Latest / Marketplace Businesses with Fexingo: Two-Sided Networks, Liquidity, and Take Rates / How Marketplaces Use Fraud Detection to Protect Liquidity
Transcript
- Lucas: So we've talked a lot about how marketplaces build trust — reviews, insurance, escrow, cancellation policies. But there is this one layer that sits underneath all of that, and it's probably the least visible to users, yet if it fails, the whole network can collapse. Luna: You're talking about fraud detection, right? The stuff that happens before a transaction even gets started. Lucas: Exactly. And it's not just about stopping credit card theft. For a marketplace, fraud is anything that breaks the trust between two strangers transacting. A fake listing, a bot account, a driver who doesn't show up, a buyer who scams a refund. If those happen too often, the good users leave. Lucas: And look, I want to pause here for just a second because these conversations — the ones where we really dig into how platforms work — they're only possible because of listeners who support the show directly. A couple of dollars a month is genuinely what keeps these going. If you've gotten something out of them, you can find us at buy me a coffee dot com slash fexingo. Luna: Yeah, it makes a real difference. Every little bit helps us stay ad-free and focused on the details that actually matter for people building things. Lucas: Alright, so back to fraud detection. Let me give you a concrete number. Airbnb says that before they implemented their current screening system, they were seeing something like one in every two hundred bookings result in some kind of trust and safety issue. That doesn't sound huge, but when you're doing millions of bookings a quarter, it's a disaster. Luna: Right, because even a small percentage of bad experiences creates a reputation problem that spreads fast. One viral story about a fake listing and suddenly everyone's nervous. Lucas: Exactly. So Airbnb built what they call their 'risk scoring' system. Every booking attempt gets a score based on dozens of signals — how new the account is, how the guest paid, the host's history, the location, the time of booking, even the device type. If the score crosses a threshold, the booking gets flagged for manual review or an extra verification step. Luna: And that's the trade-off, right? Every extra step adds friction. Too much friction and you kill conversion. Too little and you let fraudsters through. Lucas: That's the core tension. And it's why most marketplaces don't just rely on a single check. They layer them. So you might have identity verification at signup, then behavioral signals during the first few interactions, then transaction monitoring after payment. Each layer catches a different type of bad actor. Luna: I remember reading about Etsy's approach. They had a problem with sellers listing counterfeit goods or even just stealing photos from other sellers. So they built an image recognition system that flags listings that look too similar to known products. Lucas: Yeah, and that's a great example of a marketplace-specific fraud vector. On a platform like Uber, the fraud looks different — drivers using multiple accounts to game surge pricing, or riders claiming they never got a ride. So each marketplace has to tailor its detection to its own transaction dynamics. Lucas: One framework I like comes from a startup called Sift, which does fraud detection for marketplaces. They talk about 'the fraud triangle' — opportunity, rationalization, and pressure. If your platform makes it easy to commit fraud, if the fraudster can justify it to themselves, and if they feel some pressure to do it, they will. Your job is to break at least one of those legs. Luna: So how do you break the opportunity leg? That seems like the most direct one. Lucas: You make fraud harder by adding verification steps, but you have to be smart about where you add them. For example, Airbnb doesn't verify every guest with an ID. They only ask for ID when the risk score is high. That way, ninety percent of bookings go through frictionless, and the ten percent that get extra scrutiny are the ones most likely to be problematic. Luna: And that's all machine learning, right? The model continuously learns from outcomes — which flagged bookings actually turned into problems, which ones were false positives. Lucas: Yes, and the data loop is critical. Every time a human reviewer makes a decision, that becomes a training signal. Over time, the model gets better at distinguishing a genuine new user who's just nervous from a sophisticated fraudster. Lucas: But there's a trap here. If your fraud detection is too aggressive, you end up rejecting good users, especially in the early days when you don't have much data. That's why many marketplaces start with very light screening and then tighten it as they scale. Luna: That's the classic 'liquidity first' argument. You'd rather have some fraud than no transactions at all. Lucas: Exactly. And that's where the concept of 'fraud tolerance' comes in. As a marketplace, you have to decide what level of fraud is acceptable at each stage of your growth. In the beginning, maybe you accept a higher chargeback rate because you need volume. Later, you invest in more sophisticated detection to protect the brand. Luna: But that's a risky game. If you tolerate too much fraud early on, you might attract more fraudsters who then become hard to remove. There's a selection effect. Lucas: That's a really good point. And it's why some marketplaces choose to invest in fraud detection from day one, even if it slows growth. I think of a company like Vinted, the secondhand clothing marketplace. They built a very robust verification system early on because the cost of a counterfeit item on a fashion marketplace is huge — it destroys trust in the entire platform. Lucas: There's also a fascinating case from the world of freelancing platforms. Upwork had a problem with clients posting projects and then not paying. So they introduced escrow — the client deposits funds before work starts. That's a fraud detection mechanism that also builds trust. It's brilliant because it solves two problems at once. Luna: Escrow is essentially a financial fraud deterrent. It removes the opportunity for non-payment, which is one of the biggest fears on both sides. Lucas: Right. And the best fraud detection systems combine multiple signals. A friend of mine who runs a small marketplace for pet services told me they use a combination of social login, phone verification, and a one-time payment hold for new providers. They reduced their fraud rate by sixty percent without hurting conversion because the hold is only for the first booking. Luna: That's smart — using the hold as a temporary trust signal. Once you've completed one successful transaction, you're in the system and the friction goes away. Lucas: Exactly. And that's the key insight: fraud detection isn't just about stopping bad actors. It's about creating a system where good actors can signal their trustworthiness easily, and bad actors find it too costly to participate. Luna: I think that's the most actionable takeaway for marketplace operators. Don't think of fraud detection as a cost center. Think of it as a liquidity enabler. The more trustworthy your platform feels, the more transactions happen. Lucas: That's a perfect way to put it. And I'd add one more thing: whatever system you build, make sure you monitor false positives. Rejecting a legitimate user is often more damaging than letting one fraudster through, because that user tells their friends, and the negative word of mouth spreads. Luna: So the goal is a system that catches most fraud but almost never flags a good user. That's the holy grail. Lucas: It is. And the marketplaces that get closest to that grail are the ones that treat fraud detection as a core product feature, not just a compliance checkbox. They iterate on it, they A-B test it, they measure the impact on liquidity, conversion, and retention. Luna: Alright, I'm curious — what's one specific metric a marketplace should track to know if their fraud detection is working well? Lucas: I'd say the ratio of chargebacks to completed transactions is a good start. But also look at the false positive rate — the number of legitimate transactions that get blocked. And finally, the repeat usage rate of users who were flagged versus those who weren't. If flagged users come back less often, your system might be too aggressive. Luna: That's a really useful set of metrics. I hope operators listening take a look at those in their own data. Lucas: Me too. Fraud detection is one of those invisible layers that separates marketplaces that work from marketplaces that fail. It's worth getting right.