3 ms·

It sounds like you have a good handle on targeting, where you can use contextual information to show ads users are likely to be interested in, but what about fr

• ericholscher 4y ago
  Fraud detection is definitely something that is a struggle in the overall industry. We have a couple benefits that make it easier for us. First, is that we hand-approve publishers, so that we are able to tell if they are a legit project or application. We look at some social signals when they sign up to see if their traffic numbers seem realistic, and do some sanity checking. Secondly, we do capture anonymized IP's & user agents, and create a hash of the original values. This allows us to see trends across browser types and high-level locations, which helps with fraud detection. The other big thing is that we have feedback from our advertisers. We do pass a UTM code about what publisher traffic comes from when an ad is clicked. This allows us to figure out who is sending junk traffic to our advertisers. We also have a few other specific technical methods that we generally don't talk about, because we don't want to give folks guidelines. But those are mostly just to stop obvious bad actors which aren't trying very hard :) Some of these approaches work because we're still relatively small, but I think they will scale to 10-100x our size, maybe just not to Google's size. But overall, our approach to fraud is currently working based on the number of repeat advertisers we have, and the success they are seeing with our network.
  • jefftk 4y ago
    Thanks! Over time this is likely to work less well, as it becomes more worthwhile for fraudsters to target your network. Off the top of my head, using a botnet to simulate traffic with a realistic range of UAs seems like it could allow a publisher to earn, say, 30% more revenue without you being too obvious. Overall, this would shift income from your honest publishers to your cheating publishers.
    • ericholscher 4y ago
      Definitely true. Hopefully the folks who have the skillset to build a large OSS project, and build a network of fake traffic, would find other ways to spend their time with a higher return on investment than a 30% increment of their revenue. But it will certainly happen, and hopefully we can continue to minimize it. We've generally seen fraud that looks like a site with no real traffic, and trying to generate 1000% returns, which is a much easier pattern to catch. If you have any suggestions for ways we might improve our fraud tracking, definitely happy to hear them at eric@ our domain.
      • flitzofolov 4y ago
        There are products designed to solve this problem: https://www.humansecurity.com/products/ad-tech-teams https://www.humansecurity.com/products/ad-tech-teams Disclosure - I used to work for WhiteOps (now known as Human Security apparently).
        • jefftk 4y ago
          There are definitely products for this problem; it's a huge business. But they're not products that are compatible with EthicalAds' privacy-first JS-optional model.
• jethkl 4y ago
  All parties in the ecosystem tolerate some levels of fraud. The threshold for action varies over time (e.g., if ${bot-detector-company} has a press release that blames the Russians, people will care for a few weeks) and it depends on who is complaining. Detection of low-hanging fruit at small scales is often straightforward, and you can detect the clumsiest fraudsters using basic analytics tools. But detection at scale is hard, especially if you have a large number of clients that are not a homogeneous or predictable. The nature of organic traffic visiting streaming platforms, ticket sales sites, news sites, blogs, etc. are all very different. From my experience, the hardest part of the fraud detection space is getting some action in response to a fraud alert. (Disclosure: I used to work in ad fraud detection)