Riskified Ltd. (RSKD) Earnings Call Transcript & Summary

Good afternoon. I'm Reggie Smith, I lead Fintech Research here at JPMorgan. And I am pleased and happy to sit and talk with Eido Gal, CEO and Co-Founder of Riskified. Welcome. How are you?

I'm great. How are you?

Good, good, good. So I figured we would start with kind of a quick elevated pitch on the company, a little bit of background about the company, what you guys do, how you do it, who you do it for and how you go after the market?

Yes. So we help enterprise e-commerce companies, manage their online fraud. When someone places a transaction online, we help these merchants verify that it really is the cardholder and not someone who stole that credit card is using a fraudulent card that's how we started, work with over 50 publicly traded companies. Wayfair, Prada, [ really across ] Booking.com across geographies, across industries. And probably over the past 3 years, we've expanded from the problem of just solving chargebacks or fraudulent credit cards to solving other forms of abuse that merchants are experiencing around their policy and helping them also optimize their dispute process, which is how they fight the charge back with the bank.

So thinking about fraud, like how much fraud is there in the system? I don't know if you can quantify it on a dollar basis. Yes, like how large is this market? And what impact does it have on merchants?

So look, the reason merchants even care about this, and that's a great point I should have highlighted it earlier is that merchants are liable if they accept the transactions stemming from a stolen credit, okay? So if I steal your credit card, purchase something at an online retailer, you see it on your statement, you call the bank. You say, "Hey, it wasn't me. Someone stole my card." You get refunded. The money is taken on the merchants accounts. So merchants are reliable. Now if as a merchant, you're not screening every single e-commerce transaction to look for fraud and you have a loophole somewhere in the system. Fraud will go from 0 to 100 very quickly, right? And you would easily go bankrupt because of that. So because of this dynamic, every single e-commerce merchant looks at all incoming transactions, screens them and tries to identify good transactions from bad transactions because they're liable. Now what type of mistakes the merchants have because of this dynamic, they accept bad transactions that they need to pay out, right, chargebacks. They turn away good transactions that might be risky characteristics or they've been burned by a similar transaction before. And this is a complex, expensive, difficult process for them to do. And Riskified was really when we started the company, we said, "Hey, we think we can do it better than every single individual merchant, we'll have a better network effect. We'll have a better machine learning solution," and that was the thesis.

Got it. You hit on a theme. It's a very popular machine learning and I guess AI. Maybe talk a little bit about how you guys use that to enhance your fraud protection or fraud identification?

No, you're right, and that was part of the thesis that we didn't have this type of machine learning process or AI process when we started the company. So really, when we started, one of the most important things we do is we do supervised machine learning, right? And supervised machine learning, you tell the machine, the model, you tell them, "Hey, this is a good transaction. This is a bad transaction." And now you need to train on the good instances and you need to train on the bad instances and now when uncategorized transaction comes in, you need to understand if it's more similar to the good ones or to the bad ones, meaning that the tagging is incredibly important, right? It's not like a large language model that just goes online and trace based some of that information, you really need to have that accurate tagging. So the first few years of Riskified, what we were doing is we were actually tagging transactions ourselves in a very manual process. There wasn't an automated system for it yet. We're just looking at the transaction and saying, "Hey, we think this is fraudulent. Here's how we know, here's how we're going to build the feature around it and create that tagging." Over time, we've created a very proprietary unique database of taggings that we use to train our model that's only available to us. And we've slowly ramped up the automation back then based on this tagging. So it created a very, very accurate system. Something else that we do that's very unique is something called the chargeback guarantee. We understood from our merchants that they said, "Well, listen, if you're going to recommend that I accept the transaction, but it turns out that it was fraudulent. Now, I just lost a few million dollars. You're not really helping, okay?" So we came up with a chargeback guarantee, meaning that we told our merchants, look, if we make a mistake, and we said, "Hey, this is a good transaction, but it turned out to be wrong, we'll pay you back for that, right? We'll guarantee the performance of the system." And together with the approval of a guarantee -- the chargeback guarantee, we also provide an approval rate guarantee, right? Because that's the other thing merchants care about. They're like, okay, fine, you're blocking fraud, but I need to maximize acceptance rates. So those 2 things, when we were building the system, right? On the one hand, you have a lot of competing pressures. You have to be able to approve something because that's how you have the approval rate guarantee, and that's how you're paid. We're paid for an approved guaranteed transactions. But we have to be so confident in the transaction because we have the chargeback guarantee and every time we make a mistake and the chargeback comes in, it feeds straight into the system, and it's another great form of learning, which has also helped us kind of evolve the system over time.

There's 2 things I wanted to dig into there. So from the merchant perspective, working with you, like how does that, I guess, impact them financially? Obviously, they don't have the chargeback risks, but like are there other benefits to the merchant, like how does it manifest in the P&L for the merchant?

So when we come to the merchant, the standard sales process, as we help them kind of calculate their total cost of fraud. And what's the total cost of fraud? It's how much you're paying in chargebacks, how much you're paying in staffing and how much you're paying in tools, right? The staffing and the tools, you have a staff that's managing this process, that's using a set of tools. Usually, chargebacks would be 70% to 80% of that cost, staffing would be kind of 10% to 15% and the remaining would be the tools that they use. And then the merchant would calculate and said, you know what, my cost of fraud is 30 basis points. And then for that 30 basis points, my approval rate is 90%. I accept 90% of incoming transactions, okay? This is not to be confused with the bank acceptance rate, payment acceptance rate? These are transactions that have gone through the entire funnel and now the merchant is able to capture them, but they need to proactively decide, do I want this transaction? Or is it going to result in a chargeback? Okay. So 30 basis points, 90% approval rate and when you interact with Riskified, we would probably offer you either cost savings for a similar approval, right? Or an increase in approval rate for a similar fee or some combination of both. And merchants really depending on what type of merchant they are, what type of margin profile they have, that's how they would make that decision, right? If you're luxury fashion merchant, and you have 50-plus percent margin on these goods, you're going to say, you know what, I wouldn't even mind paying you 35, 40 basis points, a higher fee. If you feel you can get my approval rate up to 96%, right? I can't identify anything above 90% to me, it's kind of random. But if you're better at that, by all means. Other merchants, for example, maybe an OTA, where they have extremely low margins in some categories, they just want to minimize the fee that they're paying.

Right. That makes sense. Last kind of technical question for me. Is there a way to kind of quantify how better your systems screens have gotten over time? Like how do you think about and measure and yes...

So there's a few ways, right? The system becomes more accurate and that accuracy manifests either in better performance or margins for Riskified, right? Because our chargebacks go down for the same fee and approval rate for the merchant. Another way to think about it is the incremental approval rate that we're providing our merchants increases over time, right? And it's usually we're seeing some combination of both. So if you look at some of our filings, we released a report that shows chargeback cohorts over time. And pretty consistently, I think with every single cohort, you can see that whether they start at a slightly higher rate, or lower rate, they moved down significantly over time. And the reason being, this is machine learning, this is big data. We -- our system becomes more accurate, and we train custom models for that merchant-specific performance. So we definitely see that. And then when you think about the approval rate that our merchants enjoy, I would say, I want to be careful here, but probably the vast majority, 90-plus -- 95%, 98% of the cases, they're seeing higher incremental approval rates than they did pre-Riskified, right? So it's pretty amazing. We're able to drive higher approval rates for a lower cost. Within that, we also have pretty positive margins, right, like 50-plus percent improvement over the past few quarters. So that's how much more accurate the system is.

You just jog my memory. Let's talk about the most recent quarter and recent trends. I know you guys have turned the corner on profitability from an EBITDA perspective. I know you're very proud of that. Maybe talk about real quickly the most recent quarter? And what you guys articulated on the -- profile and guidance longer term?

Yes. So it was a great quarter. It was a culmination of, let's call it, a multiyear effort to shore up the expense base, improve the margin profile and all while kind of obviously going after the large opportunity of growth. I think what happened post-IPO or right around the IPO time frame is we did a pretty significant investment in both the global go-to-market team and kind of the product platform. And we said, hey, we think this is a necessary investment that will kind of taper off after 12 to 18 months, but it's going to get paid dividends in the years ahead. And I think we're starting to see that, right? So we saw that the machine learning platform has significantly improved. I think Q1 was 56% gross margin, predominantly by improved machine learning and chargeback to billing ratios. That's up from 53% the prior year. And we've had a similar step up the previous quarter in Q4. It's clearly articulating expanding gross margins. When we think about the revenue contribution from some of those newer geographies that we built that enterprise go-to-market motion for Latin America, APAC, really strong growth over the past few quarters. This quarter was the first quarter where we had stand-alone sales for our dispute and policy products. Those are the newer products that help you outside of this traditional chargeback model. So that's helping us from a continuous selling motion, helping drive platform revenue. So all those things within the storyline have really been clicking and we feel good about.

So we talked about risk, I guess, chargeback risk. You just mentioned that you're selling some of the other services as a stand-alone product. Maybe talk a little bit about what those products are? And is kind of shocking that you're able to get people to buy that without even the chargeback. What's going on there?

So a few years ago, we were talking to our clients and they said, "Look, this is fine, now you've solved this problem of fraud chargebacks for me, but a few things: A, I have a lot of flows coming in which might be from fraud, but it's not even related to chargebacks, right?" So just imagine this, instead of stealing your credit card, okay? I just need to call the merchant and tell them I never received my package or I received the wrong size. And they're going to send me new pair, they're going to refund me, especially, if I'm a first-time customer and they have very customer-friendly policies. So fraudsters are on to this and as you block their ability to use stolen credit cards, they're finding ways to abuse these different merchant policies and request refunds and returns in a way that's clearly abusive, right? We're not just talking about there's a spectrum. It could either be fraud, it could be someone who's ordering 20 T-shirts and only keeping 1. That's not fraud. That's a merchant decision if you want that type of customer. What our technology is great at is really identifying the great customers from the bad customers and also providing good insights into the LTV if the ones in the middle and what type of customers they are and providing the merchant, the tools to leverage this information to either block the customer to say, "Hey, you're not going to get free shipping. You have a restocking fee" or maybe someone else. You know what, you get -- the minute you click return, you get an instantaneous refund and you don't even need to wait for the item to come back to my warehouse system, right? So it's helping merchants provide a different experience based on the customer profile and merchants have really leaned into that. So that's the policy product. The dispute product, it's just chargebacks, especially non-fraud chargebacks. There's an entire process where you need to -- to represent those with the bank and you need to file what's called a compelling evidence. It's a lot of information. It's a very manual process that they were doing, and we've created a very beautiful fun AI-based process to automate that.

Sounds good. Yes, I figured it was AI to get that done. Maybe zooming out, let's talk about the broader opportunity and how you think about the addressable market? And I guess, your share today in that and where you think it can go over time?

So on the one hand, we're proud of the $120 billion in GMV we did -- we processed last year. On the other hand, when you think about e-commerce, and this is e-commerce, this includes travel, live events, delivery services, easily $6 trillion. We focus on the enterprise side of that, that's maybe $4 trillion. Massive, massive opportunity when you think about our current diversification, about 1/3 of the business is in travel tickets and live events, another 1/3 of the business is in fashion various forms anywhere from the highest end of luxury, the kind of what we call fast fashion. And another 1/3 of the business is within electronics, delivery services, remittance and payment companies. So relatively broad-based and diversified, but we think there's both expansion in each and every one of these categories, but also going after kind of some of these newer categories where we are non as traditionally strong.

When you think about -- you talked enterprise, and I think you said $4 trillion in sales. Is there a way to frame like the number of logos that is? And maybe how penetrated you are within that base?

I tend to think about it in GMV terms. I guess that in logo terms, you probably have the larger ones have an outsized share of that like global GMV but it's probably the top 3,000 that are driving the vast majority of that volume.

Got it. Okay. So you get 3,000. And then you talked about the change in your kind of go-to-market motion. What is that exactly? How has it changed? And what are you seeing from that change? Obviously, the stand-alone sales of policy protect and dispute are probably evidence of that. But what else can you point to there?

Well, what's interesting in the sales motion is that we ended up not doing a specialized sales approach, right? So we don't have a specialized like policy sales team. But we've trained the existing sales team to do more of a platform sale. And again, they say instead of saying, "Hey, fraud chargebacks, this is where you help you." We can put a wider ROI and holistic story, right? Overall chargebacks and only 50% of that is related to fraud. You're probably 10% to 20% return rate or refund rate. We can probably block 10% to 15% of that, that's what we've seen. So those are massive, massive numbers. You probably have a few dozen people on kind of the charge-back dispute team. So the value-based selling is much wider right now. And what's really great, and we've seen this quarter with the stand-alone sales of some merchants have said, you know what, I actually am committed to another vendor for the next few quarters or maybe even I don't feel like I have a chargeback or fraud issue right now, but policy? That's massive. That's my biggest problem right now. Let's go. Let's -- let's integrate and that's a great cross-sell opportunity for us as we build this trusted relationship with the merchant.

How -- thinking about policy and dispute. Like how are those services priced?

Our core chargeback product is priced as a percent of the merchant's GMV, right? Basis points. And our policy product and also the dispute, it's more of a SaaS-like pricing model, but the tiers are based on merchant volume. So larger merchants would pay more, smaller merchants, less, but it's a monthly subscription fee with overages.

Is there a way -- so you've got a nice sizable business today. Do you have a sense of the cross-sell or upsell opportunity within just your existing merchant base?

Yes. So the way we define it is and upsell is selling more charge by guarantee volume to someone who's on the chargeback guarantee platform and a cross-sell with some of those newer products. Today, within our existing installed base, we think we have in the zone of $300 billion plus in upsell opportunity, right? So this is volume that's related to clients integrated to Riskified that we're not guaranteeing right now. And over time, we have consistently seen now we're able to upsell and capture more and more volume. So that's #1. #2, on the cross-sell part, the numbers we've been seeing recently are policy deals have been priced at 10% to 20% of chargeback deals and dispute deals have been priced in the 5% range.