In person and online. My name is Cris Kennedy. I'm a research analyst at William Blair, covering the fintech and payment space. For a complete list of research disclosures and/or potential conflicts of interest, please visit our website at williamblair.com. Next up, we have Riskified. This is the second time they've attended our conference, so we're pleased to have them again here. From the company, we have the Co-founder and CEO, Eido Gal. Riskified was founded in 2013. They came public in 2021. At the core, they provide fraud management solutions that help their customers grow while managing risk. The Chargeback Guarantee product was really the core product, but they've been expanding into new areas, which I'm sure we're going to hear about. With that, let me pass it over.
Thanks, Cris. That introduction was my first slide, so I could skip it as well. All right, so nice to meet everyone. Eido, co-founder, CEO. By the way, just a bit of background. I started in the Israeli intelligence community working on online identities. Afterwards, worked for a company called Fraud Sciences in the world of e-commerce and fraud prevention. That company was acquired by PayPal, so that's how I kind of got into this world and had some domain knowledge. Being from Tel Aviv, worked in another startup on kind of more on analytics machine learning side, which is where I met my co-founder, Asaf. At some point, I said, "Hey, you know, I know that there's this problem of e-commerce fraud management. It's really challenging and costly for merchants.
They unfortunately accept bad transactions, and that results in a lot of loss that they need to pay out because they're liable in that event. It also causes them to turn away a lot of great transactions. I think that if we build, you know, kind of a machine learning-based solution to tackle e-commerce fraud, we can provide them a lot of value. I said, "You know what? That sounds like a really great idea." That is how we started to build and ramp Riskified. Fast forward to today, that was early 2013. We built an e-commerce fraud prevention company focused on enterprise merchants, went public a few years ago. Really proud of the fact that we work with over 50 publicly traded companies managing that entire process for them.
If you think about the GMV flowing through our system last year, it was over $140 billion. If you were just to think about us like an acquirer, we would actually be one of the largest acquirers just based on that, you know, definitely within the top 10. We focus on working with large strategic accounts. We believe that large enterprises choose best-of-breed solutions. They are able to go from kind of a more standard platform approach towards really focusing on what is the best solution for them in this specific category. Our pricing is risk-adjusted. That helps us work across different industries and geographies that have different risk profiles.
You can see someone like a—I'm just orienting myself around this—like a Blackhawk Networks, which, you know, kind of digital gift cards preloaded with cash, obviously extremely high risk, towards someone like an Allbirds, which is, you know, we would consider a lower risk category. The fact that we do risk-adjusted pricing really means that there's value for everyone here. Let me take a, you know, a step back. When you think about the cost structure of an e-commerce merchant for managing fraud, okay, the basis would be the first and most important piece is the cost of chargebacks. Just to give you kind of sample numbers, let's say that's 25 basis points. That basically means that for all the revenue that that merchant is experiencing, they need to pay out in chargebacks 25 basis points. That could be a very substantial amount.
That is because merchants have liability in the case of a chargeback for an e-commerce transaction. If I steal Chet's credit card and place an order online with one of these merchants, they accept that transaction because they do not know it is a stolen credit card. They ship me the goods. Chet sees it on his statement. He calls Allbirds and says, sorry, he calls Visa or his issuing bank and says, "You know what? Someone stole my credit card. It was not me." Chet gets refunded. He is protected as a consumer if you have ever filed a chargeback. Now the merchant is liable. When we say that is their cost structure, the 25 basis points, that is the amount that they are paying out because they have liability, because they accepted a fraudulent transaction. That is, let us say, 25 basis points on average. What they pay for tools traditionally was much lower than that.
Okay, it was, let's say, in the range of two basis points. You would pay a few cents per transaction for someone to screen that transaction, recommend, accept, decline, or allow you to do a review process or to do something with the transaction. You would have a team, and that team would cost three basis points. Your overall cost structure is 30 basis points. Most of it goes to paying back chargebacks, a small amount to the service and the tools that you're using, and some of it for staffing. When we started Riskified, okay, we kind of said, "Look, we don't want to be that two basis point small solution. We actually think we can provide more value, be extremely accurate, and provide guaranteed results to our merchants." We said, "We're not just going to provide them a recommendation and take two basis points.
We think we can be an end-to-end solution and guarantee it. We come to our merchants and say, "Hey, if your cost structure is 30 basis points, our fee is going to be lower, is going to be a discount to that. It is going to be 20 basis points, 25 basis points. And we are going to guarantee the transactions in case there is a chargeback." Now what happens when you start using Riskified is you are going to be paying us less than your existing cost structure of managing fraud. We are so much more accurate that even while we are providing the merchants a discount, we still have, you know, kind of very positive ROI and margins on that account. We are also providing them much more value.
When we talk to our merchants, they sometimes say, "Look, I don't want a solution that's, you know, where I wake up, come into work after the weekend, and suddenly there was a model drift and there was an issue and I just lost $5 million. I want a responsible partner that's going to manage this part of the business so that I don't need to focus on it." That is the great thing about the guaranteed performance and results that we provide. We don't just guarantee the cost side of the equation. We also guarantee the approval rate. Okay, because one of the easiest things—sorry, let me come back to—one of the easiest ways to reduce the fraud is just to reduce the approval rates, right? Then it's no problem.
Obviously, all these merchants proactively renew and continue to use Riskified because we actually help them increase their top line as well. The dynamic is you have the cost, and we said that's 30 basis points. For that 30 basis points, merchants are approving, let's say, 90% of incoming transactions. They proactively turn away 10% of volume, again, average made-up numbers, because they fear that might be fraudulent. When Riskified also guarantees, you know, the price, the cost, we also guarantee a higher approval rate. It can vary between, you know, how much savings versus how much incremental lift, and it varies by merchant. It varies based on their margin profile. When we IPOed, we looked at the top 10 accounts, and we saw that on average, they were experiencing 30% reduction in cost and I think high single-digit increase in overall approval rates.
Very meaningful ROI. Obviously, we operate in the e-commerce market, and while we're proud of the $140 billion in GMV that we've already done, it's still tiny compared to the e-commerce $6 trillion kind of TAM. I would say probably about three years ago is when we started expanding from a single product into more of our platform. Really, the order approval and fraud review is where we started, really looking at transactions that come to the merchant after they've passed through the entire payment funnel. For those of you who are, you know, kind of versed in payments, that means that there's been an auth request and now the merchant has the ability to capture them. We were talking to our merchants, and they said, "You know what?
That's great, but actually we think you can influence the bank payment authorization rate as well. We started collaborating with them as well. Now we have something called Adaptive Checkout, whereas we can send enriched data to participating issuing banks and card issuers. We do this pre-auth. Using that data, the issuers are able to provide a higher auth rate for our customers because they have more enriched data that we share with them. We are able to screen fraudulent transactions before they go into the payment stream. That helps our merchants have a better and more sophisticated mid. Merchants can enjoy a better end-to-end conversion, right? That's number one. Number two is our Policy Protect product. This is our fastest growing additional product. We shared 190% year-over-year growth in our new products Q1 to Q1.
Most of that is coming from Policy Protect. While it's still a small base, it is growing relatively rapidly. New products grew at, you know, kind of mid to low single-digit millions in 2024, and we're anticipating them to be low double-digit millions in 2025. Most of it is from policy. What policy does is that we've seen that, you know, I don't need to steal, again, let's use Chett's credit card. I can just call the retailer and say, "Well, I never received my package," or "I received the wrong color or the wrong size. Give me a refund. Send me a new pair." Nine times out of ten, a retailer says, "You know, fine, of course, you know, packages get lost. I want to create an amazing customer experience." There's a ton of fraud there. There's a ton of fraud.
We found that leveraging the same engine and data, we can run our models to find that type of policy abuse. For some of our merchants now, we're analyzing when someone initiates a refund and return request, we're analyzing that request and blocking or allowing, right? We've seen instances, in most instances, we've been able to block upwards of 10% of refund requests without incremental false positives as the merchant measures it. Usually, it's callback or complaints or subsequent chargebacks. That's a massive increase in revenue for the client, right? They used to just pay out 10% of these refund requests, which are actually fraudulent. That's one use case of the policy product. Other use cases could be around item limits, resellers, item launches.
Item launches are big, whether it's a sneaker merchant that wants to make sure that a bot is not purchasing all the inventory, or it can be a concert where you want to make sure that, you know, you have good distribution about who you're selling into the policy. Product allows merchants to create rules and logics to manage their business. They can also do positive differentiations here. It's not just about negative and finding fraud. Some of our clients are using the policy product to look at who the best customer is and not just, you know, for the abusive customers, not provide them a refund, but for the best customers, provide them an instant refund. Don't wait until the package comes back to the fulfillment sister and then issue a refund. Issue a refund at that point when they click refund.
It's a much better experience for your best customers. Some of our customers are leveraging it to provide free shipping and returns for best customers. Other customers might have a restocking fee. It's a way to differentiate your customers, leveraging our network and machine learning and provide a different experience while stopping abuse and fraud. Policy is, again, fastest growing from our newer products and our platform. Chargeback management, once a chargeback comes in, there's an entire process of disputing the chargeback with the bank. Our chargeback management software is a workflow tool that allows fraud teams to manage all chargebacks, fraud and non-fraud with the bank. It can be set up either completely automatically. It can be semi-manual.
A nice story is that, you know, as part of our management team meetings, we sometimes bring in customers to talk about, you know, their experience with using Riskified. And we recently had StockX, which is a long-standing Riskified customer that's now on the entire platform. So we asked, you know, from an ROI perspective, what product from the stack provides the most value? And he said, "Well, you know, Chargeback Guarantee, the fraud review piece." Obviously, that's kind of the biggest component and the most ROI. But for me and my team, the part that we use on a day-to-day basis and love the most is the Chargeback Management. So we really love to hear that. And we think that as we become more and more of a merchant's workflow and not just part of a decisioning engine, that is very powerful just from a usage and retention perspective.
We think that's really great. To end things on the product platform, account creation. Really, ATOs are an increasing fraud vector. This looks at when you're creating an account, logging in, and it helps, you know, kind of block bad actors from accessing that information and leveraging that to create additional transactions. That's the product platform today. Seeing a lot of traction. We discussed the 190% kind of year-over-year growth. A lot of the, on the Q1 call, we mentioned kind of that we're at a record pipeline and we're seeing increasing win rates. A lot of some of that, or a lot of that is attributable to this kind of product platform. It used to be that we would pitch merchants on the Chargeback Guarantee fraud prevention. They would say, "Well, you know, it's interesting, but I'm an e-commerce merchant.
I have 10 other priorities. I have, you know, resources to integrate three of them and you're number four. Now that we're able to solve more problems for the business, it helps facilitate additional conversations, additional ROI to get the integration resources. We're really excited about that. How do we do it? Why are we more accurate than merchants, than other systems? Number one is obviously the network effect, the flywheel effect. The more data, the more merchants, the more capabilities, the more products. It just helps us generate more and more ROI. Aside from that, I think there are a few specific and unique capabilities that I want to call out. Number one is the level of data capture that we have per transaction.
When you think about what Riskified sees, we see when you're browsing on a merchant's website, we have a beacon that collects that information and we see the products and your behavior on the website. We see how you're logging in, creating an account because of our Account Secure product. We receive the checkout information as you check out. We also receive all the information as you contact the customer support system and request an address change or anything like that. If you were to file a chargeback for fraud or non-fraud reason codes, we would receive that as well. If you initiate a refund or return request, we would receive that as well. Basically, the level of data capture is significantly higher than a payment network, a payment gateway, any one of the other players that we know in the ecosystem.
That deep data capture enables us to create a lot of very unique and compelling features. The fact that we focus on enterprises means we can also have very specific and targeted models. For some of the large clients that we serve, we would have custom models with custom features that are trained on their data. We have an autonomous training platform that helps us train and deploy models at scale. If you think about a mid-market provider that needs to have a generic electronics model that works for, you know, 100,000 merchants, we actually have the ability to tailor a model to a merchant-specific data and to do it automatically. That just helps provide a higher level of performance, especially with the data capture that we have. Obviously, enterprise scalability and everything that enterprises need.
Identity graphs are probably one of the newer capabilities that we've developed over the past two years. This has been especially helpful in the Policy Protect product. Our identity graph does a lot of clustering. It helps us understand that someone is saying, you know, emergency is a new customer initiating a return. Our clustering and identity technology actually says, well, this is a cluster that belongs to five different accounts across three different merchants, and it's the same identity. It's an unsupervised learning, machine learning piece, very unique and powerful. We think it's the core component of the Policy Protect product. That's been gaining a lot of interest from merchants as well. Obviously, everything is AI, or as we used to call it in 2013, machine learning. That was really the original thesis and genesis of the company.
You know, we continue to be focused on innovating there. A big part of the team in Tel Aviv is around data science and analytics, and that's the core culture. This is just a nice visualization of kind of the unsupervised learning that helps us understand the graph. You can see kind of the various accounts, but how we tie them into a single identity. Obviously from a merchant perspective, both because of the network capabilities that we have, but also building this level of graph is somewhat challenging. It's not something that's accessible to them. Here you can see some of the outputs of our autonomous training, right?
We just understood that at some point, as you train and create new models that are customized to newer fraud trends, to new features, to new events in a merchant's lifecycle, you end up getting good results. It used to take, you know, just a few weeks of actual data science time to train and validate and create and deploy these models. Our new kind of autonomous training hub allows us to do this at mass scale. It has been helpful in driving performance as well. Great thing about the newer products, the platform, the things that we have in mind coming up as well is that they all sit under the same data structure and merchant network.
If we think about the cost to build the policy product or the dispute product, obviously meaningfully lower than what it cost us to build the Chargeback Guarantee and great network effect capability. As we think about how we envision expanding the platform, we say, hey, we have a deep integrated relationship with, you know, 50 publicly traded companies and that number is increasing. What other problems can we solve for them leveraging our machine learning capabilities and the data network that we've already developed? That's the type of products that we continue to develop. Here you can just see some of the results, like the audited S1 material that we shared was 30% reduction in cost, 7% increase in approval rates on average for the top clients. The ROI, very meaningful.
Yeah, I would say the journey that we've had post-IPO, we've really focused on improving profitability, have actually reduced OpEx to, I think this year it's flat on the midpoint, but prior years we've been able to reduce OpEx. Basically flowing through 100% of incremental kind of gross profit dollars to the bottom line, showing the scale, the leverage, the automation in the business have continued to, we have a large cash balance, have been very active in buybacks and, you know, anticipate to continue strong activity at similar levels. At the same time, showing great growth and diversity across regions. I think you can see that historically we've been, you know, kind of more successful in some discretionary categories like fashion and luxury. Over the past few years, we felt that, you know, that's been hurting us from a same-store sales perspective.
Really focused on diversifying the business and going after non-discretionary categories. We've been able to call out good success in both the Remains categories, the food delivery business. We think we're building a more diversified and, you know, base that can withstand various economic conditions. Also seeing obviously a lot of growth outside of the U.S. I'll call out, you know, kind of APAC and Latin America as big anticipated growth drivers for us. If it used to be that a few years ago, the fashion luxury business was 45% of the business, it's now a third or even less than a third. Tickets and travel is right now the largest category. We continue to anticipate strong growth in newer categories. Here you can just see some of the adjusted EBITDA improvements that we've achieved.
I would say that around the time of the IPO, it's also when we decided to go into an investment cycle around some of the global go-to-market and the product platform. You can see that increase in spend, but also kind of the improvements made there since. Seasonally, Q4 is just seasonally a much stronger quarter because of returning buyer activity. That's why you see some of that. Very strong cash flow, you know, generated at guide over $30 million this year, slightly more last year. Anticipating continued strong momentum there. One thing I want to call out that's important, you know, we talked about the fact that we, the Chargeback Guarantee model basically means that the biggest component of our cost of sales are the chargebacks, the amount that we pay back because of the mistakes that we made.
That's the biggest impact in our gross margin. That's why it's hovering in the 50+% range and not kind of more traditional SaaS-like. The gifts and takes there is that we see continuous improvements because of the machine learning, because of the model, because of everything in cohorts over time. You can see that basically in virtually all cohorts, pretty consistently they improve. That's offset by the addition of newer cohorts, new clients, new business, new geographies that tend to start at a higher CTB rate. Okay, that's how we would anticipate the business continue to behave on the Chargeback Guarantee side. Core machine learning improvements, improving cohorts offset by the addition of new business and newer cohorts coming in at slightly higher.
To make it simple, if we go into Brazil and we have a Rimmens merchant there for the first time, we would expect to start at a high CTB, but to improve it over time. We think that's, you know, kind of worthwhile because long term it's a great geography and industry for us to be in. We consistently become more accurate as our system evolves. That accuracy, we can take it into two different directions. Number one is we can, you know, for the same approval rate, we can improve our margins, receive less chargebacks because you're more accurate. The second thing you can do is you can keep the chargebacks where they are and increase approval rates and provide more value to the merchant. That's usually some of the kind of give and takes that we have.
We need to make business decisions. Are we going to improve our margins right now or are we going to provide more value to the merchant? It is really kind of a business decision. That is it. As we think about the opportunities for growth, obviously there are same-store sales increases in growth with our merchants. We take a usage-based model. We have had a lot of success. Almost all of the growth has come from winning new merchants over the past few years. We have a lot of expansion opportunities within our existing clients. I think we have shared that for the $140 billion of GMV that we have done in 2024, we actually have another $350 billion in white space GMV available to upsell. We have shared that over the past two years from the top 30 clients, 80% have had at least one upsell.
It's kind of a recurring process that we go through. You know, started building a sales team in areas like Brazil and Latin America and Japan about two years ago. In most cases, these are starting to ramp and provide kind of more pipeline and revenue. That's on the geographic expansion, continuously targeting new categories. Example is Rimins, but we still have kind of a long list of things to focus on. The platform sale has also been helping drive growth. That's what we're focused on right now. All right. Thank you for your time. And Cris, I think I'll, yeah.
Thank you for that. We will take questions here. Don't be shy.
When you think about some of the newer verticals or geographies that you've entered into, can you just talk about kind of how the CTB kind of differs, you know, initially versus and how you manage that?
Sure. It does start higher, and we would anticipate that there are potentially more nuanced views on fraud within a new category, or there could be slightly different fraud patterns within a new geography that are unique. That is why we tend to anticipate a higher CTB there. Even when we think about like our CTB budget for the year, we know that our existing clients and what improvements we'll see. We usually peg like the new business to come in at a higher rate than that, and those are some of the reasons.
It would take a process of a few months, which we tend to try and drive that, you know, kind of quicker as much as we can, where we see the improvements. I think just the CTB charts that we shared and are also available on the supplemental information that we have on our website are the best indication of where it starts and how quickly it trends towards a different place.
Great. Thank you for that. Just talk about the margin profile of some of your newer products and how that compares to.
That is more traditional SaaS-like margins. They do not have the chargeback component. It is usually priced on a recurring revenue base, with a fee being based on the merchant tier or size. That can definitely be long-term, you know, kind of margin accretive.
Yeah, correct.
I wouldn't, criminals, people more likely to submit a, to file a chargeback on specific types of transactions, yes. Again, I wouldn't say trade, but these characteristics of a transaction are, well, no, and what constitutes a good transaction in domestic U.S. transaction for a low dollar amount groceries is not the same profile as what constitutes a good transaction on a global, you know, $10,000 luxury handbag. That could create different model scores and different perspectives. Right. That's why also it's important to train data on an individual and merchant-specific way, because a lot of times you're looking at variations from things that are considered normal within a specific account. Unfortunately, there's always going to be false positives in our system and any type of system.
Usually you get to a level where, you know, if you get to a ratio where, let's say, one out of three transactions in this band of approval rates, let's say you get to 99% approval rate. And above 99, you know that your recall rate is going to be like 30%. So you're going to have a false positive ratio of two to three. It basically means that you're incurring a very meaningful loss for each subsequent transaction in order to get to like one good transaction. You always try to minimize that, right? You also like to do use that as a training set to better improve your models, but there's always going to be some false positives for sure. Great. With that, we're going to have to wrap it up there. We do have the breakout session. All right. Thanks, Cris. Thanks, Aaron.
Appreciate it. This presentation has now finished. Please check back shortly for the archive.