Latest / Talk Commerce / Protecting Profitability in the Age of AI Shopping Agents with Jeff Otto of Riskified
Transcript
- Brent (00:01.885) Welcome to this episode of Talk Commerce. Today I have Jeff Otto. He is the CMO at Riskified. Jeff, give us your day-to-day role and one of your passions in life. Jeff Otto, CMO Riskified (00:12.27) Brent, it's great to great to meet you and thanks for having me on the show. I've been the the chief marketing officer at Riskified for over three years now. And, you know, my job that I that I really enjoy is telling success stories about how merchants not only grew their business and increased their GMV, but also did that in a way that protected their profitability. And what we try to do as a company is apply artificial intelligence across the entire surface area for where. consumers and now more and more their AI shopping agents, engage them and try to protect those merchants from bad actors, from fraudsters and abusers that are stealing their profitability. Brent (00:53.695) That's great. And any any particular passions you have outside of of Riskified Jeff Otto, CMO Riskified (00:59.856) I got I got two boys. The oldest one's playing baseball. It has been baseball, baseball, baseball. I got to, you know, be the proud dad to watch my son hit a grand slam home run. He's only seven. And and so we were all just you know cheering him on. So right now it's very much baseball is a is a passion. Brent (01:11.945) Wow. That's great. Brent (01:18.237) Yeah, awesome. Yeah, grand slam. That's that's that's exciting. seven, so is that that's post t ball? Was that Jeff Otto, CMO Riskified (01:27.744) It is, yeah. They're using the little machine pitch, you know, so yeah. And then, you know, they're figur they're starting to figure it out, right? They're they're at that great age where you start to see these moments of brilliance on the field and it's it's so fun to watch them develop and you know, from the beginning of the season to the end of the season, it's just amazing to see them play. Brent (01:34.301) Yeah. Brent (01:38.921) Right. Brent (01:47.905) I'll show my age. The next Rod Carew or the next Kirby Puckett could be coming in, huh? Jeff, before we get into content, I do this thing called the Free Joke Project. I'm gonna tell you a joke. You just give me a rating eight through thirteen. So here we go. My wife and I got into an argument over which vowel vowel is the most important. I won. Jeff Otto, CMO Riskified (01:52.472) Yeah, Kirby Puckett. There go. From the twins. Jeff Otto, CMO Riskified (02:06.669) Hmm. Jeff Otto, CMO Riskified (02:17.949) I'll give you I'll give you an eleven on that one. Very dangerous. Yeah. Brent (02:20.361) All right. Appreciate it. I f I if I knew I would have had a baseball joke lined up. all right, so give us the ten thousand view of Riskified to get started. Jeff Otto, CMO Riskified (02:26.23) Yeah. Jeff Otto, CMO Riskified (02:31.649) Yeah, so this is a dozen years ago, but basically the origin story was there was growing rampant payment fraud out there. E-commerce merchants were using tools that made you know thousands of rules-based decisions, if then else statements basically. And Riskified came along and said, you know, there is this exciting new technology, machine learning, we could train a model. We feed it a bunch of data. And guess what? It comes out with a better probabilistic decision than those rules engines would. And fast forward to today, where we are decisioning upon half a trillion dollars of GMV, you know, every couple of years. We're doing that globally. We're serving merchants all over the world as they're trying to do business all over the world and having meaningful seven-figure to eight-figure improvements on their profitability by doing that. Brent (03:29.535) Yeah, and there there's a balance there because the risk is that you're turning away a good actor with a false positive, right? Talk talk a little bit about how AI and machine learning is helping you to get over that if then, right? Jeff Otto, CMO Riskified (03:44.493) Yeah, absolutely. Yeah. The the problem used to be with the the if then else is, you know, you've got a situation where the worst thing you can do to a good consumer is falsely accuse them of being a fraudster. That is like a cardinal sin, you know. And so we use this framing of hey, do no harm, right? First do no harm. And when you can train AI models, in this case ML models, and also use a very large know, graph database to resolve identity, you just get better at making the right decision. And so a lot of this is about not just reducing fraud transactions or blocking them, but it's also about increasing approval rates and and reducing the number of those falsely accused consumers so that we're actually creating more economic success in e-commerce. Brent (04:38.077) Are you but we we in the green room we talked a little bit about customer service and I I I come from the agency background and the e-commerce background, and I can remember a client we had in Mexico back more than 10 years ago who had a whole room of customer service agents, and their job was just to make sure that the person existed. So they man instead of having an if then, they manually chased down. the the client or is is the client real? But you talked about even a worse pro product thing that's coming now is the fact that there's maybe fake people that are going to start calling in and bugging customer service. T talk a little bit about that. Jeff Otto, CMO Riskified (05:23.394) Yeah, yeah. So, you know, we we probably if if you've you've probably seen some of the the reporting on this and maybe seen examples of deep fake technology, you know, the voice, in some cases, video, very sophisticated social engineering email scripts that can be used. and what we're seeing is it's not just about protecting the checkout, it's also about protecting against and Brent, you'd be shocked at how big this problem is. A 76 billion dollar a year in losses problem that happens post-checkout just on the returns flows. So what we're talking about here is either, you know, in many cases a fraudster purchases something, then says, calls, calls the contact center of their of that merchant, you know, a few weeks later and says, Hey, I never got the package. Or they get the package, they take the good item out, they put whatever, I mean, there's Cans tuna fish, rocks, you know, whatever, ship it back, you know, fake the fake the return label and send it back. And then they've just gotten themselves, you know, something for free. And there's there is a a hundred different ways that this type of abuse can be done. I'm simplifying it dramatically. But the, you know, the National Retail Federation, you know, basically calculated this past year, this is a $76 billion problem. Now, the question might be, all right, what do we what do we do? Right. Well. Yeah, I guess option one was would be what you described 10 years ago. You know, go throw a bunch of humans against the problem and try to see, is this person calling me actually like who they say they are? even if they are who they say they are, maybe they're just being abusive and trying to steal. So you know, what we can do with new technology, and there's a technology that's incredibly powerful in this area, and it's called graph technology. And what we've spent the last five years building is an identity graph. And in simple terms, it allows you to resolve the human behind a bunch of fake accounts, burner burner phones, fake emails. So even if that bad actor goes and creates new accounts and then tries to do the same scam again, it is detected by the identity graph and they get blocked once or they get blocked again, meaning like every time they come back in. Jeff Otto, CMO Riskified (07:51.321) With a new fake account or a new fake email, they get detected, and then we we're not falling for the same thing twice. Now, what's cool about if I may, what's cool about the the merchant network is because we get a chance to work across, you know, all of these very large merchants from around the world, what we get to see is we get to see the same humans again and again. And if someone is mistreating a merchant over here and we catch them. If they try to go over to another merchant and do the same thing, well, they might be the f they might look like the first time customer for that other merchant, but we've seen them before. And so we can help that other merchant not fell fall for the same scam. Brent (08:36.063) Yeah, that's great. I you know that that sort of cross pollination is is one of the biggest advantages that any merchant would have to to learn who those who those bad actors are. How do you make sure that you're not somehow getting somebody into that list that is in fact a good actor? Do you have a way of like you know, the no fly list? I know there's a there's a way of making sure that you're you're exempt from that list or whatever if you somehow get get blocked. Jeff Otto, CMO Riskified (09:05.88) Yeah, yeah. There's there's a funny story. I was I was just doing a talk at a conference a couple of weeks ago. Are you a fan of remember the the show Better Call Saul? It was like a do you like that show? I I'm a huge I'm a huge fan. Bob Odenkirk like one of my favorite, you know, actors. And you know, on that show, for those of you who haven't seen it, you know, he's he's a he's he's multiple personalities. He's Jimmy McGill, who's like a kind of a s former scam artist turned lawyer. Brent (09:15.509) Yeah. Yeah. Jeff Otto, CMO Riskified (09:35.011) And then he he becomes Saul Goodman. And and so, like, that's a perfect example of like multiple personalities or multiple accounts or identities, right? But one human behind it. And to answer your question, in the show, he has this girlfriend, Kim, who lives with him. Now, Kim is actually not that, you know, she's she's a good egg, right? We wouldn't want just because she's made the probably the bad choice of living with Jimmy McGill, we don't want to punish her. For living at the same residence as him, maybe she uses his computer every once in a while. So you have to be very specific in the identity graph of understanding the relationship of the household and understanding these different variables so that you don't accidentally have a Kim situation where you're falsely accusing her of being a bad guy. Brent (10:26.805) Yeah, so if I go to get my Cinnabons, I'll I'll make sure that I can buy them, right? we in the green room we talked a little bit about agentic and how, you know, a lot of their at least Salesforce is now completely API forward. So that you don't even need a website. Are you are you gearing up for this idea of someday there'll be these agents that are going and buying for you and they won't be even a human involved on both sides? Jeff Otto, CMO Riskified (10:32.15) Yeah, exactly. Jeff Otto, CMO Riskified (10:54.284) Yeah, you know, agentic commerce is such an incredibly exciting area. You know, you think about how it's gonna help our industry, you know, the growth is gonna be phenomenal. The more convenience and collapsing of discovery through checkout that we can do through purchase, you know, that that is gonna drive, you know, more more consumption. And of course everybody's gonna benefit from that. You know, the the the thing that unfortunately we have to think about in our role in this ecosystem is how do we ensure that we're protecting, you know, we're protecting the the merchant from you know exploits that happen, you know, in every new channel you see this. And we won't we see the same thing in agentic commerce. I'll tell you a quick story. When when open AI and Stripe came out with agentic Commerce Protocol back last fall. One of the first things that we did is we said, you know what? All right, let's let's red team this. Let's test it out. Because we have to assume for a moment that there won't all be good consumers trying to use this to purchase. So we put together some synthetic IDs, some bad orders, and we ran it through. We put orders in that were so bad that we we were confident that any e-commerce, any mature e-commerce storefronts, fraud stack would catch it. And guess what happened? They were approved. The orders were approved. We went back, we went and called OpenAI. We let them know about it. We let the partners know. We said, hey, I think we need to maybe mature this a little bit. I think the protocol needs some maturing. We went back again in March, just a few months ago, tested it again. Same thing. Orders went right through. They were approved. And that was about a week before Sam Altman and the team said, you know what, we're going to take a step back. We're going to approach this a little differently. They kind of moved, you know, they moved back on the strategy a little bit. Now, look, there was a lot of reasons why instant checkout didn't go forward and fraud was only one of them. But what I would say is is it is a cautionary tale that we need to be careful here. And we need to advocate for all protocol, you know, protocols that are being worked on, whether that's Google's UCP, whether that's, you know, Jeff Otto, CMO Riskified (13:17.976) the work that the the Stripe is doing, the work that other protocols are doing. And we have to say, you know what, we have to help ensure that if it's an AI lab led agentic shopping flow, let's make sure that we're passing the relevant data through so we can make the right decision and so that the merchant is protected from fraud. Brent (13:39.135) Yeah, in your case it it's not it's not as important because you you want to get them at the checkout. how far into the customer journey do would you start? And I I remember hearing about systems that would try to help identify people before they even get to the cart if they're bad actors or they're or or they may may have do something that helps them to identify a little further before. Yeah, so you you have an opportunity while they're shopping to kind of learn about them. I is that part of a strategy that works? Jeff Otto, CMO Riskified (14:11.63) Yeah. Yeah, yeah. So, you know, you think about and this is not something that our our company does. We're not necessarily a perimeter player, but if you think about the folks on the perimeter that are doing bot detection and other things, that's like all obviously a nice early warning system for is this a good actor that's coming into the site to start to shop. And then once you start getting into the checkout flow, if you have some information about that consumer, maybe they were cookied previously where cookies are still available, or we're able to get some information in the checkout flow, then we can run an identity check and we can say, hey, let's go take a look at the graph, go cluster this identity off of this variable, email or something that they've just put in, and let's see what comes back. Is this a is this a a good, you know, healthy identity? Is this an identity that maybe has you know, some activity that is nefarious or something we want be concerned about. And then we can be adaptive. It's not about saying no, it's about maybe stepping up and and saying, okay, let's step up and try to gather a bit more information before we take action, before we accept payment, before we say yes to this claim, you know, things like that. Right. So I think that's all that's what it's all about. You got to be surgical with these things. And I think now the technology's matured to a point where you can be surgical. Brent (15:33.727) Yeah, that we if you're if you're you if you're talking about chargebacks, you're talking about the risk here, is the quantitative savings for the merchant not just losing the product, but also saving the maybe the reputation for chargebacks and there's some there's gonna be some kind of different level of discounts you're gonna get on your transactions. Talk a little bit about the metrics that are all involved in in in the client and the merchant. and how they can save money there. Jeff Otto, CMO Riskified (16:04.652) Yeah, you know, that's a Brent, that's a really great point because the analogy that we like to use is, you know, the fraud rate, right, or the cost of the chart back is just the part of the iceberg that you can see above the water. Right. So you have to think about all of that value stack together. It's not just about, okay, I just had to go pay back the issuing bank because I, you know, I lost I, you know, it was a fraudulent transaction. It's also, okay. I I shipped a product and I had all this, you know, logistical cost or reverse logistics if they then filed a claim. So like, you know, there's there's cost. There's, you know, all right, I I had to deal with a customer service inquiry when, you know, maybe that person that had their Credit card stolen end up calling me and being like, hey, you know, I didn't pay for this from you, you know, dah, dah. And then now my contact center had to deal with that cost, right? So it's important that you've got to look at the whole thing. And also, there's been some really great research that shows that even though, you know, it wasn't the merchant's fault that their credit card was stolen, it's they will blame the merchant. Because they saw that fraudulent charge at that, you know, from that merchant. And they and that actually changes the way that their brain thinks about the brand of that merchant. And so that's another thing why it's so important for us to, you know, protect not only the consumer identity, but also think about the cost of fraud in a more comprehensive way. Brent (17:43.926) Yeah, and a lot of that cost is just around reputation and how that spreads once it's happened and if it's happened on scale, the merchant ha the client has less le less ideas that they would ever want to purchase from you. W what what what are you seeing coming in now? What's what's exciting for you in this next couple of quarters in terms of risk and protection? Jeff Otto, CMO Riskified (18:10.094) Yeah, you know, we're we're excited about a few things. You know, there's the the ability to use AI agents within you know our platforms interface to help merchants see patterns about their consumers that are interac interacting with them. That's incredibly exciting. we're very excited about what we're seeing protecting beyond checkout. You know, I talked briefly about the contact center agents. You know, they've got such a tough job, Brent. You know, you imagine you got these CSRs, they're metric on making people happy. You know, it's like it's customer satisfaction. It's net promoter score. It's they got to do it quickly. They're they've got, you know, they're they're timed on their call resolution. And at the same time, There's some cohort, smaller cohort of bad people that are going in and calling them and trying to socially engineer them and take advantage of them and say, hey, I want a refund. I'm, you know, mad or whatever. I didn't get X, Y, and Z, and lying through their teeth. And these poor CSRs are are forced to concede that. And that, yes, they they did the right thing based on how they're metric, but then the company lost money. The company was able to give a fraudulent person a a a fraudulent refund, you know, and that's what we're trying to stop. So if you can in that moment, that critical moment where that, you know, person that fraudster calls in or they start a chat session or they're doing whatever, if you can provide an identity signal at that moment and help that agent make the right call, or even route that person away from the agent, take it to a specialist team, you know, that that that deals with these hairy situations, that can really help. That can really help protect the brand, protect the profitability. Brent (20:10.421) Yeah, the the rise of the AI agent who's helping customers on the chat is is definitely escalating now. Are you helping to train those agents as well to maybe steer calls or steer conversations in the right place? When maybe you've you you should be able to identify somebody that's chatting with a bot and they should go somewhere or just get kicked off. Jeff Otto, CMO Riskified (20:27.694) Yeah. Jeff Otto, CMO Riskified (20:33.282) Yeah. Jeff Otto, CMO Riskified (20:37.912) Yeah, you know, you know, I I think the there's there's a trade-off that I that I see and and this is when I talk to executives at merchants that we work with, we often say, Okay, you have a you have a choice of how to invest your time and energy into agentic commerce. You can delegate this and and take advantage of what the AI labs are doing, and you can create you know, a a product catalog that it that you know the protocols can consume, you know, whether through it's you know an MCP server or whatever. And and you can do that. And that and that will result in a couple of things. One, you're gonna lose a little bit of your brand identity because now someone is shopping inside of Chat GPT, right? They're seeing just the way that that information is curated inside of the large language models interface. The alternative though is why not also invest in a genetic shopping on Your very mature e-commerce storefront. You own the experience, you own the the data visibility, your CDP is under the hood, so you can do segmentation and you understand better who you're dealing with. and you own the loyalty piece of it. And then you also, guess what? You can take advantage of your existing fraud tech stack that you've already invested in for the last however many years. So, you know, you you you know that all of that prior investment can be utilized and connected in. So, and we're seeing this, right? I mean, whether it's you know, you got Rufus with Amazon, you've got, you know, Ask Macy's has been a phenomenal success story over there. you know, merchant-owned AI shopping assistants can really boost conversion rates. And they can do it in a way that is safer and protects profitability. From the bad actors. Brent (22:37.045) Yeah, that's I mean, that's such a good good point there. And they they can also make it worse if they're a poor agent, right? If if it's a poorly trained agent. And I can think of a couple of big brands that have decided just I'm not even gonna bother with their chat because it's so bad. It's like Chat GPT three or something. Jeff, we have a few minutes left. as I close out the podcast to give everybody a chance to do a shameless plug about anything they'd like. What would you like to plug today? Jeff Otto, CMO Riskified (23:06.448) well, you know, we we are very proud of the work that we've done in detecting payment fraud. and what we've done is we've expanded that into the post checkout areas with a product called Policy Protect. And Policy Protect is about all of those instances where a merchant can be abused on their refund claims, their return claims, perhaps their promotions that they're doing, right? Offering loyalty discounts or things to new, maybe new customer discounts, things like that. And what we're doing under the hood is we're using identity intelligence to figure out who is it that's actually requesting that, whether they're doing it directly on the on the merchant's website, maybe their contact and contact center. And what we've done is we've rolled out a platform that allows e-commerce leaders CX leaders, fraud leaders to look at this and and they can and we can help them make the right decisions. And this has saved certain merchants between, you know, seven and eight figures annually on on their profitability. So it's it's a like NRF, the NRF said $76 billion problem. We can take a meaningful bite out of that and really help merchant profitability. Brent (24:29.791) That's perfect. Jeff, how do people get in touch with you? Jeff Otto, CMO Riskified (24:33.624) Yep. with me personally, reach out to me on LinkedIn. I'll Brent, I'll get you my LinkedIn profile so anybody can send me a a a note or an email. And for riskified, you go to riskified.com. You can learn all about us. Brent (24:47.563) That's perfect. Jeff Otto is a CMO at Riskified. Thank you so so thank you so much for being here today. Jeff Otto, CMO Riskified (24:53.283) Thank you.