Ad Fraud, Part 2: Dr. Augustine Fou on Hidden Fees, Phantom Outcomes, and the Agentic Al Trap

Watch on Youtube

Listen on Spotify

In Part 1, we talked about bots. In Part 2… we follow the money.

Brett and Rio sit down again with Dr. Augustine Fou, and this time the conversation goes deeper—and gets a lot more uncomfortable. Because the real issue isn’t just fraud. It’s the system that quietly profits from it.

We break down what actually happens between the moment a marketer places a bid and the moment a publisher gets paid. The gap is bigger than most people realize. In one example, a $1.50 CPM turns into just $0.15 for the publisher. In another, increasing your bid doesn’t get you better inventory—it just means someone in the middle keeps more of your money. And a lot of what gets reported as “performance” can be completely fabricated—clicks, conversions, even analytics data.

This episode is about that hidden layer most teams never see: take rates, pass-through opacity, spoofed supply, and the growing gap between what advertisers pay and what publishers actually receive.

Then we turn to what’s coming next: agentic AI.

Because if your inputs are broken, automation doesn’t fix it—it just scales the problem faster. The same systems that already struggle with transparency are now being handed more autonomy, more budget, and more control.

We get into why blended averages hide the truth, why the industry’s long-standing “1% fraud” narrative doesn’t hold up, and why so many optimization systems are quietly steering budgets toward cheaper, lower-quality inventory. We also unpack the difference between “attention” and actual human engagement—and why so much of what’s sold today is closer to performance theater than real outcomes.

By the end, this isn’t just a critique—it’s a playbook.

If you’re a marketer, this episode will change how you think about where your dollars are going, who’s extracting value in the supply chain, and what you can do—right now—to take back control.

Because the real question isn’t:“Is there fraud?”

It’s: “Who’s getting paid—and how much are they taking?”

---

Read full transcript bellow: Brett House (00:01.575)

Hey everybody. Welcome back to Signal and Noise. I'm Brett House joined by my co-host, Aria Longacre and our special guest, Dr. Augustine Fu. Dr. Fu has been on the podcast before. Welcome back. Always a thrilling conversation. For those that don't know Dr. Fu, he's one of the most persistent and provocative investigators of digital advertising fraud. Three decades, you've worked at the intersection of digital marketing, cybersecurity.

Rio (00:13.636)

He's back.

Brett House (00:30.275)

measurement and fraud detection. Everybody knows you for that. You do a lot of posting on LinkedIn, challenging assumptions, benchmarks and black box reporting the industry has long relied on. You're the founder of Foo Analytics, an analytics and media verification platform built out marketers and publishers independently inspect campaign quality and fraud and other things. And earlier in your career, for those that didn't listen to episode number one, I do encourage you to do that.

You were the chief digital officer at Omnicom's healthcare consultancy group, right? And you taught digital strategy at NYU and being in Jersey, Rutgers, that's great to hear. And you really built a reputation, I think. know, Rio and I certainly are well aware of it. We respond to a lot of your posts, whether it's on X or on LinkedIn. As one of the first people that was really willing to publicly say that the fraud problem was a lot larger.

Augustine Fou (01:06.766)

My first,

Brett House (01:23.953)

than what the ANA and other organizations were reporting as the classic 1 % rule. And you didn't just say it, you proved it and you continue to prove it again and again and again and you show it and sometimes you get in a little trouble. So we promise to keep you out of trouble, Dr. Poo. Welcome again to Signal of Noise. Thrilled to have you, thanks for joining.

Rio (01:35.074)

He brings receipts every time. Yeah.

Augustine Fou (01:47.448)

Thank you, Brett. Thank you, Rio. Glad to be here. We got some important topics to cover and things the industry may not be comfortable talking about a whole lot, but we're going to talk about it today.

Brett House (02:00.965)

Yeah. And did I cover your background well enough? I think for those that don't know yet. Yeah.

Augustine Fou (02:08.43)

Let me say a few words. So I've also spent time on the agency side at both Interpublic Group and then Omnicom. And I've been on the client side at American Express doing digital marketing for their small business cards division.

Brett House (02:24.497)

There we go.

Rio (02:25.786)

Awesome. Well, great context again, thrilled to have you back for those who did not check out the first episode. I don't know if it's required listening to in order to listen to this, but they're definitely related because the first episode we really covered what I would call the fundamentals. Like how much ad fraud is out there challenging some of these like industry beliefs or perceptions that there's not a lot out there. there, there is a lot. We talked about that. We talked about how it's even going to places that are maybe considered safe, like CTV, right? Where premium and what's quality inventory can sometimes be, fake or not quality or, or really AI generated slop. Right. We talk about that and how the incentives, a lot of times it's really an incentive problem that's driving a lot of these behaviors and decisions people make. this time we really wanted to go deeper and really look at the economics and the accountability question. Right. We wanted to look at if ad fraud is still endemic. And I think we all agree it is. And if the open web is where a lot of it hides in plain sight, then who's actually making money off this? Who's scraping the dollars out of the system? How much of that is being obscured by blending averages, by black box reporting, by just general lack of transparency? And then there was that recent audit drama, which we're going to talk about a little bit with the DSP that put a brighter spotlight on the question, like many marketers are.

I think still, Brett, mentioned this, like afraid to ask what happens? Where's the gap between what people are paying and how much is actually taken out of working media? I mean, I read a study recently that's around 39 % actually goes into media. The rest is siphoned off a scrape by different players, right? Some of it fraud, but a lot of it not. A lot of it's just take rates and different players putting their hand in the cookie jar, right?

We wanted to talk about this. I really think this matters because if we want the open web to succeed and we want it to be, we want it to stay alive and it's under a lot of pressure right now. If, as you say, Dr. Fu, this is where lot of the fraud, fake inventory, leak engine, hidden extractions remain most endemic and maybe even hardest to see. And then you add like these recent changes in agentic AI and how it's easier to create more content. It's easier to mimic human behavior better at scale.

Rio (04:44.122)

I think if anything, the stakes are going up. And if the source of truth is already compromised, automating this with agentic trading, I don't know.

Augustine Fou (04:53.942)

is going to just accelerate the take rate and the fraud. So you hit on many different topics, but let me try to break it down. I'm going to start with an example that's based on my concrete data. So a year ago, I ran an experiment for Ozone Group, which is a bunch of UK publishers that includes The Guardian, Daily Mail, BBC.co.uk.

Rio (04:58.49)

100%.

Augustine Fou (05:17.078)

And the main question I wanted to answer was exactly that, which is what portion of the dollars that I paid as an advertiser actually went to the publisher. So I basically played the role of an advertiser. asked them, tell me the CPM that's going to be above your floor CPM. So I make sure they clears and I can get the ads. And I executed the campaigns in DB 360, which is Google's platform. And I specified a one domain inclusion list for each of the publishers. So I'm going to use the BBC example.

So bbc.co.uk was the single domain I put in the inclusion list. I wanted to make sure, first of all, what percentage of my ads actually made it to the one place that I wanted them to go, and then ultimately what the publisher made. So I bid $1.50 CPMs. Can you guess what the publisher actually got?

Brett House (06:11.937)

Ten Cents.

Rio (06:12.762)

50 cent CPM.

Augustine Fou (06:15.854)

Okay, it's right in between those two. It's 15 cents.

Brett House (06:20.199)

Oh, there we go. I was a little closer. I think I won that. Right? Well, I did a little homework ahead of time, but 15 cents and this was the total CP was this incremental. See, this was just a buck 50.

Rio (06:20.477)

my god.

Augustine Fou (06:23.426)

Yeah, yeah. On a dollar 50. No.

Rio (06:27.354)

15th. Is that how much that wasn't a winning bid? was how much they actually got in the end, right? Okay.

Augustine Fou (06:35.39)

Yeah, exactly. So I paid a dollar fifty CPMs and they got fifteen cents of it. So ninety percent of it went somewhere else. And the reason it's that bad and nobody

Brett House (06:35.76)

Yeah.

Brett House (06:41.817)

And you verified and validated that. Yeah.

Rio (06:48.186)

Was that supply path? Was it issue or was it just endemic across? I guess you're gonna explain that now. I'm sorry, I cut you off.

Augustine Fou (06:53.474)

Yeah, it's a bunch of different issues, but the key issue is that the publishers never get to see what the advertisers bid on their inventory. And then the advertisers never get to see what the publishers got for their inventory. There's other issues which I'll get to in just one second. In this case, this was supposed to be the simplest possible scenario with the fewest supply paths. Right. Cause we all know that, you know, people are doing SPO supply path optimization, because when you have hundreds of thousands of different supply paths, obviously the bad guys can get in there and steal your money. Now, one of the key issues, um, so not only the publisher only got 10 % of what I paid, but there were, there was another issue, which is the supply path problem. So from the data that I got from DV360, I said, how many supply sources actually did I buy from because I didn't know ahead of time, right? This is all put out to pit. So it reported nine different supply sources. I went back to the publisher and asked, how many SSPs do you actually sell through? They said five. Okay, so what the heck are those other four that had no relationship with the publisher? Why the heck are they allowing the domain bbc.co.uk to be sold? So let me reemphasize, none of this is BBC's fault. This is where the bad guys...

Brett House (08:03.804)

Yeah.

Brett House (08:11.911)

Yeah. So they, so they have relationships with the DSPs, the five DSPs that BBC chose. Right. So they must have relationships with them. I mean, I'm sorry, SSP. Sorry. Yeah. Yeah. They're adding hops. Yeah.

Augustine Fou (08:20.064)

No, these are SSPs. again, yeah.

Rio (08:21.53)

SSPs. So they're just adding hops and adding the row of markups,

Augustine Fou (08:27.79)

So in this particular case, these SSPs, the four others that don't have any relationship with this publisher, were allowing bad guys to pretend to be bbc.co.uk. This is called domain spoofing, and that basically diverts money away from the publisher in a way they can't even see, because it never got to them. So sometimes the publishers are saying, OK, we get accused of high fraud. We look at our site.

Brett House (08:40.397)

Okay. Yep.

Augustine Fou (08:55.438)

there's no bots or anything. We're not buying traffic. What is going on? What they don't understand is that all of that high fraud, could be 100 % bot traffic is occurring on the fake sites that are pretending to be them. Okay. So let me, let me be clear on how, how this kind of grift happens. If the SSP doesn't actually catch the domain spoofing at the very first hop, it will become laundered by the second hop onward. Because in that very first hop, if the fake site,

Brett House (09:07.974)

Yep.

Augustine Fou (09:25.144)

pretend or says it's bbc.co.uk, they're still going to leave their own seller ID in there because the fake site wants to get paid. So they're just lying about the domain. If that discrepancy, which should be caught by ads.txt, if ads.txt were not enforced by that very first SSP, then by the second hop, some legitimate reseller could put their own seller ID in there. So then from that point forward,

Brett House (09:32.614)

Yeah.

Brett House (09:50.041)

Yeah, which is what you mean by laundered. gets obfuscated and it's suddenly.

Augustine Fou (09:53.006)

Yeah, you can't see it. So by the second hop onward, you won't be able to catch that discrepancy. So the problem is that a few of the SSPs are literally looking the other way and bad guys know exactly which SSPs to onboard their fake bid request. Once it gets on, it starts distributing through the ecosystem because there's so much reselling going on, then you lose track of where it came from. So to kind of bring this back to the BBC story.

All they had to do was go to those four SSPs that they had no relationship with and said just block our domain because no one from your SSP is allowed to sell that because we don't have a business relationship. Blocking that domain, bad guys can't pretend to be them. So we solved that.

Brett House (10:31.356)

Yeah.

Rio (10:36.538)

But did the publisher, like the publisher wouldn't even know these SSPs are working with them at all. They might not even know that, right?

Augustine Fou (10:41.922)

because it's not even in the reporting that they would see. So that's what I'm saying.

Brett House (10:44.549)

Yeah. And because, and because of the spoofing, the SSPs that they have relationships with have no contractual relationship with these additional SSPs. That's, that's yeah.

Augustine Fou (10:52.706)

Correct. Yeah, they probably wouldn't. But again, because everyone is interoperable with everyone else because they just want the volume. That's where this problem comes in. So nobody has run the experiments that I run in the past. You actually have to work with the publisher, see what the floor CPM is, specify one domain inclusion list. These are highly specific experiments where we actually got the data. And this wasn't a BBC problem. Every single publisher has this issue.

Brett House (10:58.63)

Yeah.

Rio (11:20.282)

Dr. Vood, no one has one domain inclusion list. So what you try to keep it as clean as you possibly could and you still, you know, the publisher still lost like 90 plus percent.

Augustine Fou (11:26.862)

Yes, yes, because

Brett House (11:29.551)

Yeah. Well, and imagine the compounding complexity when you start, you know, doing a real buy, right?

Augustine Fou (11:34.124)

Yeah, exactly. Right. Nobody does a one domain inclusion list and nobody cuts down all their SSPs down to one supply source. So basically I've told my advertiser clients the way to simulate a direct buy is to specify one SSP, right? Just talk to the publisher, which SSP do you prefer to sell through to specify one and specify one domain or one seller ID. So that way you can simulate a direct buy while using programmatic pipes.

So that way other people can't come in and just pretend to be that publisher's domain. That's the main problem. And let me tie it to your second point Rio, is where the agentic stuff comes in, right? So for 15 years, I've seen bad guys just use WordPress templates and stolen content to create fake websites by the tens of thousands. Now with agentic, they don't even need to use a WordPress template, right? The agent will just make the website for them and populate all the content for them, because it's all just generated.

Brett House (12:18.341)

Yeah.

Rio (12:31.45)

You can clone a website, just have your agents do it in the background now.

Augustine Fou (12:35.244)

Yeah, exactly. So basically, we've seen a whole bunch of these recipe websites and it's all AI generated photos and recipes, none of it's real. But you might wonder why recipes. It's because on recipe sites, you don't have keywords that would trigger brand safety filters like war or shooting or whatever. So it's easy to get by everyone's filters and start making money.

Brett House (12:44.188)

Yep.

Augustine Fou (12:59.736)

So that's why we've seen a whole bunch of these fake recipe sites come into the supply recently due to agentic.

Rio (13:07.514)

wonder how much of open web spend is actually recipe sites these days. you're right, there's a proliferation of them. And if you do a search for recipe, you'll get like a gazillion come up, right? I think most of them are, you see, most, most of them are completely made by AI. When you read them, you can tell.

Brett House (13:11.887)

Yeah, that's...

Augustine Fou (13:17.334)

Yeah.

Brett House (13:21.189)

Yeah. Yeah. And that was validated by the Zygorski at DoubleVerify. They were saying that that's where a major source of the problem is because it's very little of it's human generated.

Augustine Fou (13:21.336)

Yeah, it's just exhilarating.

Augustine Fou (13:31.734)

And it doesn't trigger the brand safety keywords, right? Like a new site would or any other site like a medical journal where it has the word blood in it and it's automatically blocked, right? So our recipe sites just get through much easier and that's why bad guys are focusing on that.

Brett House (13:41.062)

Yeah.

Rio (13:46.458)

And more than 40 % of I think the legit news content is blocked by brand safety due to keyword blocking, domain blocking, and all the filters you just mentioned. At least 40%.

Augustine Fou (13:56.758)

Yeah, super rudimentary. Back in 2020, when the pandemic hit, we have screenshots of the home pages of CNN, Washington Post, New York Times, be blocked by the brand safety vendors. And that's not correct, right? The highest traffic volume page on those sites, those publisher sites is the home page. And because it had the word COVID in the URL or on the page, they got blocked and they got demonetized.

So all of those, and it's documented, right? These brand safety vendors then blamed their own advertisers for not whitelisting those new sites. It's like, well, your technology is supposed to be better than just the keyword, right? They claim, yeah, it's a natural language.

Brett House (14:39.397)

Yeah. Well, yeah. Dynamic keyword block list optimization. That's got to be a dynamic process because they get stale and old fast based on new cycles, based on political cycles, all that sort of stuff. Yeah.

Rio (14:51.49)

Well, yeah, Brett, was that recent article, Anthony Vargas wrote this article in Ad Exchange. don't know if either of you have seen it yet, but it was good. It talked about how more than half, in fact, this is coming out, we put this in the show notes, but it used 2026. The problem hasn't gotten better with his, like his conclusion, more than half is being blocked today. And these are brand suitable pages, right? They were flagged by like, keyword block lists.

Augustine Fou (14:55.022)

Thank

Brett House (15:04.985)

Yep, I did see that.

Augustine Fou (15:12.142)

And on the flip side, yeah, and on the flip side, right on this side, the good publishers are being demonetized. On the flip side, these same brand safety vendors are not blocking ads going to porn sites, piracy sites, or even worse. So they're basically allowing the funds to go through to all these inappropriate places, even though their claim to their own customers, the advertisers that we're going to help by buying this brand safety services from us, we're going to help you prevent your ads from going there. And there's been others have documented all of these problems over the years, right? Not just me.

Brett House (15:45.499)

Yep. Yep.

Rio (15:45.87)

And you wonder too, like if a legit BBC page got blocked, right? And then like the spoofed, because you know the scammers when they spoof a domain and have the bogus website up there, it's not gonna have any of keywords in it.

Brett House (15:58.61)

Yeah. switching to, so we talked about SSPs, we talked about your experiment, which was super interesting. Did you do a LinkedIn post about that one, Dr. Pham? Yeah. Another one that I read recently was on DSP take rates, which I thought was super interesting. And I'm sure you're pretty passionate about this. Yeah.

Augustine Fou (16:07.042)

Yes, that was, yeah, I'll share that.

Rio (16:15.822)

This is one people don't talk about this one enough. I love to get into this. Yeah.

Brett House (16:19.843)

And you argue, let me just say, let me read it back to you to make sure I understood exactly what your argument was. certainly read it. You argue that blended averages kind of hide the real story. And in one case, a leading DSP captured 99 % of incremental CPMs as the bids increased. Is that the way to interpret it? So the bids increase, so there's this amount over the ask price, right? That.

Augustine Fou (16:35.96)

Yes, exactly right.

Brett House (16:41.959)

the publisher at the end of the supply chain didn't see any of that 99 % 99 cents of every dollar for that increased bid amount didn't go to the publisher. So tell us a little bit of how that works and why aren't

Augustine Fou (16:51.576)

Correct. let me talk about that experiment. in that particular case, what we did is we upped the CPM. So we started with a lower CPM. And the theory was that if you increase the CPM bid, you're supposed to, in theory, get your ads on better publisher sites. Because the good publishers can't afford to sell ads at a very low CPM, only the fake sites can.

Brett House (17:16.037)

Yeah, premium inventory, right?

Augustine Fou (17:19.702)

So in that particular case, as we up to the CPM, the take rates also got dynamically adjusted up so that every cent that we were willing to pay extra went to the DSP in the middle, not to the publisher.

Rio (17:34.17)

So you weren't actually advertising on better websites. You ended up just giving DSP a higher take rate.

Augustine Fou (17:40.395)

Exactly. So even if you pay a higher CPM, so a lot of a lot of advertisers think, if we're paying a premium CPM, we got to be getting good publishers. Then you look at the data, it's all the same crappy publishers. The reason for that is in the middle, the DSP can be dynamically adjusting their take rates on a per impression basis. This gets hidden in the average, right? When you look at their public filings, their quarterly investor reports, they'll say our average take rate is this. I'm not even going to say

Rio (18:08.794)

20 they'll say it's some of them like will say so it's 20 percent or it's whatever it is right so you're so you're saying that's just average that in certain instances is much higher.

Brett House (18:13.723)

Yeah.

Augustine Fou (18:17.454)

Yeah, 99%. So you don't ever see that because not all of it's 99%. But for certain campaigns, all of the incremental CPM that you paid went to the DSP and not to the publisher. So you didn't get any better.

Brett House (18:31.205)

Yeah, it just, it seems like fundamental short-termism. You and I were talking about this on LinkedIn because if a DSP that's servicing the open web, right, it's to their benefit to grow the pie, right? To grow the size of that inventory pool and the validity of that inventory pool, lest in the future threaten their business. Yet that seems, this type of thing undermines that, which to me has to be just a short-termism profit drive versus...

Augustine Fou (18:59.118)

And it is pervasive because every single ad tech company that is a middleman, that is a toll taker has to do this, including Google and including Facebook. And what they do is whenever they need more profit for themselves, they just keep more of it. It's a total black box. You can't go back and track what they kept. If you look at some of the Reddit threads, you will see some of the publishers say, anyone else notice that the, the Was it called the revenue share rate just dropped dramatically this quarter? So basically moving from Q4 of last year to Q1 of this year, anyone notice a 30 % drop in revenue share? It's like, what the heck's going on? The publishers have no control over that because whoever is the middleman, they can adjust the take rates to whatever they want to maximize their own profits. So that's not a DSP issue. It is every middleman can do that because

Rio (19:52.43)

Who else is doing that? So DSPs are doing this. Is every DSP doing this? Are some of them? And then I'm not going to ask everyone.

Augustine Fou (19:57.807)

every platform, literally everyone can do that. So it's like, do you want to rely on them to be trustworthy and to be honest and above board? You shouldn't because every single one of these ad tech companies have shown that they can go do that and just keep more money for themselves.

Rio (20:16.218)

But what about DSPs with owned and operated inventory? Do you think that's different? Do you think because they the, they're actually, they have the owned and operated inventory, they're, you they're actually selling or an inventory through their own platform. Does that make a difference or have you seen the same thing?

Brett House (20:16.261)

Yeah, and in the-

Augustine Fou (20:28.29)

Yeah, this is different. mean, own and operate it. You're assuming you're buying direct from that particular publisher or seller. This pertains to middlemen, right? Toll takers. So just think of it in that context. Whoever's a toll taker, they can dynamically adjust whatever they want. And you don't have the data to see how they did it and what they... It's all black box. So they can do whatever the hell they want. And then there's no explanation that they

Brett House (20:32.665)

Yeah, direct buys. Yep.

Brett House (20:52.561)

So on the average advertiser said, what would you inspect first when you're auditing the supply chain, right from DSPs to SSPs to.

Augustine Fou (20:58.606)

So if you're buying direct from the publisher, you at least know, okay, somebody, one or more, mouse to feed in the middle took 15 % and another 15%. So at least the publisher got 70 % of what you paid. And that should be relatively constant. If there are middlemen, then you can't tell literally, you can't tell. That's where in my case, I bought one domain inclusion list, but still the publisher got 10 cents of what 10 % of what I paid. So If you don't buy direct and you're buying through a DSP with an SSP and there's unknown supply paths, that's where the problem comes in. That's because the bad guys are pretending, you know, they're putting their fake stuff in and pretending to be a legitimate website. And also, I don't know if you know this, the SSPs are bidding on someone else's inventory because all they're doing is arbitraging. So as an SSE, like you're not supposed to be bidding on someone else's inventory, right? You're supposed to help.

Brett House (21:52.423)

But yeah.

Brett House (21:56.71)

Yeah, and that's just an audience extension play where they get they the arbitrage and they make the margin difference. Yeah.

Augustine Fou (21:59.339)

Exactly.

Rio (22:00.388)

Well, is this just an issue for like open exchange or is this like, like, say, let's say if a, you know, there's a trend towards more PMP deals or, or, or let's say programmatic guaranteed. mean, are there specific buying model, buying formats that can minimize this?

Augustine Fou (22:12.448)

Yeah, let me, me, bye.

Augustine Fou (22:18.286)

PMPs will not do that. PGs won't do that. Okay, let me describe why. So PMP means private marketplace. It is only as good as the publishers that are in that private marketplace. So if you have a PMP that is a bunch of junky or crappy websites and mobile apps in it, it doesn't prevent any of the fraud. So the buying model doesn't do anything. You have to look at who's in that PMP. The second thing is...

Brett House (22:43.335)

Well, if you have legit publishers of the PMP, can they still be subject to spoofing like we just discussed?

Augustine Fou (22:48.494)

Yes, yes, right guys, can you just pretend. So let me put it very simply. If you're the owner of fakesite123.com, you will never put your own domain in the bid request because you will never get any bids. You will always have to lie and pretend to be some well-known site like CNN, ESPN, whatever. So in those cases, now it's also ties to the MFA sites, right? So ANA came out with their 2023 or 2024 report saying, oh, there's a whole bunch of MFA sites.

And then Jones Media started selling a list of MFA sites to block. The moment someone bought that list of sites to block and added to their block list is the moment those sites started lying and doing it. It's meaningless. They can just get around the block list just by lying about the domain. And again, they'll catch us if you can. If you don't bother to check the ads out TXT, it's not our fault that you let it through. But the thing is after people bought the block list.

Brett House (23:26.769)

Yeah, it's defunct. It's meaningless, right?

Augustine Fou (23:43.651)

They saw that in their place reports, all those MFA sites went away. They did not actually go away. They simply don't show up in the data because if you're an MFA site and you started saying it's ESPN, all the fake stuff and now commingled in that same row as the real ESPN inventory. So it just gets hidden. So the fact that the MFA sites went away and they published a study, a follow-up study a year after saying, yeah, the MFA site, MFA problem is solved because they all went away. Okay. It's all still there. It's because they're looking at log level data and place reports that use the domain and the bid request, not the domain where the ad actually went. So all of that was misleading. And I had to write back to ANA and said, no, that's misleading. The MFA sites are still in there. I can see it in the data. It's just that no response because they didn't want anyone else to hear about that.

Brett House (24:21.231)

Yeah, yeah.

Rio (24:29.572)

How do they respond?

Brett House (24:29.574)

Yes.

Brett House (24:33.333)

interesting. So your conclusion, mean, back to Ryo's question is that the curated supply, PMPs and other, it doesn't really reduce waste or fraud. It just kind of rearranges that a bit. Yeah.

Augustine Fou (24:42.208)

No. Correct. And the only way to really solve it is to measure it with a post-bid JavaScript like mine through analytics, where you can actually see where did the ad actually go. So then you compare that to what they said in the bid request. And if it doesn't match up, you know they lied to you. Simple as that.

Brett House (24:53.628)

Yeah.

Brett House (25:00.463)

Yeah, so, so, know, click. Yeah.

Rio (25:00.942)

Sounds like it's a whack-a-mole problem though, right? I mean, you have to be like constantly surveilling it, identifying where the ads come up. And if you identify that there's a potential problem, investigating it more, letting your vendors know.

Augustine Fou (25:13.486)

But the thing is, when you have post-bid JavaScript, you can actually see, okay, because these bad guys rotate through a whole bunch of domains and mobile apps. So we can see, oh, it's the same seller ID doing all of this. You just block it at the seller ID because we know that seller is a bad actor. So then you block it at the seller ID level. And then no matter what domain they spoof or what app they spoof, we've blocked them. So that's how DSPs are using food analytics right now as a feedback loop, right? We don't do anything in the pre-bid because Pre-bit filters don't work at all. We measure post-bit, so where the ad actually served. And then if we can see the same seller is doing all these shady things, right, using bot traffic, lying about the domain, whatever, then we just cut them off. We kick out that seller, so we never bid on it again. So then no matter how many domains or apps they put in, we've cut them off.

Brett House (26:05.223)

Yeah. So back to kind of buying type, mean, there's a lot of talk around Google's Performance Max, Meta's Advantage Plus, Amazon's Performance Plus. They're also black box systems. Fundamentally, they're walled gardens. Do you think these are fraud multipliers? Do you think that this is a safer environment for advertisers to be spending against? What's your perspective on that?

Augustine Fou (26:28.014)

Nope. So there's a couple of things you have to understand the walled gardens in two parts. The main property, say for example, Facebook app and Instagram, that's the main property. And then the audience network. So Facebook has something called FAN, Facebook audience network. And those are all the outside sites and apps that also run ads for them. And those are the sites and apps that get a share of the revenue. they have, yeah, anything that's not on the main sites.

Rio (26:52.282)

That's all the offsite inventory. Correct. Yeah.

Brett House (26:54.683)

Yeah.

Augustine Fou (26:57.902)

So all of them have the motive and the means to commit as much fraud as they want. So if you turn off the audience network, you would have avoided 90 % of the fraud. Of course, on Facebook itself, on Instagram, on YouTube, there's still a whole bunch of fake accounts. But if you want to advertise on those platforms, that's the kind of cost of doing business you have to accept. But at least not the 90%, right? That's how right? Yeah.

Brett House (27:19.973)

Yeah. You're safer going direct again in filtering out the audience extension plays outside of their networks. Okay.

Rio (27:22.212)

Well, I'll-

Augustine Fou (27:27.384)

Yes.

Rio (27:27.674)

Okay. Do you think, and I'm to explain why in a second, why I'm asking this question. So if Meta is an advantage plus, do you think it's steering buyers towards, let's say the shader inventory and like the bad behavior. The reason why I asked this is I remember I was running a campaign with a client last summer, this is for an auto OEM. And we were specifically uploading audiences to Meta of people who purchased certain vehicles because they would qualify for certain posts, you know, let's say post purchase service, right? These are not big audiences 10,000 at a time, right? And we were specifically targeting them with personalized creative based on their make and model and the specific offers that availed of their geo, right? It was interesting. Meta, without asking us, flipped on, AI flipped on, and we went from a few thousand impressions to tens of thousands. The CPMs dropped a lot, so we weren't really paying much more, but we had nobody who was seeing these ads. We had no idea if they were even, like, there was potential lawsuits. We shut it off immediately. We couldn't believe what happened. So curious, has that been like kind of your experience and why do think that's happened?

Augustine Fou (28:36.462)

100 % because, in fact, I'll use a simple story. A person was managing his own music accounts. He was selling music and memorabilia and that kind of stuff. So he was managing his own Facebook campaigns for five years. At the beginning of the five year period, 90 % of the ads were run on Facebook itself, 10 % on audience network. By the end of that, the algorithms had just put 90 % of his ads on the audience network and only 10 % was left on Facebook itself.

What they're doing is they're optimizing for larger volume, lower CPM.

Rio (29:11.076)

And larger volumes gonna be offsite and lower CPM is gonna be offsite.

Augustine Fou (29:13.806)

Correct. There's just unlimited volume offsite, right? There's still a finite number of people who are logged into Facebook all day long, right? So to get the massive volumes, you have to go offsite. So that's where the audience comes in. Yeah. So you were seeing, I'm getting like 10 times the number of impressions, 10 times the number of clicks, but my sales went away. So I said, go look at your interface. Is the checkbox for audience network checked or unchecked? It is checked by default because Facebook wants

Brett House (29:16.742)

Yeah

Brett House (29:23.907)

And his performance. Totally. Yeah.

Rio (29:42.35)

Right.

Augustine Fou (29:43.619)

So he unchecked it. And then he got a frantic call from Facebook saying, your DMs are going to go up, your volume is going to go way down. We're not going to be able to deliver in full for you. you see, Facebook wants them to use audience network because they can just run through all their money as quickly as possible, generate a ton of clicks. And if you're not smart enough to look at outcomes and just look at, yay, we got a whole bunch of clicks, you're going to say, wow, that's looking like it's so good. But it's completely useless. All of the clicks are fake.

Rio (29:47.788)

yeah.

Augustine Fou (30:13.196)

So again, going back to just turn off the audience extension, audience network to avoid 90 % of the fraud and at least on these walled gardens. So Facebook has the benefit of people logged into Facebook and Instagram all day long, right? So if your ads remain on the main properties, you'll get your ads in front of humans. Same thing with Google, humans Google things. So if you leave it on Google and then turn off all the outside search partner network and all that, you're gonna get your ads in front of humans.

Rio (30:41.252)

So it's a Facebook example. So like with that, they rip through your budget quicker, right? So like it just trashes your pacing, right? But I guess the thing is for all of the off-site traffic, do have to pay, they have their own audiences and own an operative inventory, which is Facebook, right? But I guess you're right. There's only so many people logged in. it's more questions, they do have to pay out to these other advertisers for the site they're saying. they're, right.

Augustine Fou (31:04.75)

Yeah, so it's less lucrative. It's less lucrative, right? They get half the revenue share, right? Because the other half goes to the site, but it's incremental money for them. So they don't have enough owned and operated on the main property to be able to absorb all the irrational budgets that P &G, Unilever, all these guys did. So they have to use the audience to just burn through all of it.

Brett House (31:14.278)

Yeah.

Brett House (31:21.191)

Yeah, this is, you made that point in the first episode, that these budgets don't justify the actual available inventory, right? And so you have these...

Augustine Fou (31:32.813)

Yeah, there's not enough humans on earth. There's not more than 24 hours in a day. And even with parallel consumption, where every human is using their phone while they're watching TV, you still can't account for the trillions upon trillions of ads being sold through programmatic today. And I have that chart that shows by 2013 or so, humans consumption of all these types of media have completely maxed out and plateaued.

Brett House (31:50.011)

Yeah, and the key is that you...

Augustine Fou (31:59.587)

whereas the digital spend curve keeps shooting up like a hockey stick. So that gap is all made up with bots and other forms of fraud that you can't explain by human activity.

Brett House (32:11.227)

Yeah, well, you just did a webinar. shared the deck with me that was literally called the agentic AI area in ad fraud. What changes when fraudsters, intermediaries, buying systems all get access to the same sort of class of autonomous tools like in it?

Augustine Fou (32:26.354)

the, the, the term I use was kind of a democratization of the fraud. So it's like, if anyone else can get this, yeah, they don't need any skills like programming skills. Right. So yesterday I saw on YouTube, was it called vibe hacking? Okay. You don't need any hacking skills. You just tell the AI to go hack this competitor site for you and they'll do it for you. So that's really the problem because a lot of these humans, they don't even have the right skills to do this.

Brett House (32:32.436)

That's a negative way to talk about democratization, right?

Brett House (32:49.573)

Yeah.

Augustine Fou (32:55.406)

But now with AI, it's democratized ad fraud to literally any script kiddie who wants to get into it. So again, hopefully the advertisers are a little bit smarter asking harder questions and don't just let the platforms use their black box and spend your money. And then you have to rely on their reports, which is basically them grading their own homework. Of course, what are they going to tell you? Oh, everything's fine. Don't worry. Just spend more.

Brett House (33:02.139)

Yeah.

Rio (33:19.704)

I saw this post by, by Tim Rowe, who's president of state of streaming. We've had him on the pod before too. And like, he was talking about how, I mean for streaming anyway, for CTV, that he was looking at the amount of content that was supposedly viewed. And it was, it would have meant everyone on earth had viewed several hours of content in a day. was just an impossible number. like, I mean, I know you've posted some stuff about that too, Dr. Fu, about how these numbers just don't add up. Do you have any numbers you can share with the viewers about what you're seeing and how it just doesn't make sense?

Augustine Fou (33:50.136)

Yeah.

So I'm going to cite a chart from TVision. So they basically have a little set-top box that measures people sitting there in front of TV watching streaming. So the curve that they published is that most humans watch streaming after dinner. So between the hours of 6 PM and midnight or maybe 1 AM, that's when the majority of the streaming occurs. And then they also publish, OK, these are the actual shows that people are watching. So those are great data points to have most humans are watching like those 15 actual CTV streaming apps like Peacock or Paramount or Netflix or whatever. Netflix and YouTube being the largest two. What the heck are these hundreds of thousands of other CTV apps and nobody's ever heard of before, right? So then we get into the ratios. So I've said before, these two statements are true, right? They seem a hundred percent opposite, but they're true at the same time.

How can CTV be 100 % fraud-free because fraud is impossible at the same time as CTV is 100 % fraud? How can both of those statements be true at the same time?

Brett House (35:05.543)

Yeah, I think I know, but answer it for our audience.

Augustine Fou (35:05.73)

The answer is so simple once you hear it.

Rio (35:06.874)

It's not just...

Augustine Fou (35:10.572)

Okay. So basically when you're buying ads from Hulu, from Disney, from Paramount Plus directly, there's no fraud in CTV because it's impossible because bad guys can't get money out. So they're not in that stream, right? If you're buying from Paramount and you uploaded the creatives to them, they're running it for you. That's it. And Paramount's probably not trying to rip you off. Whereas if you're buying anything CTV on Big DSP, it could be 100 % fraud.

Brett House (35:29.414)

Yeah.

Augustine Fou (35:40.531)

And that comes from the fact that the earliest case of CTV fraud was where Grindr, the mobile app, simply lied and said it was CTV. They just falsified the bid request. So instead of bidding out for display ads, they started lying and saying it's CTV, just change the device type. And they got away with it until they were caught. So I don't think Grindr was the only mobile app doing that because you're getting way higher CPMs for CTV than Display or OLV.

Brett House (35:52.316)

Yeah.

Brett House (36:10.172)

Yeah.

Augustine Fou (36:10.316)

So every mobile app, every JavaScript in an ad slot, right? Even smart refrigerators are fabricating fake CTV bid request. So that's how you're getting the tens of billions of CTV impressions being bought and sold.

Brett House (36:23.687)

CTV that's not CTV but charged at CTV prices.

Rio (36:24.89)

Well, yeah, here's what Tim, here's the Tim had probably, this is great. found it. So he was Nielsen tracked 60.7 trillion minutes of television streamed in 2025. That's 62,081 minutes stream per person who's over 16, which is 5,173 minutes stream per person per month.

Augustine Fou (36:28.301)

Yes.

Rio (36:47.386)

which is 170 minutes per person per day. There's no way that's even remotely accurate, right?

Augustine Fou (36:53.378)

No, I mean, some basic macroeconomic views like that will tell you. I mean, I did the same with just basic display ads. And it's been, you for the 15 years since programmatic became mainstream, it's been completely out of whack with reality. But no one wanted to hear that. And they preferred to hear that the legacy fraud verification company said, it's 1 % fraud. Don't worry about it. Just keep spending. Right. So that's been the coverup that's been happening for the last 15 years. Now, I don't necessarily think the ANA or the advertisers or the publishers are being bad guys, right? It is simply that they were not given the right information. So what I tell them recently is like, okay, if you're using the legacy fraud verification companies, they've told you fraud is 1 % for the last 10 years. Obviously there's nothing to do if that's so low, right? It's within the noise level, right? So now if you upgrade your tools and you have better analytics, if you can see better, you can do better.

Brett House (37:41.148)

Yeah.

Augustine Fou (37:49.741)

And so that's where when they upgrade to food analytics, they can say, wow, I never saw that there was 50 % red in my campaigns. No one ever told me that before. So now I said, don't panic. Now that you can see where the fraud is coming from, you can systematically clean it up. So for example, one week you add the top 10 bad guys to your block list. Next week you add the next top 10 bad guys to your block list. And that way you can progressively clean your campaigns. And in doing so, your CPMs are going to creep up because you're buying better inventory. But the quantities that you buy are going to go down because you don't need to buy that much. So overall, you're still saving money.

Rio (38:28.954)

Dr. Food, turning back to the DSP take rates again, because we talked about it little bit. I know they vary quite a bit. I think Amazon was saying, because they have owned and operated, I think they're saying 1%. I don't know if that's true or not. then other ones, see others have said they're around 20%. Clearly, it's opaque and the numbers maybe can't be trusted. How can brands be sure that they know how much DSPs are taking?

How can they pick the right one? And even looking at like the SSP problems, how can they make sure when they work with publishers that they're limiting the number, only working with legit ones, ones that aren't doing this behavior? What are some best practices you can call out?

Augustine Fou (39:08.972)

Yeah, so when you're dealing with the DSPs or any kind of middlemen, you just can't know because they're not going to give you the data, even if you insisted, even if you sued them for it, whatever, because there have been cases where the advertisers have asked for log level data and asked for the data behind the verification firms. And they'll say, no, your contract doesn't allow you to see that data. Sorry. Okay. So if you're dealing with middlemen, there's many ways for them to avoid giving you the data So the best way, if you still want to use programmatic pipes, is just do what I said before, which is go talk to the publisher and say, we know you sell through multiple SSPs. Just tell us the one you prefer to sell through. And you specify that supply source and to the domains. So then you don't let any other domains get in there. So you're essentially doing an inclusion list, not just of the domains and apps, but also of the supply source.

Rio (39:50.53)

and just limit it to that one.

Augustine Fou (40:04.718)

one SSP, not 45, not 98 or whatever, right? And I've done those experiments as well. If you don't limit that, then there's all these supply path complexities which allow fraud to get in. Okay, so that's one. The second is actually go back to buying direct. So Disney, Paramount, Hulu, they all have self-serve interfaces where you can literally create an account, log in, upload your 30 second ad, and then pay the fee.

Brett House (40:21.167)

Yeah.

Augustine Fou (40:33.794)

That's the most erect

Rio (40:34.714)

So buy through their interface or their DSP instead of like a, let's say a generic DSP. Okay.

Brett House (40:35.399)

Duh.

Augustine Fou (40:39.062)

Anyone else? Yes.

Brett House (40:40.101)

But does that scale when you're thinking about a multi-channel, multi-media buy?

Augustine Fou (40:45.964)

Let me ask you the whole, yes, of course. So let me ask you, how many CTV apps do you even know about? Okay. It's less than 20. If you look at the TVision data, it's basically 15 that most humans actually view. So, okay, you can set up 15 deals. You can log into 15 sites. I mean, some of them are actually together like Disney and ESPN and Hulu are all the same. So you just log into one place, upload.

Brett House (40:53.765)

Yeah, it's a small number, right? The legitimate, yeah, less than 20.

Brett House (41:04.433)

Yeah.

Rio (41:09.076)

They bundle, yeah. But that's a lot of swiveling though. Like in all fairness, there's a lot of swiveling back and forth. That's a lot of different logins. mean, for let's say, I guess an agency can handle it, but it's still a lot even for an agency because people have to do this. But maybe in the future, once this is like, if we could build some of like meta DSP agentic interface that can kind of, because they all have APIs now, in theory you could build one on top of all of them. Is that starting to happen now?

Brett House (41:09.753)

Damn.

Brett House (41:33.895)

to go direct buy to all of these automated from the brand side.

Augustine Fou (41:37.595)

If the agency's not willing to upload to 15 places, then you should fire the damn agency because they're not, they're not, shouldn't be paid for. Yeah. What else the heck are they doing? So, so long story short, it's just simulate a direct buy as best you can using programmatic pipes directly by where you can to avoid all the fraud. And that gets back to my statement before. Fraud is impossible in CTV because you're buying from the direct publishers who are not deliberately trying to rip you off.

Brett House (41:43.833)

Yeah. What else are they doing?

Brett House (42:06.181)

Yeah.

Augustine Fou (42:06.382)

But then fraud, if you're buying through large DSP, you literally don't know. So you can consider that 100 % fraud until proven innocent.

Brett House (42:15.355)

Yeah, you're forecasting sort of a future that for all this to fix itself, the middlemen and you you've heard it here on Signal and Noise, the middlemen are the problem Right. And it doesn't seem like the incentive structures, the profit motives are aligned to force them to change. It's the incentives. mean, they clearly know they can get gravy on top of incremental ads then from a take rate perspective. They, know, without anybody knowing and it just boosts their numbers for their publicly reported.

Rio (42:28.43)

Well, it's the incentive. Brett, incentives are the problem.

Augustine Fou (42:38.242)

Yeah. Yeah.

Rio (42:45.125)

Well, it makes you think if we could rewind the clock 25, 30 years, right? And say, okay, like the whole incentive problem, the fact that media is compensated based on tolls or take rates. I mean, think that incents a lot of, not to say there wouldn't be problems if we had a different cost model, but I think when you're looking at a lot of things we're dealing with today, it's because there's a perverse incentive to spend more and more and more, you're taking maybe not a fixed amount, but you're always taking a cut of

Augustine Fou (43:06.584)

Yes.

Rio (43:12.25)

Kind of the panel in all these middlemen who jump in and doing that.

Augustine Fou (43:12.75)

Yeah, you're taking a percentage of and most of those pricing models are CPM plus, which means they add something on top of the media CPM. And that's where it went wrong. Right. So let me now layer in one more thing, which is the agencies. Okay. So if you thought that the dollar 50 going to 15 cents was bad, what actually happens is the agency is charging the advertiser, their own client, $15. Okay. That's the CPM that advertisers pay for display ads.

Brett House (43:22.265)

Yeah. Yeah.

Augustine Fou (43:42.799)

then the same agency holds, yeah, it's $15 CPMs for display ads. The agency turns around and bids $1.50 so that they pocket the other 90%. And then that $1.50, only 15 cents goes to the publisher. So the reality is actually the advertisers paying $15 and the publishers getting 15 cents, which is one cent on the dollar. And again, that

Brett House (43:43.847)

A 10x on the base CPM rate.

Rio (44:08.346)

An agency is keeping this on top of their agency fee. Let's say it's 15%. Right? So they're taking, they're charging 15 % to actually plan and buy the media, but then they're taking.

Augustine Fou (44:17.462)

No, it's actually hidden now. So a lot of the agencies I've seen recently say, we won't even charge you any percentage because they want to win the business. Where do you think they're getting that money? They're getting it from hidden undisclosed margin. for example, through

Rio (44:31.586)

It's funny. I actually saw an RFP this several years ago that actually said, we want you to do it for zero for nothing, because we expect you'll find the money, I guess. But that's like almost saying we know that you'll take rebates. We'll know that you'll take, you'll mark up things at undisclosed amounts. mean, that's, mean, it's almost whose fault is it then? it the

Augustine Fou (44:50.318)

So I blame the advertisers and specifically their procurement departments, right? You have forced the agencies to buy fraud and do shady things because you're not compensating them properly. Right? So advertisers have gotten into the habit of we want 10 times more quantity at one 10th of CPM price. And essentially you're in. Yeah.

Brett House (45:08.729)

Yeah, it's all as much reach for his little dollar investment.

Rio (45:11.672)

It's just because they were picking the agency based on CPMs. Like with big pubs, that's it, right? Okay.

Augustine Fou (45:14.814)

exactly and that's been yeah.

Brett House (45:16.367)

Yeah, it's weird. It's almost like diametrically opposed to how television advertising work television advertising work. They could could demand a premium price because they've got a premium captive audience on a big screen.

Rio (45:26.522)

Well, you would up front where you'd sell 90 % of your inventory, then the scatter market was for that remaining 10%. The scatter market was usually more expensive too, which is interesting, right? So I think digital, it's on its head. Like most of it is programmatic, 65 % is programmatic.

Brett House (45:32.145)

Yeah.

Augustine Fou (45:40.551)

Yeah, even more than that, I think it's 80 % now, but in TV, there was actually finite supply. So it made sense for your hold code to go say, we have all these big clients, we're going to bundle all their spend together and then go negotiate against NBC or ABC or whatever that work. But in digital, it's a complete opposite because there is no limit on the supply. have seen 15 years that the bad guys can generate.

Brett House (45:41.168)

It

Brett House (45:47.452)

Yep.

Brett House (45:56.828)

Yeah.

Brett House (46:03.001)

No, there is is limit. Yeah, there's limit. No, but minus sans the bad guys there technically is a limit. That's the thing that for good supply there is a limit. And that's what's broken right now.

Rio (46:12.675)

A good supply.

Augustine Fou (46:17.134)

Yes. And so for example, the reason a lot of these advertisers got convinced to stop buying from New York Times and buy from exchanges is New York Times is so little inventory because it's based on real human visitors. Okay. So it's finite and it hasn't grown. In fact, it's declined percentages year after year. So all of the growth came from fake sites and fake apps. So that's why the advertisers, that's what I mean by unlimited inventory. So if you look at the last 15 years, and let me talk about the economic

Brett House (46:43.271)

Yeah.

Augustine Fou (46:46.592)

macroeconomic level. There have been billions and billions of dollars shifting into digital from traditional, right, offline media, whatever. So if there's that much demand coming in, then the prices should have gone up, right, based on basic economic theory, right? If the demand goes up and supply is finite, prices should go up. But in fact, we've seen 15 years of price declines because the supply grew even faster than the tremendous amount of dollars shifting into digital.

Brett House (46:54.616)

Yeah, it's 70 % of all ad dollars spent right now. Yeah.

Brett House (47:05.671)

Yeah.

Augustine Fou (47:15.406)

Okay, so all of that gap is made by fraudulent fake inventory. Now on top of that, I published an article two months ago when we look back at the last 15 years of revenue of 255 public companies in the US, the largest ones like P &G and Unilever that shifted the most aggressively into digital actually grew zero or negative, okay, revenue growth. So we now have

Brett House (47:41.672)

Interesting. And just let me just interrupt you for a second, just to give people a size, like an idea of the scale of this, that in 2026, it's about 830 to 860 billion globally is digital, digital specific of a total one to $1.25 trillion ad market, right? 70 % of all those ad dollars. So 70 % of all ad dollars, right? 413 billion in the U.S. Yeah.

Augustine Fou (47:58.521)

Yeah. Yeah.

Rio (48:03.65)

And increasing every year, like every year, like traditional goes down.

Brett House (48:07.143)

So this is a large piece of the overall pie of total ad spend. Yeah, it's really interesting to get the scale of this problem.

Augustine Fou (48:14.444)

And then if we look at just the US companies, because they all have public filings, right? We can see it in their public filings. And then some of them, you can look up the ad age numbers, which is their split between traditional and digital. The ones that are majority digital, where it could be 90 % spend in digital and 10 % in offline. Those are the companies that have grown the least categorically. But then you have other companies like Progressive and Geico, right? You all see in their TV ads.

Brett House (48:40.369)

Go to hell.

Augustine Fou (48:40.568)

They maintained a sizable share of their spend in TV. And those are the ones that grew in some cases double digits compounded annual growth over the last 15 years. So it's in the future.

Brett House (48:51.164)

The linear advertisers, the television advertisers listening to the pod are just like celebrating right now. They're just, right?

Augustine Fou (48:56.044)

Yes, yes, because they want, they want all the guys who thought they were innovative and say, we're aggressively shifting everything into digital, you know, because they thought, it's precision targeting. It's one ad to the right person at the right time. Right. They didn't account for the fraud. So between the inaccuracy of the targeting and the 90 % fraud, all the money spent in digital didn't do anything for their actual revenue growth.

Brett House (49:07.96)

Wonder, yeah.

Augustine Fou (49:23.822)

Thank you.

Rio (49:24.154)

Would CTV save the open web? I mean, it's been the biggest growing... No? You don't consider it?

Augustine Fou (49:28.494)

No, CTV is not the open pod. CTV is not the open pod. There's only a finite number. No, I mean, there's Peacock, there's Disney, there's Hulu. Roku is really dangerous. So I'll leave that to a second different pod. But there's only a finite number of streaming channels or apps that humans use. Those are the only ones you need to buy from, right? So Netflix now has ads. Buy on that. Amazon Prime has ads. Buy on that. YouTube. Humans are logged into YouTube buy on that. That's how you get your ads in front of you.

Rio (49:59.918)

So because of its limited inventory, in effect, you can go direct to these streaming services, publishers, right, and you can buy, so you're not including them, because many people would say CTV is part of Open Web, you don't think it is.

Augustine Fou (50:12.236)

I don't think it is because on the open web, when someone visits a recipe site, they're not logged in, right? They just want that one recipe and they leave. That to me is open web where the users are anonymous, they're not logged in. And that's what gave rise to surveillance marketing, which is where they're tracking users by cookies. And that's when the targeting went to crap because in the beginning, when you have a very finite number of sites, you could probably guesstimate the person, okay?

If they visited Victoria's Secret, Maxim, Playboy, you those kinds of, you can probably guess it's some dude, right? It's some guy. So you can guess that.

Brett House (50:40.165)

Yeah.

Brett House (50:46.087)

Does anybody visit playboy.com?

Rio (50:49.55)

What about retail? What about retail media? mean, because they kind of like don't are, I mean, they have some of the elements of, know, of, you're logged in generally because you're shopping, right?

Augustine Fou (50:57.102)

So I wrote an article called Retail Media Saint or Sinner, and they're mostly sinners because from the get-go, they blended in audience extension. So I said, if I buy from retail media, I want my ads on the screens in Kroger. Don't give me all those audience extensions. Yeah. Yeah. Yeah.

Brett House (51:08.039)

Yes, yes, exactly. was sort of like, yeah, they're acting like retailers when you're in their prop.

Brett House (51:16.773)

Yeah, yeah, exactly. It is a retail experience. You're in Walmart, but you're in a digital environment, right? And so you want ads within that shopping experience, just like you would see an in-capped in a store. And yet they're going out to the open web and they're pushing this stuff for audience extension. And that's when everything fails.

Augustine Fou (51:32.162)

Yeah. Goes to crap and it went to crap.

Rio (51:35.38)

Do know what percent is audience extension versus, let's say, on-site for some of them?

Augustine Fou (51:39.598)

90%.

Brett House (51:41.095)

90 % of retail media is audience extension.

Augustine Fou (51:42.477)

Of course, because they want the volumes, in order for them to match the CPMs and the volumes, they had to do that. And so, no, they basically spoiled. I don't know specifically, but we have seen, I've measured some of these retail media networks and we had to go to them specifically because I work with a big agency and said, cut that out. That's not where we're buying, right? If we're buying.

Brett House (51:50.959)

Is it the same across the board? I mean, is it the same across the Amazons and the Walmarts and the... Yeah.

Brett House (52:05.383)

Yeah, because you think that 80 % of the spend is in Walmart. I think it's 80 % of total spend is in Walmart and Amazon combined. And then there's 200 other, 198 others that the smaller players just are not inherently going to have the inventory. So they're going to almost be forced to do that from an economic model perspective.

Rio (52:16.164)

That's the new one every day, yeah.

Augustine Fou (52:22.862)

And it creates the incentive for them to do audience extension to get more inventory. Same thing happened with local media sites. So for example, a television station has a website. There's finite number of people that go there, right? If you don't live in that town, you're not going to the television company's website. So they were from the get-go, they were tempted by all these ad tech companies. hey, look, we can give you 10 times more inventory if you do audience extension. So that's when all of that went to crap.

Brett House (52:30.897)

Yeah.

Brett House (52:48.71)

Yeah.

Augustine Fou (52:50.986)

And we actually found some of these monetization partners actually juicing all the numbers. So the monetization partner was the one buying all the bot traffic and laundering it through the local TV station website. It's like, how many people live in this town? How can there be 10 million people visiting this local TV station website? Okay. That's not even possible. So all of it came from the audience. What do call it? The monetization partner was cheating and then using their good name to launder

Brett House (52:51.376)

Yeah.

Brett House (53:08.679)

Yeah, yeah. So you're...

Rio (53:11.578)

Alright.

Brett House (53:19.353)

Yeah, so your hypothesis that when advertisers shifted to digital, right, let's say digital and aggregate, they oftentimes saw no improvement gains in performance versus what they originally did in linear. I wonder, and tell me if there's an experiment in place already, if advertisers said, we're going to focus on direct buys direct with publisher buys, premium inventory, if you would see a comparable type of performance outcome as you did traditionally in the linear world.

Augustine Fou (53:47.625)

So I'm going to recommend we don't use the word performance because in lot of advertisers minds performance meant number of clicks they got. Okay. That was wrong. Okay. We're to talk about sales. Okay. Not even outcomes because outcomes are easily fake now due to attribution reporting and all that kinds of, so I can cite a case study. My client, Georgia Pacific, they sold paper towels, toilet paper, paper plates, and other paper products on Amazon product pages. Okay. When they did CTV.

Brett House (53:55.963)

Yeah.

Yeah.

Augustine Fou (54:16.0)

when

Rio (54:25.05)

How did they calculate that? Did they actually get data from the retailers and they did a report to look at? This is on Amazon.com. So for a big CPG, must be challenging, right? So they're buying media at different places. And let's say they don't have brick and mortar, right? They sell through channels. I've gone through those exercises before where you're trying to get data from the retailers. It's really challenging. They don't like sharing it. It's usually trailing. It's usually incomplete. You have to model it out.

Augustine Fou (54:29.394)

Amazon.com on Amazon.com. So they're doing that.

Brett House (54:33.934)

Okay.

Rio (54:53.432)

I mean, I think this makes it harder to go through the exercise you just described because it's difficult to understand cause and effect, showing people ads versus is it actually driving behavior,

Augustine Fou (55:03.662)

Yeah, 100%. I've dealt with this for 15 years as well. So grocery stores, right, if you're just selling soup and soda, it's sold in grocery stores and the manufacturer doesn't even have contact with the grocery store because they sell it to a regional distributor. So again, they don't get that data back soon enough to know how to optimize their campaigns. In this particular case, they had Amazon product pages. They could see the lift in sales once they started optimizing away from crap and optimize towards humans.

Brett House (55:18.406)

Yeah.

Augustine Fou (55:32.398)

The other thing they can do is regional turnoff experiments. So this is what eBay did in 2011, 2012. They turned off the entire West Coast and then all the searches came in it still, right? None of it diminished. So yeah.

Brett House (55:37.147)

Yeah, yeah, yeah. That's where you build your control. Yeah.

Rio (55:46.732)

It's a classic holdout test, right? I guess like what's old is new, right? These are things we've been doing in marketing forever, right?

Brett House (55:48.486)

Yeah, A geo holdout. Yeah. So now that we're on the topic of outcomes or intention or whatever we would have called performance, right? You've drawn a line, I think it was in your webinar, between attention and offenders love to sell, which seems to like a reach metric to what you call attentiveness, which can be observed through kind of stronger behavioral signal.

Augustine Fou (55:52.098)

Yeah.

Augustine Fou (56:12.044)

Yeah. Yeah, let me explain.

Brett House (56:17.895)

How do advertisers optimize towards that and what does that mean?

Augustine Fou (56:22.274)

Yeah. So very easily. Yeah. So let me start with kind of the evolution. So we went from viewability to then attention to I'm going to recommend the next step, which is attentiveness. So viewability has to do with the ad itself and whether 50 % of the pixels were in view. it has to do with whether the ad was viewable. Attention is a nice concept because these attention vendors have done eye tracking studies in the lab. Okay. So for example, a larger ad would get more attention than a smaller ad. Totally makes sense.

Brett House (56:40.742)

Yeah.

Augustine Fou (56:51.82)

an ad that covers your entire screen would get more attention than a tiny banner ad. Totally makes sense. But the problem is these attention vendors cannot actually measure attention when they're running real ad campaigns. Because when you have an ad running and you have a pixel tracker in there, you can't actually turn on the camera of your phone or your laptop to see if the person was actually looking at the screen, because that's going to throw a user prompt and you can't do that. So they can't actually measure.

Brett House (57:18.534)

Yeah.

Rio (57:19.578)

That's also super creepy.

Brett House (57:21.031)

That's true. Yeah, yeah.

Augustine Fou (57:21.772)

Yeah, yeah, like an ad should not be turning on your camera. Okay, so they can't measure that. And so essentially when you're optimizing for attention, meaning higher attention, you're just optimizing for larger ads compared to smaller ones. You're optimizing for ads above the fold compared to ones that start off below the fold. That's not good enough. Okay. So that's when I

Rio (57:39.16)

unrelated or related, not, but question Dr. Fu, these, are some of these devices actually, you're talking about surveillance capitalism, right? Are they, are they listening? Are they filming? Are they, mean, have you seen evidence of that? Or is that all conspiracy theories? What's your opinion?

Augustine Fou (57:56.313)

I think a lot of people give the permission without knowing they gave the permission, right? So Google has this thing that listens for songs and helps you identify what song is playing. The speaker, the microphone's on all day long. Okay. So you don't know that because that's how it works in order for you, but that's kind of a side topic. Let me, let me just finish the topic of attentive, attentiveness. So what we can measure is actually on the landing page of the advertiser. So when the person clicks through from the ad and gets to the landing page, Basic common sense tells you if a human deliberately clicked on an ad, like a search ad, they're looking for something. So they're probably going to do something like move their mouse, click something, at least touch the screen, scroll up and down. So attentiveness is the percentage of users who arrived on the landing page that actually did something. And I can give you a very simple contrasting example. From Google Search, the people who clicked through from Google Search that arrived on the landing page, and we see UTM source equals Google 70 to 90 % of those users actually did something else, like move the mouse or click something. That's great. Users coming from Programmatic or users coming from TikTok, only 1 to 2 % of them did anything on the landing page. So if you're an advertiser, which is the more valuable visitor for you, the ones that actually did something? So now to tie this back to attention and viewability. Attentiveness already takes into account both of those because the ad would have had to have been viewable. The human would have had to pay attention to it to then get influenced by the creative message and then want to click through and then come to your site. So if the person came to your site, first of all, it's human and they did something else. We've already accounted for viewability and attention. And this is directly measurable on the advertiser's site. And it gives them a direct actionable optimization, which is we're going to spend more money on the pay channel that are to users with higher

Brett House (59:50.263)

On Google, Google search wins again. Right? Last click attribution is back. Yeah, that particular example.

Rio (59:56.666)

It's greatest business ever built. Google search, yeah.

Brett House (59:58.439)

Really?

Brett House (01:00:05.551)

Yeah, yeah. that same logic can apply back to like a chat bot once they start introducing advertising, well, they have already. It's the same kind of path that you're following through to see what they do when they actually arrive and how does that compare versus other types of traffic, which you just outlined.

Augustine Fou (01:00:12.588)

Yeah. Yeah.

Augustine Fou (01:00:22.36)

Yes. Yes.

Rio (01:00:24.044)

Interesting too, because you think about like what you just described could obviously be mimicked by bots pretty easily, but obviously with some of the filters that you're talking about, you could filter out or not count bot traffic, right? But then, but as people start having at some point, if you think that we're going to have our own agents going, doing things, we've seen a lot, like even on Signal of Noise today, we've seen a lot more bot traffic going, clicking around on things. It's interesting. So in the future, I

Augustine Fou (01:00:38.958)

Thank

Augustine Fou (01:00:46.414)

Well, need to add analytics on your site because we can actually tell that apart. So there have been campaigns from PMAX where the bots actually are faking the clicks on the landing page, but not only could label them as bots beforehand, but we can look at the click patterns and say, okay, that's not a natural human click pattern. And it's simple thing like.

Rio (01:01:03.576)

Right. Or it's the same click pattern again and again and again, right?

Augustine Fou (01:01:06.668)

Yeah, yeah. there are things now the bots will eventually optimize that. And I mean, we've seen bots actually copy off the human mouse movement patterns and just replay them. That's a simple way to get around detection, but they would have to fake a dozen different things correctly for me to not catch them.

Brett House (01:01:24.911)

Yeah. And can you catch them? You've also mentioned that sometimes fraudulent reporting passes over through to like Google Analytics, like GA4. I mean, have you seen that?

Augustine Fou (01:01:32.611)

Yeah. So the problem with Google analytics is that they are required to filter out the bots. And you might think on the surface, that might be a good idea, but actually it's not because you actually want to see that there are bot clicks coming from different paid campaigns so that you know you have a problem. If GA just doesn't show you the bot clicks, you won't know that you have a problem with those campaigns. So that's where clients are adding food analytics on site tags next to GA, not in place of.

Brett House (01:01:55.664)

Yep.

Augustine Fou (01:02:02.191)

So when their discrepancies either too high or too low, they can use food analytics as a source of truth to help troubleshoot those discrepancies.

Rio (01:02:09.08)

And they're comparing the two. They can see like what GA4 is reporting and see what Foo Analytics is reporting. So, okay, there's a spike here and all Foo Analytics is reporting, you know, increase in bot traffic correlates with this. Okay. Makes sense. But I bet the bots are getting much...

Augustine Fou (01:02:17.848)

Yes. Exactly.

Brett House (01:02:20.167)

Nice product placement there, Dr. Fu. Nice product placement. I like that. That was good. That was pretty slick.

Augustine Fou (01:02:23.138)

Yeah.

Rio (01:02:24.484)

But I bet the bots are getting way more sophisticated though, right? I mean, like as AI improves, mean, how do you keep up with it? I mean, every day there are new advances.

Augustine Fou (01:02:33.486)

Well, I'll give you little secret here. And I've said this before, so I don't worry about the fraudsters listening to this. So it's usually easy for a bot maker to tell the bots to click on a bunch of stuff. And you would think that that's basically a random pattern. So if we see that the clicks are spread too evenly across the page, that's not humans. On the flip side, it's easy for the bot maker to say, click on this button or this XY pixel repeatedly.

So again, that's not human. So if it's too ordered or if it's too random, it's not real, right? Humans are somewhere in between. So I've said this before, we can still measure it. Again, they would have to fake a dozen different things correctly in order to get by my detection.

Brett House (01:03:07.879)

Interesting.

Brett House (01:03:18.533)

Yeah. So, so.

Rio (01:03:19.226)

And I bet when looking at patterns, I mean, bet that the bot traffic ends up being very similar over time when you're looking at it in the aggregate, right? Because I mean, that's what these, I mean, when you look at what these LLMs do, I mean, they're averaging kind of everything they see. Yeah.

Augustine Fou (01:03:29.55)

So we play the same patterns, yeah, basically.

Brett House (01:03:33.031)

So, yeah, so let's get to the sort of so what. What can marketers that are listening to the podcast do this quarter and like the immediate future? If you were briefing a CMO, you do this all the time, right? A procurement leader, immediate lead, whatever it might be. What three immediate changes would you mandate this quarter or let's say the first half to reduce waste and improve signal quality?

Augustine Fou (01:03:52.751)

First is ask harder questions. Don't assume that any vendor that you're buying from, right, whether it's the verification vendors or the DSPs or whatever, is telling you the truth. Not because they're malicious. It's simply that they have no incentive to tell you the truth, right? They're going to tell you what you want to hear. So you have to have independent measurements. So this gets to point number two, which is like, have the vendor show you the data. And I'm...

Brett House (01:04:10.748)

Yeah.

Augustine Fou (01:04:19.36)

asking that specifically because if the verification vendor and you ask them, can you explain why you marked this as a bot or why you didn't mark this as a bot? They'll say, we can't explain it to you because it's our secret sauce and we don't want bad guys to know. They don't know what they're talking about. So stop buying from those vendors. You need a vendor who can actually explain everything to you, give you the supporting data so you can understand why something is a human versus why something is a bot. So once you have that, then you can have confidence in the data that you're seeing.

Then you can say, well, obviously these are bad sites and bad apps eating up my impressions. And then the third step, very simple, start blocking these bad guys. I wrote an article called the maturity level of advertisers using food analytics. First is lower the red, right? Start optimizing away from the fraud. Second is start optimizing towards the humans, right? Increase the blue, because you want more ads to be shown to humans. And then the third is optimize for attentiveness on the landing page.

Brett House (01:05:11.686)

Yep.

Augustine Fou (01:05:17.718)

and outcomes if you can. So if you're an e-commerce site, you can see the outcomes, the sales right there. If you're CPG or soup and soda, it takes a little too long because you're selling in grocery stores, but still you can use attentiveness as a gauge for, if that person was interested, they would actually do something on your site.

Brett House (01:05:31.481)

Yeah.

Yeah, you might be getting lagging indicators, but it's still going to give you better insight than if you're not doing that as a CPG player to plan ahead for your next buying cycle.

Augustine Fou (01:05:40.814)

us.

Rio (01:05:44.282)

Dr. Foo, what percent of overall ad spend is fraudulent? mean, just thumb in the air.

Augustine Fou (01:05:50.543)

No, I'm not going to say that because that would essentially be an average. Okay. So I will give you the range. The range could be 1 % fraud or 100 % fraud. And I've seen all of it. So again, because I don't have 100 % measurement for the entire digital ecosystem, it's irresponsible for me to extrapolate and tell you a number. Whereas everyone else does that. Okay. Those numbers are not true because they're just extrapolating from what they can see.

Rio (01:06:17.609)

Actually, to piggyback on what you said before, coming average is maybe part of the problem, right? Because if you think about it, like you can hide a lot of bad behavior if you come up with an average. If let's say the average, let's say the overall number is 20, 25 % of the, oh, that's bad, but not that bad, right? But, in certain campaigns to your point, it could be 99 or a hundred percent.

Augustine Fou (01:06:22.798)

It is.

Augustine Fou (01:06:26.616)

Exactly.

Augustine Fou (01:06:32.43)

100 %

Brett House (01:06:33.927)

Yeah, I and I like the question dr. Fu because I mean I think Rio is leading you to say 99 % right and just make the most controversial statement that will put in the LinkedIn post dr. Fu claims

Augustine Fou (01:06:39.362)

Yeah, yeah, no,

Yeah.

Rio (01:06:45.39)

We wouldn't do that, come on. People probably will.

Augustine Fou (01:06:46.542)

No, you can actually do that. You can actually do that and we'll get a lot more views, right? Dr. Fu claimed it's 99%. So all the haters will come out of the woodwork and start hating on us. But, Rio, this actually goes back to what I, while I got into this 13 years ago, when I was at Omnicom, we were actually seeing average click-through rates on search campaigns to be 9 to 15%. That's the average. But then when you got place reports that broke out each individual site, some of those sites, the fraudulent ones were 100 % click-through rates.

Brett House (01:06:50.641)

Yeah.

Brett House (01:06:54.158)

Yeah.

Rio (01:07:01.134)

Yep.

Augustine Fou (01:07:16.174)

Again, the problem was hidden in the average. So you say, yay, my search campaign was working great. It's 10 % click through rates. But then when you broke it out, you could see the problem. So just to close this out, you're right. Averages have not done anyone favors because it's actually hidden the problem from view.

Brett House (01:07:33.958)

Yeah, and you can get that in your most basic platforms. I we've run email campaigns, for example, and I've seen, what I've been talking about this a lot is, you see a pretty healthy click through it, but then you dig in, I noticed that it was like two domains in my entire email list that were driving 90 % of the clicks. I'm like, and so yeah, it was like 500 clicks from one domain, from one email address identifier. And I'm like, that has got to be a bot that's actually, this is actually an anti-fraud.

Rio (01:07:51.032)

Yeah, we thought we had really good results at first.

Brett House (01:08:03.271)

protection bot for the company. But that was the activity that we sort of pulled out. Yeah. So should we go to quick hits now that we're at the hour or just above the hour? What do you think we're at?

Augustine Fou (01:08:15.131)

didn't we already do that, which is like the three recommendations. I think that's, mean, unless there's something else you want to talk about.

Brett House (01:08:23.131)

Well, we have quick hits where we give you kind of quick, short answers to very simple questions. So I'll give an example. So one metric you would ban from marketing decks tomorrow.

Augustine Fou (01:08:31.595)

Okay.

Augustine Fou (01:08:38.414)

click through rates or view through conversions.

Rio (01:08:42.468)

What's one place in the stack where a lot of the bad behavior happens that you could just nuke if you were able to?

Augustine Fou (01:08:49.07)

It's going to be, if we're talking about programmatic media buying, it's going to be a few SSPs that let the majority of the fraud onto the ecosystem. And if you cut some of those off, it's going to be a lot better.

Brett House (01:08:59.943)

Yeah. And the final one, because I know you got to run, we're at the hour and we all have meetings, is one question procurement should ask every DSP from a contractual perspective immediately.

Augustine Fou (01:09:12.246)

what percentage of the dollar that I'm spending actually goes to the publisher? And I can guarantee you 100 % of the DSPs will not tell you.

Brett House (01:09:21.691)

Wow, well, that's great. Well, Dr. Fu, awesome having you again. Thanks for spending another hour with us. It's always super insightful. Love the research and just the things that you support. And I think it's for the betterment of the industry. It's for the betterment of the open web.

Rio (01:09:23.012)

Fantastic.

Augustine Fou (01:09:26.35)

Thank you.

Rio (01:09:26.66)

Totally.

Rio (01:09:36.858)

Yeah, your previous episode, that's one of the ones people continue to come back to. see people watching it. So we hope more people will go back to that one, watch this as well as this one too. This is a great discussion. This is lived up to the building. Thank you.

Augustine Fou (01:09:49.582)

Thank you very much, Rio. Thank you,

Brett House (01:09:49.915)

Thank you. And for everybody listening that's made it this far, follow us on YouTube, Spotify, Apple Podcasts. You can find all of our content editorial podcasts and otherwise on www.signalanoids.ai. And we will see you next time. Thanks everybody.

Augustine Fou (01:10:08.719)

Thank you.