OfferUp's head of trust and safety explains how the marketplace built confidence at scale: manual curation early on, then machine learning to catch banned items, government ID verification badges, and partnerships with local police. Trust, they say, is never finished.
Chapters
Could you share a little bit more about how, you know, maybe some of the ways you thought about building trust in the early days and maybe some of the learnings from that?
You know, in the early days you got to build stuff and even though you don't have a lot of issues necessarily, but eventually you get scale, then you have issues. So I remember the first time someone robbed somebody on OfferUp, like we were going to get pitchforks and lanterns, we were going to go to their house. I'm like, okay, this is crazy. And so we started, we very quickly, we were You know, I felt pretty good, like we put enough in this space, and then we were on our heels because we started scaling so fast. And again, to be clear, like, it's very rare anything ever happens on OfferUp, but sometimes they do. I mean, if you look at the scale that we're at, like, you don't have any major city in the US without some kind of crime, no matter how much you police it. And so we've done a whole bunch of things. Number one, we started doing things at the time of posting. So we look at items and pictures, and we do some Image recognition. There's a lot of AI stuff we do and machine learning under the hood to figure out like, okay, what is getting posted? Like, what is in this photo? So you name it, I've seen it. Body parts, drugs, weapons, you name it, I've seen all of it. And, you know, eventually you just train models. So initially when we were scaling, we didn't have time, we had to go build the model. So luckily I had an old business partner of mine in India. I called him, I said, I need people. I just need you to manually curate stuff right now, 24/7, because we can't keep up. So in the early days, it was, we would literally, we would literally divide the team up per hour of the day. And we say, you guys have to curate. That's all we had to do until we had a solution. Then we started putting bodies offshore as we built out our own solution. So that was what we did. Then we started building out profiles and ratings. And then we ultimately also built out this program we call True You, which is you can scan your government ID. And step 2 is It will— you can up— it'll do some image recognition. And if it's a match, then you get this badge that says you're a TrueView member. So that was important. Our community also reports a lot. So there's rules and systems we do. Like, let's say I post an ad, let's say I post a Rolex watch for $400, clearly fake. If our system doesn't detect— it should detect that because we look at watches and we say, okay, if it's off by this price, remove. But if it misses it, then the community can report it. So let's say it gets like 3 reports within an hour, automatically gets pulled out, right? So these rules and these systems, you just build on them over time. You're constantly building on them. But in the beginning, in the early days, there's a lot of manual work with a lot of these. So anyone building marketplaces, don't— something wrong with that. Go ahead and just manually do stuff, but figure out like where you're spending the time and then start to automate. And then fix that and start to automate. Like you just have to You have to think about that as a long slog, and there's not like one size fits all. And they're always going to try to go around the process, always. And we have third parties now. Like, we're very sophisticated when it comes to, I'd say, trust and safety locally. We've partnered, we've trained 5,000 police officers around the country how to use OfferUp. We've rented auditoriums. I went down to police stations and met people. Which is awesome because we've always heard from them like, no one's ever done this before. Craigslist never came here. eBay never came here. You're the first company to actually spend time in doing this. So we do spend a lot of time on trust and we're never going to be perfect, but I think we're pretty good.
