← AdExchanger
AdExchanger
Addressing Addressability, With ID5 CEO Mathieu Roche
00:46:31
When Google reversed its decision to deprecate third-party cookies, the first word that popped into ID5 CEO Mathieu Roche’s head was an expletive. But it doesn’t even really matter what happens with cookies on Chrome anymore. “Most of the...
Loading summary
Transcript87 lines
- [00:00]
A
Foreign.
- [00:13]
B
Welcome to Ad Exchanger Talks, the podcast devoted to examining the issues and trends in advertising and marketing technology that matter most to you. I'm Allison Schiff and you're listening to or watching Ad Exchanger Talks. The identity of my guest this week is someone who knows a lot about identity, Matthew Roche, CEO and co founder of alternative identity provider ID5. We'll nerd out in the extreme on third party cookies, addressability, probabilistic versus deterministic matching, what ID5 is referring to as adaptive identity and and lots of other nerdy goodness. But first, and speaking of nerdy goodness, get your ticket to Programmatic IO New York. If you're an ad tech, you've just got to be there, plain and simple. It's taking place in New York City on September 29th and 30th. So if you've been waiting to get your ticket, don't get them while they're hot. And make sure to save a little dough. Not a cookie reference. While you're at it, use the code pod, crush, all one word and all in caps at checkout for 25% off the cost of your ticket. See you there. Hey Matt, welcome to the podcast.
- [01:34]
A
Hey, listen, thank you for having me.
- [01:36]
B
All right, so let's do what I always do. We'll start it out. My favorite question, what is one thing about you that not a lot of other people already know?
- [01:47]
A
So it's a very personal, personal comment. I did not have any kids when I was 30 and I had three when I was 31 because my daughters are 15 months apart and I had twins. So that made up for a very interesting few years. Wow. Before starting your company, which was an all other level of challenge.
- [02:09]
B
I was going to say it's like you have four children because you also have your company. Not to compare the beauty of the parent child relationship to birthing a startup, but you know what I mean.
- [02:24]
A
Yeah, there is some of that. I think when you are lucky enough to be like the parents, quote, unquote, I'm doing quote I know it's a podcast. Nobody's going to see that. But air quote the parents of a company that has been successful enough to be now almost eight years old and over 60 people and everything, it doesn't feel like your little precious thing anymore. It's more of a shared responsibility. Right. We've got investors, we have a board, I have a leadership team, an executive team. So it's less of a. Yeah, it was very much like that in the early days. It's really your thing and you want this to be successful and to grow up and be healthy. Now it's not like we're too big to fail, but like we've felt like we've passed that and the kids are off to college, so less of a responsibility.
- [03:18]
B
Well, how did you, I should know this about you because you and I have known each other for, for a long time by now. But how did you find yourself in the world of ad tech and online advertising? Like what's your origin story?
- [03:30]
A
Yeah, that's interesting. I, I, so I've been in this industry over 25 years, which you know, dates me now. And I started really, I, I discovered the Internet and E commerce and online advertising during my MBA in, in Atlanta, Georgia Tech. And when I got back to, I moved back after, after college, to France, to Paris to take a job in consulting. Very boring. And the whole thing at the time, late 90s, early 2000 was well, is this Internet thing ever going to take off? And I came back, I was coming back from the U.S. right. So it felt like I had, I had seen the future. And so I lasted all of six months in this boring consulting job and then joined an early stage investment fund model after the incubators of MIT Media Lab, cmgi, those type of things, and started doing investment in early stage companies that were all at the time pretty bubble burst Internet companies. And one of the early investment we did was in a, in a marketing software company doing audience measurement analytics. And that's how I got my, my first kind of exposure to the world of ad Tech and MarTech and yeah, and then five years later actually joined that company, spent 12 years with that company to the point where I left to go start ID5. So it's been a long time, a long history now of going through the ups and downs of our industry, right. The pre, you know, the early days of online advertising, banners on sites, the days of ad networks, the days of data dmps, the days of programmatic, agency, trading desk, independent trading desks, pre bid, all of that where to the point where we are now where Agentic, Premium, cross channel, Omnichannel, tv, digital television, all of those things are now kind of common knowledge and baseline basically for everyone in this industry.
- [05:49]
B
That was a lot of buzzwords all at once. But yes, this interview, I'll just give.
- [05:54]
A
You a quick history of the buzzwords right here.
- [05:56]
B
The Internet thing has definitely caught on and like you said, you've been doing the ID5 thing for quite a while now. Eight years. Founded in 2017 and even in 2017. I think the writing was starting to be visible on the wall in terms of the need for alternatives to third party cookies. Obviously that's why you founded the company. Better privacy protection, like more sustainable ways for publishers and advertisers to make money and connect with their audiences. That was all brilliant brewing. And maybe it was like obvious to you, but I think there were some people who weren't as aware of it as they should be. I think that there were a lot of people that didn't think that cookies would ever truly disappear. Not that they did. And we'll talk about that. Oh my God, the saga. But yeah, I mean it feels like a very in the weeds industry problem. And when you founded the company, it was a little early in terms of like needing alternatives to third party cookies. So why that? Why was that on?
- [06:57]
A
Yeah, it was very early. It wasn't actually. I mean the driving. You kind of hit on all three topics, right? That we're addressing where we were addressing signal loss, we're addressing data privacy and data protection and we're addressing performance. The first one was really a second thought, right. The origin story of ID5 is how do we enable publishers, let's call it the Open Internet. Right? How do we enable the Open Internet to compete with the walled gardens who already back then when we started thinking about ID5, so 2015, 2016 already were starting to really capture a much bigger part of the investment than we thought they should. So there was a performance driving mechanism there. Like we need better because cookie matching doesn't work. Right. Cookies are not a good enough method, so we need to do better. The second driver was GDPR is going to become a law in Europe next year. No way. Third party cookies drop like 165 times per page on most publishers across Europe are going to cut it from a privacy compliance standpoint. So performance compliance and data protection were really the two driving mechanism. The third one was, oh, and by the way, Safari has started to limit access to third party cookies. Maybe they're not going to be available anyway. So let's also keep that in the back of our head. Right? But the main arguments and the main reason for launching 85 was we need a better infrastructure for the open Internet and we need something that is actually compliant with the upcoming regulations. Those were the two drivers of that led us to launch the company. Third party cookies happened to be an opportunity for us with all the limitations that we see across browsers, Safari and others. But it was really driving better performance because advertising works better if you can recognize who you advertise to. Right? Plain and simple. And also enabling compliance for the whole ecosystem. Because if you don't have access to personal data lawfully, you're in big trouble.
- [09:12]
B
Lawfully is the important word. And it did prove pretty prescient like that idea. It became more and more relevant as the years went by. And I do want to talk a little bit about the cookie craziness, but before we do, I have to ask, and I want you to be honest, what was the first thought that went through your head on Monday, July 22, 2024, 3:00pm Eastern Standard Time? Which is when Google announced that they weren't going to deprecate third party cookies in Chrome after all anymore. What, what was your first thought?
- [09:44]
A
You know what wasn't that date? It was the. I don't have the date and the hour as well as you do, but it was the, it was the one in July 2023 when they said, actually we're not going to deprecate unilaterally, we're going to ask users for permission. This was the one. We're like, oh shit, pardon my french. Because this was I think the real nail in the coffin. The next one where they say, well actually we're not doing anything. I think everybody by then had given up trying to understand. Everybody had already given up about the privacy sandbox. So that wasn't even like a concern. It was just like, okay, yeah, yeah, I mean, yeah, it doesn't matter anymore, right? This kind of reversal about like, well actually it's not going to be a unilateral deprecation anyway because we supposedly we care for the industry and so we don't want to do like there's a bunch of, you know, BS excuses. But this was the real, the real change I think. And this what led us to change. And this is where we saw a lot of changes from our clients behavior as well. Because until then it was pretty consistent. And that was the first major inconsistency in the whole Google narrative that then led to them just removing the entire thing altogether. But I think the date you mentioned is more the end of the privacy sandbox initiative. The date I refer to is more the end of third party cookie complete deprecation. It's just partial deprecation if you look at it from an industry standpoint because they are available on pretty much half of web traffic. The other half is invisible with cookies. So that was like, okay, it's only going to be a 50% thing now and not a potentially 100% thing. But in the end, I can't share revenue growth numbers with you, but it didn't impact the trajectory of the business that much. And so it's been a bit of a nothing burger, a lot of waste, you know, hours that will never get back. For sure, all of us in the industry, but I think very little impact in terms of how our vision has changed for the need for a better identity infrastructure for the industry, the need for better data compliance, the need for better performance, all of that remains the same. In effect, consumer behavior has actually made it more important because people use more devices. The fragmentation, the difficulty of accessing identity signal, the difficulty of recognizing people in general, technically, legally, all of that has just gone up. Whatever Google did didn't really change that. And so the need for someone like ID5 to solve that problem, create that abstraction layer on top of that technical and that legal complexity, it has never been stronger than it is today.
- [12:55]
B
It's funny how many nails there were in this coffin and yet third party cookies still live. Isn't that funny? And you're also. This is such a dumb question, but you're also fluent in French, of course. So did you think oh shit or did you think au merde?
- [13:13]
A
I've lived in the UK for 15 years and now I spent a lot of my time in New York, so I'm starting to dream in English. So I might have said, oh, shit.
- [13:22]
B
I mean, how much time did this whole deprecation drama waste? Because I feel like the original news, like way back in, whatever it was, 2019, late 2019 or early 2019, February.
- [13:33]
A
2020 was the official announcement.
- [13:35]
B
Yeah, that was a catalyzing event. But then, like, you were just describing all of the back and forth and all of the delays and the navel gazing. People got tired, it took the wind out of everyone's sails. And now, amazingly, people who didn't collect nuts for winter, basically, they're fine. Like they didn't starve. Not doing anything was almost an okay strategy in a way. I mean, of course you need to think about performance and think about future proofing your business, all of those things. But do you think it would have been better for the industry if Google had gone the safari route and like not involved everyone and asked for feedback, just like yanked the rug out and said, said, sorry, not sorry, guys, deal with it or what?
- [14:16]
A
I think third party could give people the impression that they're fine and give them a way to operate at a. I'm not even going to say acceptable, but like give them a way to Operate without having to ask the tough question or without having to do the tough, you know, take the tough decisions of that, that actually benefits their clients, but that are difficult. The difficulty of this whole thing is that publishers suffer massively of not being able to monetize their inventory correctly because half of it is invisible. Brands suffer massively of not being able to engage with their customers because half of them are invisible. In the middle, ad, tech, data, buyer, seller, all of that. Yes, they suffer because their clients suffer, but they don't suffer as much as the principals, if you wish. Right. And so there's a real difficulty in making this industry move because they have this kind of, this clutch that they can, this crutch that they can still kind of, you know, lean on and, and that. I don't know if we can say it works well enough, but like, it works to some extent. It works up to 50% of the time on the web. Right. And so, like, that's. So, yes, it's a bit of a, you know, in French you say like a band aid on a wooden leg. Right. It doesn't cure the patient for sure.
- [15:47]
B
That's an amazing phrase.
- [15:48]
A
But it feels like, it feels like it's, it's, it's, it's an answer. It's a very bad answer, but it feels like an answer. And so like, yeah, let's put up, let's put a band aid and, you know, pretend we're covered. So I mean, Obviously, as the CEO of ID5, I would say, yeah, sure, we should have removed cookies. I think it was mostly a lot of waste for our clients. A lot of effort, a lot of time spent, a lot of money wasted in trying different things that never took off. A lot of snake oil vendors, all of a sudden coming up with, oh, contextual is the new kind of magic. Okay. It's always been smart to put the right ad in the right place. Right. But if you can put in the front of the right person, it's better. So, like, let's not pretend right place is good enough just because, you know, so, you know, the whole, yeah, the whole kind of, you know, cookie less buzzwords created so much noise that there was some good, you know, urgency with it, but there was also a lot of bad shortcuts and kind of, again, snakehole vendors and people trying to come up with alternative ways of doing things that weren't really sustainable. And I'm not even talking about privacy sandbox. So a lot of noise. I'd say 30% positive, 70% negative.
- [17:21]
B
That's not the greatest ratio. There's a When you were talking about a band aid on a wooden leg, I'm going to mess this up a little bit. There's a Yiddish phrase, a helfen aphatoiten beynkis I think is the phrase. And it means it helps like cupping a dead body, which is, you know, like cupping therapy. You put cups on someone's back or their stomach or their arms and like you create a vacuum and like the suction force is good for some reason. But yeah, I guess it's not going.
- [17:51]
A
To revive the dead body for sure. And the band aid is not going to cure the leg for sure.
- [17:57]
B
So I love idioms.
- [17:58]
A
That's how I feel about third party cookies, to be honest. And a lot of people were not doing anything, I think. I mean, I don't know anyone who's not doing anything anymore. I think everybody's going to move past Chrome or no Chrome. Doesn't really matter. I think most of the industry has moved past the notion that cookies were good enough to target and measure advertising on the web, which is a good thing for their clients, right? For the brands and the agencies and the publishers. But it's been a very long winded way to get there while, as you said, the writing was on the world back in 2017, 2018. And yeah, it might have been a more an healthier way to go about if Google hadn't created this kind of storm in a teacup and and try to reinvent everything with privacy sandbox. And yeah, that was quite a bad.
- [18:54]
B
Nothing much, much, much ado about. Not very much. So I want to ask one more question before we take a quick break and just change gears a little bit. Talk about adoption, the adoption of alternative identifiers. So not to be vain, but I'm just going to read the first couple of lines of a story that I wrote that we published earlier this year. It was based on a conversation on stage and an event that ID5 hosted in the city that I went to. So if an alternative identifier is present in the bid stream and no one transacts on it, does it make a sound? The availability of alt IDs and open auctions has steadily increased, but demand has yet to catch up with supply. So the crux of that conversation was that Even though alternative IDs were being deployed, DSPs weren't transacting on them at scale. And one of the reasons for that is because demand was lacking. And demand was lacking in part because cookies were still available. So that story was From February. Could I still write that story today? Have things change? Are DSPs transacting on alternative IDs more now?
- [20:08]
A
So I think there's a number of ways to look at this. Yes, DSPs are transacting on alternative IDs. There is a question of availability. A lot of the adoption, naturally, before products are on the shelf, nobody's going to buy those products. And so you need to put products on the shelf. So you need publishers and SSP to adopt and pass IDs for buyers to realize that, okay, there is untapped opportunity in here that I should use. Right. And so it's a, it's, it's a natural progression. Right. From sellers to buyers. Right. Again, no products on the shelf, no one's going to buy them. It's taken a few years for us to get to a, an adoption level where, I mean, right now there are more ID5 IDs in transactions than cookies. Third party cookies. Right. If you listen to the whole bitstream.
- [21:01]
B
Interesting.
- [21:02]
A
Or more ID 5 IDs than third party cookies. The problem is very few people listen to the whole bitstream because all of the algorithm have been trained to listen to only the bitstream that has a cookie id. Right. And so we have to unlearn how to optimize against cookies. Right. And that takes a lot of time because there's a lot of bit shaping that takes place at all levels of the industry. And so we need to unshape the bid request to make sure that people can listen to everything because there is a lot of hidden gems in the thing that they've not listened, that they've been listening to because it's usually valuable users, iPhone users, for instance, browsing the web on Safari, very valuable, affluent. But because historically they didn't have cookies, people were kind of shaping traffic away from them. It's also cheaper inventory because there's less competition and it's an auction process. So if there's less competition, prices go down. Meaning as a buyer, you have opportunities to create a lot of that. So it takes adoption, it takes reshaping of the traffic, and also it takes the understanding that this is not something that the end buyer will ask for because they don't care. They don't care which ID is going to be used to transact. They care about reaching their users. They care about reaching their users at the right price. If you can make it happen by embracing onboarding, universal ID solutions like a lot of platforms have done, great, they'll be happy. But if you wait for General Motors, or McDonald's or Marshaka to say, oh, by the way, I would like my ad to be served against an ID5ID or against a ramp ID or against a lot of me panorama ID. They're never going to say that. It would be like, oh, I would like the brand of the tires on my car to be this brand or that brand. No, I buy a car, I expect the car to take me from A to B if it's got Goodyear or Michelin tires on it. You're the car manufacturer. You've decided that this brand of tire was better for performance and security. Great. It's exactly the same thing with DSPs and data companies. You're the technical specialist. You've decided that this, these alternative IDs were better for performance and security. You've onboarded them by default into your system. I'm just going to use that. And so there's been a lot of back and forth between publishers saying, like, oh, I'm not going to put products on the shelf until people are ready to buy them. Well, they won't be ready to buy them until you put them on shelf, right? So back and forth between buyers and sellers and then even with buyers like DSP saying, well, I'm not going to integrate with ID5 with ID solutions unless the agencies or the brands ask for them. They're not going to ask for them. Right? And so there's been this, this kind of, you know, kind of catch 22 situation. You know, the Spider man memes, right? I'm waiting for this guy to ask me. No, I'm waiting for this guy to ask me. I'm waiting for the. In the end, it's a technical decision. Identity is a means to an end. Nobody buys identity. They buy targeting, they buy optimization, they buy frequency, right? That's what you're interested in, measurement.
- [24:07]
B
They want performance.
- [24:08]
A
Identity is a means to providing performance. And believe me, when you can bid on users that nobody bids on or users that are super valuable, we have data that shows that the performance, the reach, the scale of your campaigns, the CPA goes down, the scale goes up, all of those things work. And publishers make more money as well because it's better than zero, because their traffic is shaped out and nobody bids on it. So it is just like lifting the value for the entire ecosystem. If you all of a sudden you can consider 100% of users rather than just 50% of users, but it's a collective migration. If you just keep pointing fingers at the other guy down the chain to make the decision for you, nothing's going to happen. Fortunately, we're past that stage now, but it's taken the better part of the last five years to kind of get the market to a point where this was obvious.
- [24:59]
B
It's been a process. Nothing comes easy to us, apparently, in the ad tech industry. So we're going to take a quick. And when we're back, we're going to talk about ID5's new brand positioning AI and I'm going to force Matt to do a lightning round. So that's the first csearing of it, but that's what we're going to do. So stick with us.
- [25:17]
A
Love surprises.
- [25:31]
B
All right, we are back and yeah, let's spend a few minutes talking about ID5's positioning. So in June, I think it was at Cannes, actually, you guys launched a refreshed branding and the positioning that I want to talk about more than the branding, which you're calling adaptive identity. And the idea there is that. And I feel like I should put some drama in my voice when I say this, like the movie trailer voiceover guy, but in a fragmented world and a fast moving world where people constantly switch between apps, browsers and devices and locations, identity cannot be static. But yeah, I mean, like, people engage with the world in very complex ways and that makes addressability complex. There are lots of different signals. People have preferences you have to honor. It's a lot. So the way I understand it, and then I'll, I'll let you explain it because it's actually your thing. But the concept of adaptive identity is technology that can evolve to meet the moment, to adapt to privacy changes or whatever changes happen, as opposed to cookie syncing or something that just is more static and kind of old school. And to do that, you've got to use something like machine learning, otherwise it's not really feasible. So I feel like I've teed you up very nicely there, but explain the concept of adaptive identity and tell me also a bit about some of the investments you're making in machine learning.
- [26:53]
A
Yeah, you've done a really good job of summarizing and describing what it is. You're hired. Yeah, I mean, again, we talked a lot about cookies, but cookies are really relevant in about a third of the time people spend interacting with digital content and services. The other two thirds are spent on smart speakers and mobile phones and connected televisions. And cookies are not relevant. Right. And so when we think about identity, we think about recognizing people everywhere. And that requires us to be able to understand signals coming from all of those Devices and then to cluster them together. We're going to get a bit technical here, but cluster those signals together to recognize.
- [27:38]
B
You do know you're on ad exchanger talks? Go for it.
- [27:41]
A
I can go nerdy. All right, please. So what we've been doing for the past almost eight years. And by the way, you said eight years. It'll be eight years at the Mexico in Germany in three weeks. So we'll be. Happy birthday, celebrating our eighth. Eighth birthday there. What we've been doing for the past eight years is just like collecting a ton of signals and processing them through clustering algorithm to make sense of the people that are behind the devices. The devices and the people that are behind the devices. And so we're building graphs, if you wish, graphs of signals that we are able to apply to recognize users. And our goal being to recognize everybody everywhere. And so that requires, as you said, adaptability, that requires agility, that requires learning, that requires improvement. There is a lot of things that are changing in our world. The technical context of interaction. Whether you're on a rework WI FI network or whether you're on a 4G 5G network on your phone in the car, or whether you're at home on a much more static WI fi, whether you're on a television, in an app or on your phone, on a website or on a computer on an audio speaker, all of those things. Whether you're in Europe or in the UK or in the US in different states in the US all of the conditions that we need to embrace to be able to recognize you in those different contexts are. I mean, the number of combination is surreal, right? And so only through machine learning and through an algorithmic process are we able to do that. And I mean, Google knows what you're going to type in your email before you type it. This is algorithm, this is recognition patterns or pattern recognition. That's what we do at ArtScale to recognize patterns of behaviors so that we can tell our clients, yep, this user here on that device on that day and that other user on that other device on that same day are the same user. And so if you want to do frequency capping, it's very useful. If you want to do retargeting, it's very useful. If you want to do targeting across devices, very useful. If you want to do measurement, understanding, all of that powers those capabilities that really make advertising valuable. And so the notion of adaptive identity encompasses all of that. We need to adapt to legal circumstances, we need to adapt to technical circumstances, we need to adapt to evolving legal and technical circumstances. And we need to do that at the speed of, I was going to say at the speed of AI. Adaptive identities initials are also AI. But like we need to do that in real time over I think about 7 billion times a day, every day. And no static methods of recognizing users will scale to that level. And so that's why we've kind of branded what we do, adaptive identity, to kind of convey that message, that value that we've embedded into our technology to provide identity capability to our clients.
- [30:55]
B
There's something I've been doodling on maybe you can help me with. Does AI or how can AI improve the accuracy of probabilistic identity matching without compromising privacy? Meaning can machine learning techniques help advertisers better connect the dots between different signals and devices? But like still respect privacy constraints, consent frameworks, all of that stuff. Because AI isn't inherently privacy friendly or even perfectly accurate. So like navigating privacy alongside like AI powered identity resolution seems like a tricky balance. Not doable, but hard.
- [31:35]
A
Algorithm are not perfect as you said. But again, back to the example of the Gmail example. Gmail is pretty good at guessing what the necro is going to be in your sentence. This is algorithm. So you can tune in an algorithm, you play between recall and precision. So recall is scale, precision is precision, and it's always a balance between the two. If you maximize scale, everybody is the same user. If you maximize precision, every interaction is a different user and you need to find the sustainable points on that curve between scale and precision. But I think again we've proven that in the age of AI, we know that algorithm are really, really good at recognizing patterns and at guessing things. And that's what we apply. The question of privacy is very different. It's a yes, no. You can have privacy compliant probabilistic method and you can have non privacy compliant deterministic method. The question of access to data is really what data protection regulation structures, right? And so it's a question of like notice and choice. Tell me what you're going to do with my data and give me a choice to accept or not accept that transaction, right? Those are the fundamental pillars of data protection regulation everywhere in the world. Japan, Brazil, India, us, any state, Europe. It's always the same logic, notice and choice. And so it's basically a black or white, it's a yes or no, right? Can I use that signal? Can I use that piece of personal information, yes or no? If I can, it doesn't matter that I use it. Probabilistically or deterministically, I can use it. If I can't, whether probabilistic, deterministically, I can't. And it's something that we really use on our side at the very entry level of our infrastructure. When we get a piece of information, and again, we get 7 billions of them every day. When we get one piece of information into our system, it comes with a flag that says yes or no. If it says no, we don't keep it. If it says yes, we keep it and we process it. And then we can use it for the benefit of our customers. The reason why it says yes or no, because the user up front of that interaction has said yes or no. And so if the user said opted out, I don't want for you to process my personal information, we're not going to touch it. So it's not a. There is a how do you process that information? Right. Deterministically, probabilistically. And there is, are you allowed to process that information? If you're not allowed, it doesn't matter how you're going to treat it, you're not allowed. If you're allowed, it doesn't matter how you're going to treat it, you're allowed. And then it's a question of scale and precision. And how are you going to play along that curve of the scale to precision ratio?
- [34:20]
B
Okay. No, that's very helpful. Would you say though, that a probabilistic model for identity is, and I'm doing air quotes for the people not watching the video, that it's better than deterministic? Arguably, you don't really need perfect accuracy as a buyer. You just need to know that you're growing in the right direction or spending in the right direction. Like perfect accuracy feels like a fantasy to me. So I guess my question is, is good enough actually pretty much just good enough, like when it comes to digital identity.
- [34:55]
A
So I first, it's in the eye of the beholder, right? And we build different graphs and we apply different levels of precisions and recall for different clients based on their needs. Second, considering that deterministic is more accurate than probabilistic, can be proven wrong very often, right? I mean, john doemail.com is probably an email address that we get, you know, 10,000 times a day in our system. And if I use that to match IDs across devices or domains, I'm going to be wrong every single time. And on the contrary, if you're at home on a fixed WI FI connection dedicated line with a stable IP Address over seven days. Is that probabilistic? I don't know, but it's fairly accurate. So the trade off between accuracy and scale is really a question of outcome. What do you want to achieve? If you want to reach an audience of 10 people, maybe you'll be good enough making sure that you have the same email address on both sites to authorize the connections to be made. But no advertiser wants to reach 10 people. So advertising is a scale game. And the only way to get to scale is, is to apply machine learning and algorithmic recognition because there is not enough quality stable hard signals that can be reconciled deterministically to power the Internet and online advertising. And again, I think everybody understands or trusts the ChatGPT's answer when they ask a question. It's algorithm, so we shouldn't consider the algorithm are bad. And lookups, which is another form of another way to call deterministic matching, are good. I think there are some pretty good examples of algorithmic data processing that have convinced everyone that it was valuable and efficient.
- [36:56]
B
I will say though, I have asked perplexities some pretty basic questions and I've gotten wild answers.
- [37:02]
A
Absolutely not saying it works every time wild.
- [37:06]
B
For example, I put in a link just this morning to a thread that was initiated by Eri Paparo talking about transaction IDs and I was feeling a little lazy, so I put the link into Perplexity and I said can you summarize this for me? And it provided me with a three paragraph dissertation on Abilify, which is, I believe, a drug for depression. I was so confused and I said this has nothing to do with anything. They're like, well actually Abilify. And they just kept going on. It was, I was very confused. So sometimes it doesn't even make sense.
- [37:42]
A
Of having the importance of having good truth set to validate your model and retrain them constantly is undoubted.
- [37:53]
B
So now I would love to do a little lightning round, but these are the rules. So I only want a yes, a no, or a three to five second answer and no one, literally no one takes me seriously when I say I want to do a lightning round, they just answer at length. So I've never done this before and I'm sorry that you're going to bear the brunt of my frustration before we even get started. Well, I downloaded an app called the Buzz app and literally all it does, I'm going to open it right now is make a buzz sound. So if you go over, I'm going to do this. I'm opening the app now.
- [38:36]
A
Me after five seconds.
- [38:37]
B
Yeah, I'm going to buzz you after five seconds.
- [38:40]
A
I'm gonna talk even faster than I usually talk, which is going to be terrible.
- [38:44]
B
Are you ready?
- [38:44]
A
Everybody listens at 1.2.
- [38:51]
B
Or you can just say yes or no. We could just be really provocative. All right, so I'm ready with my Buzz app. Is Apple's approach to privacy better than Google's?
- [39:02]
A
No.
- [39:04]
B
Thank you. There's a lot of fragmentation in the identity space. How many alternative identity providers does the industry actually need? Just give me a number.
- [39:16]
A
I think three to five is where it's going to land.
- [39:19]
B
Okay. Do you believe that hashed email based IDs are sustainable long term? Yes or no?
- [39:28]
A
Sustainable, yes. Sufficient? No.
- [39:30]
B
You were so good at this. I feel so bad for scolding.
- [39:33]
A
I'm so afraid of the buzz started.
- [39:37]
B
Will identity graphs be obsolete in five years?
- [39:41]
A
No, there'll be more needed than ever.
- [39:44]
B
Will browser fingerprinting ever be fully eradicated?
- [39:49]
A
That's going to be impossible in five seconds. I hope so. Yes, but it's unlikely.
- [39:54]
B
Let's opt out of the lightning round for a second because I think that's actually pretty interesting. Then I have a few more lightning round questions. But talk to me about browser fingerprinting, because Google basically just did an about face late last year. They're fine with fingerprinting now. Apple supposedly is anti fingerprinting, but it feels like it just goes on, it just continues even though everyone's like, no, it's a bad thing.
- [40:18]
A
So I think what's bad is, I mean, fingerprinting is very different if you look at it from a sell side or a buy side standpoint. Right? Fingerprinting from, I mean, recognizing users right on the sales side. Whatever signal you use with their consent is completely fine. If you use the time of the day multiplied by their IP address, multiplied by a hash of their URL and you store that as your id, it can be considered fingerprint. But if you've said to the user, hey, I'm going to use information from your device to recognize you when you come back, is that okay with you? Yes, it's okay. Okay, great. Thank you. Thank you, Mr. User. You can call that fingerprinting. I think it's perfectly fine in the course of a compliant declared transaction to do that same thing as putting a cookie on their browser. Basically, if the buyer or a bad actor is listening into the bitstream, aggregating some of the signals that are being passed in the bitstream to then recognize people who are valuable audiences coming from CNN or People.com and then using that information to go retarget those users as people.com visitors on cheap inventory, effectively stealing intelligence from the publishers and making money by using that intelligence elsewhere, then it's bad. So the problem is not the practice, it's how it's used and how transparent it is to the publisher and to the user. If you do it non transparently without properly compensating the publisher and getting the agreement of the user, it's in some countries just plainly illegal to go to jail and very, very unethical. How you do it doesn't really matter. Doing that with cookies is equally unethical. Doing that with a combination of signal coming from the device isn't worse or better. It's equally unethical. Doing that without the user consent in Europe is illegal. Doing that if the user has opted out in the US is illegal. So that's the problem that fingerprinting presents. Fingerprinting on its own is not a problem. It's how you use it, how you apply it to different practices that becomes a problem.
- [42:41]
B
That was such an interesting answer and yet I wanted to buzz you. Only for comedic value, but then I didn't do it. Okay, so that was, that was interesting. And we're going to do a little bit more of the lightning round. What is one ad tech trend that you think people are ignoring?
- [43:02]
A
Who? Measurements. Oh, I don't think people are ignoring. I think they're not investing enough on it. I think if you don't measure, it doesn't matter.
- [43:11]
B
Interesting.
- [43:11]
A
And, and a lot of people focus on activating like targeting audiences and showing ads. I don't know. It's going to be more than five seconds you're going to buzz me. But if you don't measure it, it's useless. And so we need to spend a lot more time on measurement.
- [43:25]
B
This one's a yes or a no. So third party cookies weren't deprecated in Chrome. Can we assume that the gaid, the Google advertising ID won't be deprecated either?
- [43:36]
A
Yes.
- [43:39]
B
What is one ad tech trend you think people are paying too much attention to?
- [43:46]
A
What's the latest buzzword? I don't know. I'd say agentic AI to be a bit provocative. I don't think AI is a fad, but I think there's just way too much ink being spent to talk about something that is right now mostly slideware.
- [44:07]
B
What ad tech buzzword makes your skin crawl? You're like, there are just too many. I Don't know.
- [44:22]
A
There's so many ad tech buzzwords. If you had a list of five, I would definitely pick one. I would say maybe something around contextual 2.0.
- [44:37]
B
I hate when people plop a 2.0 on something that feels pretty cheesy.
- [44:44]
A
I know. It feels like contextual of a sudden was the best thing since sliced bread, as if it was rediscovered. I mean, it's always been important to put the ad in the right place, right? Like if you. If you advertise on a. On a billboard that nobody sees, it's just bad advertising. You know, if you put an ad on content that is inappropriate or unrelated or irrelevant or then it's just bad advertising. So contextual has always been important. It's not become more important now than it was before, and it wasn't less important when you could target users at greater scale because cookies were available more broadly. So, like, I think it's just a. Yeah, it's a false resurgence.
- [45:28]
B
Oh. Oh, no.
- [45:29]
A
You almost buzzed me.
- [45:30]
B
I almost buzzed you. But this stupid app is ad supported, so now I'm watching an ad I can't even hit the buzz button for.
- [45:37]
A
I mean, advertising. Advertising is a great business model.
- [45:44]
B
Okay, well, we are out of time, but there was one extremely important question I wanted to ask you to round us out. We spent a lot of time talking about cookies. I spent a lot of time writing about cookies. Cookies, cookies, cookies. Favorite cookie to eat. Let's just talk about delicious cookies.
- [45:59]
A
I'm a classic chocolate chip cookie guy. I can't get any better than that. Kudos if they are my daughters, because she's an amazing patisserie chef.
- [46:09]
B
Well, that's adorable. And I'm gonna end it there. So thanks for humoring me and I apologize for making you my buzzer guinea pig.
- [46:21]
A
No problem. Always a pleasure, Alison. Thank you.