able to distinguish what a real face actually is.


Brian: [00:09:23] Right.


Phillip: [00:09:24] And that's even in video we're talking about like 4K Ultra HD video, we're heading on a trajectory where it's going to be very difficult to to tell reality from a generated reality, which is...


Brian: [00:09:40] For people, let alone AI. You're going to have to, like you said, you're gonna have to have AI because we're not good enough.


Phillip: [00:09:45] No, not at all.


Brian: [00:09:46] We're not good enough to be able to tell.


Phillip: [00:09:48] Ok, so let's talk about how this works in commerce, because I'm trying to drive to this killer thing that blows my mind. Yes. There is this gif that went around a year or two ago where this guy was like, "All I want is to be able to see if I look good in this pair of shoes." And he had this picture that he cut out of himself standing to the side and he's like scrolling on his phone, like a category listing page on like a web site. And he's like putting himself, it's cut off at the ankle... He puts himself in the shoe like, you know, he's physically, like a physical piece of paper with his body printed, he's like putting himself in that shoe.


Brian: [00:10:24] Man, this is what I've been talking about.


Phillip: [00:10:25] Right. Exactly. This is exactly what you've been talking about. But there is a real business, a startup called Superpersonal, which is doing this in AI now. And they're taking the same technology that you would have with deep fakes, which is a stream of hundreds of thousands of images, or a long, lengthy video of your face, and they're mapping it onto a model. Have you seen this, Brian?


Brian: [00:10:49] Yeah, briefly.


Phillip: [00:10:50] This is, when you look at that, and we'll link it up in the show notes.


Brian: [00:10:54] It's stupid amazing.


Phillip: [00:10:55] It's stupid good.


Brian: [00:10:56] Yeah.


Phillip: [00:10:57] That you could put yourself into the images on a web site, and the entire... When you talk about personalization, it's about product selection or prediction around what you might like... Screw that, personalization in 2020 is the whole web site is literally you.


Brian: [00:11:13] It's you.


Phillip: [00:11:14] It is you.


Brian: [00:11:15] Exactly. Exactly. We don't even need models.


Phillip: [00:11:17] You don't need models.


Brian: [00:11:19] We need models for aspirational content. And that is it. Everything else is you. This is Episode 8, man.


Phillip: [00:11:26] Yeah. This is exactly what you've been talking about for two years. I'm finally onboard. It only took me two years to get there.


Brian: [00:11:31] No, you went for it.


Phillip: [00:11:33] It sounded so far fetched though. Two years ago it sounded farfetched.


Brian: [00:11:38] Not to me.


Phillip: [00:11:38] And in six or eight months time, we actually see real application of actual technology.


Brian: [00:11:45] This type of technology, this acceleration of machine vision and AI and sort of this application of body data, this is the new super computer. This is the new supercomputer.


Phillip: [00:12:02] Right.


Brian: [00:12:03] What I'm getting at here is, I feel like technology accelerated so hard on the hardware side for so long, now... That new acceleration is all about machine vision and learning. And we're going to see leaps and bounds at a level that I think people don't even understand, and it's going to be applied in some freaky ways. In China, it's already being applied in some freaky ways. People are you know... The Chinese government is using facial recognition to shame jaywalkers. Right?


Phillip: [00:12:43] Wow. Yeah.


Brian: [00:12:43] Yeah. Yeah. This is a real thing. And so initially it wasn't very good. And like, sometimes people would get shamed that weren't supposed to get shamed. But they've gotten a lot better at it now. And so, you know, in short...


Phillip: [00:13:01] Yeah. Yeah. There's always going to be some, you know...


Brian: [00:13:04] There's crazy, crazy application here, but...


Phillip: [00:13:05] Right. Collateral damage, you know, the collateral damage splash zone, if you will, of the application of these types of young technology.


Brian: [00:13:14] Right. But I mean, that sounds... Like a few years ago people would've said no way.


Phillip: [00:13:18] Right. For sure.


Brian: [00:13:19] Now, this is here and now... It's something you as a consumer, you as a retailer, you as a technology provider, you are gonna have to game plan for this technology. Kind of like people had to game plan for mobile like this is the next wave.


Phillip: [00:15:27] You know, you combine this with a lot of AI.... One of the things that I've sort of been skeptical about AR... And, you know, there's been a lot of apps, like try on apps, for augmented reality, and they're not terribly good. One that is terribly good that we saw recently was the Warby Parker, the new Warby Parker IOS app.


Brian: [00:15:48] Yeah.


Phillip: [00:15:50] So it uses the depth map sensor, which Apple just opened up to developers, but it uses the actual the same face ID depth map sensors that you would have for the face unlock for the iPhone. It's using that to get realistic depth map and size and scale mapping, so that you can have actual like real try on. It's kind of shocking. It's so shockingly good. It kind of looks like you're looking in the phone camera as a mirror, and it maps it perfectly. It doesn't jump around, and it shows you if the glasses are probably going to be too small for your face. It's crazy good. And I wonder if the next generation of user generated content...


Brian: [00:16:39] Bingo. You're right on point.


Phillip: [00:16:41] Right.


Brian: [00:16:41] I was thinking about this like, for glasses it's not a big deal because that's something that's very easy to get a hold of and try on. But what about things that are a lot harder to get places?


Phillip: [00:16:51] Right.


Brian: [00:16:51] So UGC for like couches. Right? Things that are hard to ship around and move around and get to a photo studio or to, you know, to collect images on. It's gonna get way easier.


Phillip: [00:17:07] I think about how the... So there is an interesting intersection here. So we just saw some of the opening sessions, opening keynote, and the opening session with Levi's in particular, which I could talk an entire hour about... But one of the opening sessions around Levi's, which we've talked about some time ago, the retail press picked up the Levi's customization story and their whole initiative and the workshop around personalization of jeans and designing of jeans and some of their retail stores. They picked up that is, "Oh jeans with frickin laser beams." Right?


Brian: [00:17:44] Right.


Phillip: [00:17:44] Which is, I think, such a sad miss because it's actually a really smart move by them to reorient how they think of themselves instead of "We are a finished goods brand." They are saying, "We are no longer thinking of ourselves as a 'We make something, and we have to anticipate the demand model of who wants it,'" and they're now thinking of it. "We hold unfinished goods and finish them just in time for delivery.".


Brian: [00:18:15] Exactly.


Phillip: [00:18:16] And that's a complete departure of the relationship with the customer. It's a completely different business model, which I think is stunningly cool.


Brian: [00:18:24] We're just at the start of this.


Phillip: [00:18:25] Oh, sure.


Brian: [00:18:26] Like the Nike Innovation store on fifth. That was like...


Phillip: [00:18:31] The House of Innovation.


Brian: [00:18:34] The House of Innovation. That whole blank shoe thing was like a flagship sort of experience, right?


Phillip: [00:18:39] Right.


Brian: [00:18:39] But this is going to become commonplace. Man, getting back to Episode 8 again, where we talked about the idea of maybe you even have just like, you're neighborhood customization station, where it's like you get a different level of quality based on brands of like blanks and then you take it in and it gets customized, or it gets shipped to that customization depot. It gets customized, and you pick it up or you can pay to have it delivered to your house.


Phillip: [00:19:10] I see that being able to happen. What Levi's talked about is because it's such a departure, they've eliminated some of the steps around the product customization that took manual effort, and they're replicating that with things like lasers and treatments and finishes that happen closer to its efficiency and supply chain. But this is the thing that blew my mind... Their fulfillment centers are no longer just distribution centers that box up and pick pack and ship products. They are part of supply chain now.


Brian: [00:19:50] Right.


Phillip: [00:19:51] Because they're actually manufacturing.


Brian: [00:19:52] Right.


Phillip: [00:19:52] It's not anymore just about, "I'm good at logistics and shipping and doing direct to consumer shipping." It's, "I am a fundamental part of the manufacturing process for customized goods," which is a completely different world.


Brian: [00:20:06] Right.


Phillip: [00:20:06] And when you think about the technology... I was saying this earlier to somebody, and it doesn't matter who... we'll have that person on the show, I'm sure at some point in the future... But I was just talking to them about one of the hardest, one of the biggest things that a company is faced with right now is trying to understand the omni channel customer and attribute omni channel behavior and stitch together customer journey. And every major brand is giving that data away to people who already have solutions for it. I think that the long term play is that if the more that a brand retains their own data and is able to operationalize it as a technology company and be able to do that for themselves, the actual like... That's what Google has done with their cloud platform. That's what Amazon has done with web services. That's what the 20 year behemoth players, the unicorns in our space, don't give their data away to other people for short term efficiency and for short term gain. The long term is if you can take a longer view on it and find a way to maybe do something like site personalization, like Superpersonal, but you're not necessarily partnering directly with someone to do with speed to market short term opportunity play, and instead you invest in bringing that in-house. We just saw recently... Who was it that just hired an AI officer? Was it Levi's?


Brian: [00:21:49] Yes, I think it was.


Phillip: [00:21:50] Was it Levi's hired an AIO? an AIO?


Brian: [00:21:53] Yes.


Phillip: [00:21:54] And by signaling to the market that you're investing to put that technology in-house and make it part of your culture instead of farming it out to someone else, I think that's the real differentiators in our space, that's what they're doing now. And some of them, it'll require them to move a little slower, and some of them will fail in the process. But I think the ones who actually emerge as the most victorious and the differentiators and the disruptors, those are the ones who have put that work in.


Brian: [00:22:22] I feel like Chris Homer from ThredUp had a really good model here, which was test internally, do things internally, make sure you're on the right path, and figure out which tools you need to bring in.


Phillip: [00:22:34] Yes, for sure.


Brian: [00:22:36] Definitely keep your data. For sure. I don't think your customers are gonna watch you give all your data. A hundred percent there. But like, have your own practices, and then find out where you need to bring in third parties and be thoughtful about it.


Phillip: [00:22:51] Right.


Brian: [00:22:52] Fail fast internally before you go out other places.


Phillip: [00:22:56] Oh for sure. I think that the scary thing there is that you will fail.


Brian: [00:22:59] Yes.


Phillip: [00:23:00] You're going to fail at some point. So you need a model of resiliency as a business to be able to survive the failures and sort the wheat from the chaff, if you will, and figure out what is working for you and double down on that.


Brian: [00:23:18] That has to be a discipline.


Phillip: [00:23:20] Oh, everywhere, not just technology.


Brian: [00:23:22] It's an organizational problem. And actually, this is exactly what Chris talked about. Basically, you need to have a system for how to deal with your data that is organizational. It's a cultural thing. And if you don't have that, you're not going to be able to capitalize on what you talked about, which is the long tail. So I thought that was a phenomenal recommendation from him.


Phillip: [00:23:45] So I we haven't seen any...we need a better word than deep fakes, but we haven't seen any deep fake type companies that are applying AI to facial mapping here at Shop Talk, yet.


Brian: [00:23:59] Well, I mean, we've hardly touched it. I mean, it's really the first day. You know, we're gonna go walk. There's a whole section here dedicated to AI. So I'm really excited to get in there and see if we can find anything that's like retail ready there. But there are some cool brands who I've already seen, some that we've mentioned on the show before. I'm really excited to see what else is out there. I always love to see, you know, I feel like Shop Talk attracts a lot of the new retail technologies that are really starting to get their feet on the ground or like get their feet under them.


Phillip: [00:24:32] Right.


Brian: [00:24:33] And so this is a good spot. If you're a retailer, this is a great spot to get out there and find something new that can help you grow your business or solve a problem.


Phillip: [00:24:43] Yeah, OK. So what are you looking forward to? It's not really Shop Talk centric show. I'm sure, we'll talk about that. But what are you looking forward to for the rest of today? Is there anything in particular you're looking at?


Brian: [00:24:56] Well, I'm really excited about our next 30 minute segment here where we're going to interview Lively. That would be really cool.


Phillip: [00:25:02] Very cool brand.


Brian: [00:25:03] Yeah, very, very cool brands. And there are a few sessions that look interesting. I'm also excited that I'm going to be spending time this evening with GGV, which will be really interesting to see, you know, kind of what they have going. And I go to dinner with them every Shop Talk. So that will be interesting.


Phillip: [00:25:26] That's awesome. Well, stay tuned for more content from Shop Talk and more interviews. You know, we're just scratching the surface here. It's going to be an awesome couple of days. And, you know, we want to know what you think about all this stuff. Do you see...if you are running a direct to consumer fashion brand, maybe there's application. There seems to be obvious application for facial mapping and site personalization. But is there a creative use outside of that? I want to hear from our audience and you can do that the best by going over to FutureCommerce.fm or follow us on social. We're very active on Instagram these days and on Twitter. And that's Commerce Future. And we want to hear from you. So lend your voice conversation, and let us know what you think. How can you put this stuff to use for good in your business?


Brian: [00:26:14] For good. For good.


Phillip: [00:26:16] Yeah. For good in your business. I'd love to hear more about that. Send it on over. All right. Until next time, retail tech moves fast...


Brian: [00:26:24] But Future Commerce is moving faster.


Recent episodes

LATEST PODCASTS