Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintechs Use AI to Detect Fake Reviews and Social Proof Fraud
Transcript
- Lucas: So earlier this week I was reading a report from the Federal Trade Commission, and one stat jumped out: they estimate that fake reviews cost consumers something like $152 billion globally every year. That's not just lost money on bad products — that's the cost of being misled into buying something you otherwise wouldn't. Luna: And $152 billion — that's the kind of number that makes you wonder: how much of that is hitting fintech products specifically? Like, are people signing up for neobank accounts or lending apps based on fake five-star ratings? Lucas: Exactly. And that's the angle I want to dig into today. A lot of fintechs live and die by their app store ratings and social proof. A new digital bank might have a 4.8 rating on the App Store — but how many of those reviews are real? The same technology that makes it easy to scale a fintech also makes it easy to scale fake endorsements. Luna: So there's a whole sub-industry of review fraud detection now. Companies like Yelp have been doing it for years, but newer startups are using AI to catch patterns that humans would miss. Lucas: Right. Yelp's algorithm has gotten pretty sophisticated. They look at a reviewer's history — if someone posts five five-star reviews for five different restaurants in an hour, the system flags it. But fintech reviews are trickier because the purchase cycle is longer. You don't review a credit card the way you review a taco. Luna: So what's the state of the art? I know there's a company called Fakespot that analyzes the language of reviews using natural language processing. They look for patterns like overly generic wording, repetitive phrases, or too much emphasis on a single feature. Lucas: Fakespot is a good example. They essentially build a linguistic fingerprint of a fake review. For instance, real reviews tend to have a mix of positive and negative observations — even if the overall rating is five stars, someone might mention 'the app crashed once but it's fine.' Fake reviews are almost always uniformly glowing. They also tend to use more first-person singular pronouns — 'I' and 'me' — because the reviewer is trying to sound authentic. Luna: That's interesting. So the AI is basically looking for over-authenticity. Like, the fakers are trying too hard. Lucas: Exactly. And then there's the social graph approach. Some detection tools look at the network of reviewers. If a batch of accounts all joined on the same day, all reviewed the same five fintech apps, and all gave five stars, that's a strong signal of a review ring. Amazon has been fighting this for years, but fintech is catching up. Luna: I read about a case where a fintech company was using a service called AppWiz to generate fake reviews. The detection tool caught it because the IP addresses all came from the same cloud hosting provider. That's pretty basic, but it works. Lucas: Right. And the arms race continues. The fraudsters are now using real human reviewers — paying people in low-cost countries to post genuine-sounding reviews. That's harder to detect because the language is natural and the IPs are diverse. So the AI has to look at subtler signals, like the timing of reviews relative to product releases, or the fact that the reviewer's entire history is only reviews — no other activity. Luna: Let's talk about the FTC's role here. They updated their guidelines on testimonials in 2023, and now they're going after companies that buy fake reviews. There was a big case against a company called Cure Encapsulations that had to pay more than $12 million in penalties. Lucas: That case was a warning shot. The FTC is saying: if you're a fintech and you knowingly use fake reviews, you're on the hook. But the problem is enforcement capacity. There are thousands of fintechs and millions of reviews. The FTC can't police it all, which is why the platforms themselves have to build detection. Luna: So what does this mean for a small fintech startup? They don't have the resources of Yelp or Amazon to build an in-house AI detection system. Lucas: They can use third-party tools. There are companies like ReviewMeta and Fakespot that offer APIs. You can plug in your review feed and get a risk score for each review. Some are even starting to use multimodal AI — looking at the text and the image of the product together. For a fintech app, that might mean analyzing screenshots of the app's interface that accompany a review to see if they're doctored. Luna: I want to loop back to the consumer angle. As a listener, how do I know if the reviews I'm reading for a budgeting app are real? Is there a rule of thumb? Lucas: Look for reviews that mention specific features with nuance. A real review might say 'the YNAB app helped me track my spending on groceries, but the sync with my bank is sometimes slow.' A fake review will say 'Best app ever! Changed my life!' Also, check the reviewer's other reviews. If they've only ever reviewed one product, that's a red flag. Luna: And on the flip side, if all the negative reviews are about the same issue — like a bug that was fixed months ago — that could be a coordinated attack from competitors. So it's not just fake positive reviews; fake negatives are a thing too. Lucas: Absolutely. Review fraud goes both ways. And some fintechs have been victims of smear campaigns. I remember a case where a peer to peer lending platform was hit with a wave of one-star reviews all mentioning 'poor customer service' but the timestamps were all within 24 hours. The platform used AI to flag them and got them removed. Luna: So the tools exist, but they're not perfect. The false positive rate — legitimate reviews being flagged as fake — is still a problem. Yelp has been criticized for that. Lucas: It's a trade-off. If you're too aggressive, you suppress real voices. If you're too lenient, you let fraud through. The best systems combine multiple signals: linguistic, behavioral, and network-based. And they update their models constantly because the fraudsters adapt. Luna: Let me ask a bigger question: how much does this actually matter for fintech adoption? If a fintech has fake reviews, does it really move the needle on user acquisition? Lucas: Studies show it does. One academic paper found that a one-star increase in a product's average rating on the App Store can lead to a 25% increase in downloads. So the incentive to game the system is huge. For a fintech competing in a crowded space like budgeting or neobanking, those stars can be worth millions in acquisition cost savings. Luna: Which brings us back to the arms race. As detection gets better, the fraudsters get more sophisticated. I saw a report that some fake review farms are now using AI to generate the reviews themselves — so you have AI writing fake reviews, and AI detecting them. Lucas: It's literally algorithms fighting algorithms. And the cost of generating a fake review is dropping fast. With large language models, you can produce thousands of unique, human-sounding reviews for pennies. That means detection has to get more advanced — looking at metadata like device fingerprints, GPS spoofing, even keystroke dynamics. Luna: One thing I find encouraging is that some fintechs are turning this into a competitive advantage. They market themselves as 'review-verified' or 'fraud-free reviews' to build trust. It's a differentiator. Lucas: Yeah, I've seen that. There's a fintech called Trustpilot that started as a review platform but now offers a verification badge. And some neobanks like Chime and Varo have started displaying a 'verified purchase' tag on reviews. It's not foolproof, but it helps. Luna: Speaking of trust, I think this is a good moment to mention something about how this podcast stays on the air. We don't run ads, and that's intentional — it means we can talk about specific companies and cases without any conflict of interest. Lucas: Right. And that's only possible because a small number of listeners support us directly. If you find these conversations useful for what you're building or running, you can chip in at buy me a coffee dot com slash fexingo. It's a simple way to keep this independent. Luna: Yeah, and there's no pressure at all. But even a few contributions make a real difference in terms of the time we can put into research. So thank you to those who already do. Lucas: Back to the topic. I want to look ahead: what's the next frontier in fake review detection for fintech? Luna: I think it's going to be around video and voice reviews. As more platforms allow users to post video testimonials, fraudsters will start generating deepfake review videos. Detecting that will require entirely new forensic tools. Lucas: Exactly. And some startups are already working on it. They use things like facial micro-movement analysis and audio spectral analysis to check if a video is synthetic. But that's a few years out for widespread use. For now, the focus remains on text and basic metadata. Luna: So if I'm a fintech founder listening, what's the one action I should take today? Lucas: Audit your reviews. Use a third-party tool to run a baseline. You might be surprised how many of your own reviews are fake — either planted by overzealous marketers or by competitors. Then set up ongoing monitoring. It's not expensive, and the cost of reputational damage from a fake review scandal is far higher. Luna: And for consumers, the takeaway is: don't trust ratings blindly. Cross-reference with independent sources, look for verified purchase tags, and read a few actual reviews — especially the three-star ones, which tend to be the most balanced. Lucas: Yeah, the three-star review is often the truth teller. It's the sweet spot between hype and outrage. So next time you're choosing a fintech app, spend two minutes reading the mid-range reviews. It might save you from being part of that $152 billion statistic. Luna: And it might help you find a genuinely good product. Alright, that's a good place to end. Thanks for listening.