บทถอดความฉบับเต็มของ 51% Accurate: Half Your Identity Matches Are Wrong

00:05:27.839 So there are many challenges, and everyone that deals with it understands that they need to always be on top of regulation. 00:05:28.000 --> 00:05:32.560 For instance, in the US, there are several types of regulation, just in the US. 00:05:32.639 --> 00:05:32.879 Yes. 00:05:33.040 --> 00:05:34.160 And then you have Europe. 00:05:34.560 --> 00:05:43.759 They need to think of a long term, like they need the experience of the user in their website relevant, not just selling more, because eventually it will come back to them. 00:05:44.079 --> 00:05:45.839 Identity, privacy. 00:05:46.000 --> 00:05:48.639 We've been talking about it for a while. 00:05:48.959 --> 00:05:51.360 Why so central now? 00:05:51.519 --> 00:05:55.920 Is it the omnichannel is the driver, or what is what is change? 00:05:57.040 --> 00:05:58.160 You have to be compliant. 00:05:58.319 --> 00:06:15.120 At the end of the day, there are many moving parts that require a holistic solution because you might be finding a user belonging to one audience based on IDA, and then there is another audience which is on ID B. 00:06:15.360 --> 00:06:22.639 So the publisher needs to accommodate multiple identifiers, sometimes over 20, in a single auction. 00:06:22.959 --> 00:06:28.639 And we found ourselves as one of the value is reducing the hustle of operation. 00:06:28.800 --> 00:06:38.720 If you think of a publisher, if they need also to manage other than content, the different relationship with different identity, because every SSP has their own ID. 00:06:38.879 --> 00:06:40.480 Every DSP has their own ID. 00:06:41.040 --> 00:06:51.040 So I think what we found is that you need to have a very reliable house of identity that could provide you the solution that would fit your needs. 00:06:51.360 --> 00:06:57.199 Providing identity, but not limiting my needs as a publisher to one area or another. 00:06:57.439 --> 00:06:58.879 I don't want to peak. 00:06:59.120 --> 00:07:00.480 I want a flexibility. 00:07:00.879 --> 00:07:04.399 And I think this is very important in a technology event or of identity as well. 00:07:04.879 --> 00:07:20.000 That's also changing, as you pointed out, that that the different focus and the different questions publishers need to be asking themselves and the different stress on them dealing with the identity as we go across more and more devices. 00:07:20.240 --> 00:07:24.000 So how would you actually then describe intent IQ? 00:07:24.560 --> 00:07:34.000 So we evolved from starting on a device graph that we were uh selling to uh DSP as a service. 00:07:34.480 --> 00:07:45.519 And then we come to realize that not every entity that needs identity can ingest this large volume of data or to ingress the data. 00:07:45.680 --> 00:07:50.879 So what we we found is you need to build solution on top of identity. 00:07:51.199 --> 00:08:09.839 So for instance, in our case, we have a product called Beat Enhancement, and Bid Enhancement is all about helping to get the product

of identity very easy to use instead of hosting huge volume of data and hosting huge volume of processing units. 00:08:10.240 --> 00:08:18.879 they just need the solution that helped them at the right moment, Like we call it a millisecond before the bid leaves the auction. 00:08:19.360 --> 00:08:29.199 And that's where we found ourselves providing a solution on top of identity, the mechanism, accuracy, and everything that we we can uh discuss. 00:08:29.680 --> 00:08:44.000 But providing the solution, whether it's an attribution, whether it's onboarding, or for publisher, a service that connects in real time, making the technology very accessible. 00:08:44.240 --> 00:08:49.120 If like looking back five, six years, not every publisher can take the solution. 00:08:49.360 --> 00:08:58.399 He needs to host a big solution, and now you're taking like the last mile of identity, just matching real time and provide me the IDs. 00:08:58.559 --> 00:09:07.360 Today a publisher needs a set of IDs, and tomorrow they have a need for uh, for instance, ID that are based on deterministic login data. 00:09:07.519 --> 00:09:12.240 Okay, I have my ID as a company, as an identity with my own demand. 00:09:13.200 --> 00:09:19.120 The publisher doesn't need to pick or commit to one because there is no single bullet of ID that works. 00:09:19.519 --> 00:09:29.840 That's where we fit ourselves in a solution that is very flexible and it's like an identity agnostic with 60 different IDs in one solution. 00:09:30.240 --> 00:09:35.600 We talked about what matters more also in an omnichannel world. 00:09:35.759 --> 00:09:47.200 So a recent identity infrastructure 2.0 report um point out that a loan just hashed email to postal linkages, which is very basic. 00:09:47.919 --> 00:09:53.679 Averaged only 51% accuracy, which means 49% inaccurate. 00:09:53.840 --> 00:09:57.120 I mean, that is a startling number for me right now these days. 00:09:57.279 --> 00:09:59.200 I mean, how would you explain it? 00:09:59.279 --> 00:10:00.320 What do you make of that? 00:10:00.720 --> 00:10:08.080 But if you think about a very simple life that you say I'm going to buy an ounce of gold, right? 00:10:08.639 --> 00:10:13.919 So it looks gold, it's shiny, but what if it was like diluted 50%? 00:10:14.320 --> 00:10:15.440 Would you still buy it? 00:10:15.759 --> 00:10:16.320 Not at that moment. 00:10:16.879 --> 00:10:17.600 Not exactly. 00:10:17.759 --> 00:10:30.559 So it's like when you think about what you've just mentioned, there is a real problem of matching real deterministic hashed email to a person, not even speaking on a house for like 50, 49%. 00:10:31.360 --> 00:10:36.399 Think about the challenge to connect it to a person on an in-up moving outdoor. 00:10:36.960 --> 00:10:39.519 So the accuracy becomes really a challenge. 00:10:39.679 --> 00:10:49.360 We keep monitoring, and we see if you don't do the monitoring, you are not aware of the fact that it's like taking half of your dollar and throwing it away. 00:10:49.679 --> 00:10:54.639 So, how is it possible that publishers are really confusing? 00:10:54.720 --> 00:10:57.279 You know, I can match more because we can do more. 00:10:57.519 --> 00:10:58.720 We'll get to AI in a moment. 00:10:58.799 --> 00:11:00.720 You know, scale, it's all about scale. 00:11:00.960 --> 00:11:05.279 When actually it's more like, have I matched correctly? 00:11:06.080 --> 00:11:06.559 Exactly. 00:11:06.720 --> 00:11:14.960 So as a publisher, they have a lot of pressure to gain performance and numbers and management want to see growth. 00:11:15.279 --> 00:11:23.519 And as the the the media evolve and more lost of signals, they want to push for coverage. 00:11:23.759 --> 00:11:38.000 If as a publisher you are ignoring the fact that maybe 50% is wrong, then the experience of the ads that are being sold on your site would come back to you as a boomerang. 00:11:38.399 --> 00:11:43.919 Trying to maintain 92 to 95% accuracy, it's not an easy task. 00:11:44.320 --> 00:11:46.000 You're measured on your ad experience. 00:11:46.159 --> 00:11:48.960 That's part of your performance, that's part of your brand performance. 00:11:49.039 --> 00:11:53.600 If you are a brand involved in that as well, what is the expense? 00:11:53.679 --> 00:12:04.480 You know, what what is the number that matters most if it's not about the expense of keeping the customer, losing the customer? 00:12:04.559 --> 00:12:07.759 I mean, what is at stake if you're not matching correctly? 00:12:08.159 --> 00:12:16.159 First, on the advertiser buy side, they're actually getting less and paying more. 00:12:16.879 --> 00:12:19.519 So their performance because they bought the diluted gold. 00:12:19.840 --> 00:12:23.679 They they're like spread and pray, and that that's wrong. 00:12:23.919 --> 00:12:32.720 So publisher in the short term may gain value by saying, I don't care as long as the number grows, but eventually it's like what happened on MFA. 00:12:32.960 --> 00:12:36.720 MFA three years ago was like a jungle. 00:12:36.879 --> 00:12:42.960 Everyone was buying user from social and then selling ads, yeah, and now it stopped. 00:12:43.200 --> 00:12:51.759 So think what happened if all this idealist traffic on publisher will suddenly say, okay, let's take a look. 00:12:52.000 --> 00:12:55.039 Is this the real person that you are selling or not? 00:12:55.200 --> 00:12:56.960 And then the party will stop. 00:12:57.279 --> 00:12:58.639 It's like a music chair. 00:12:58.720 --> 00:13:04.159 Everyone is now playing in that, but once the movies the music stops, you're losing. 00:13:04.879 --> 00:13:14.559 So I think we as an identity vendor putting this accuracy and the understanding that we need to measure ourselves deterministically very important. 00:13:15.039 --> 00:13:16.720 Don't trust, measure. 00:13:16.960 --> 00:13:17.919 That's the takeaway. 00:13:18.240 --> 00:13:20.879 So let's stay with the buyer for the moment, right? 00:13:21.440 -->

00:13:30.320 Buyers out there, three identity companies are pitching and they're saying, hey, we have scale, we have reach, we have accuracy. 00:13:31.200 --> 00:13:33.200 What helps make the right decision? 00:13:33.279 --> 00:13:44.960 I mean, is there a non-negotiable here for someone who's really just starting to understand and wants to make certain that they have to prove what they offer? 00:13:45.440 --> 00:13:51.200 What they have, they have precious data that their customer brought them when they are buying. 00:13:51.279 --> 00:14:12.399 So if you take a true set and you want to evaluate three vendors, and you take a small set that you know for sure, and you can provide uh like a hidden test saying, okay, I'm giving you 5%, but in it's part of like 100% of data, and I want to see how you're gonna match this unique data. 00:14:12.480 --> 00:14:14.879 Like it's your own first-party data. 00:14:16.399 --> 00:14:41.120 It's be foolish not to evaluate based on your and it's very simple, just an offline test could demonstrate whether this vendor is saying what they are, and if this 5% that you know for sure is matched, whether it's uh a zip code, like you said, to a hashed email or mobile ID to other IDs because the the user purchased, they gave you all the information, evaluate before you decide. 00:14:41.679 --> 00:14:44.000 So you know this so well, you're up for the challenge. 00:14:44.240 --> 00:14:44.879 Yeah. 00:14:45.360 --> 00:14:51.519 What about buyers when they're thinking about, you know, because scale, scale is attractive. 00:14:51.759 --> 00:14:54.159 And they're saying scale versus accuracy. 00:14:54.240 --> 00:14:59.440 You know, I can live, I I talk with many publishers also that come on mobile group. 00:14:59.600 --> 00:15:00.879 What do you say to them? 00:15:01.360 --> 00:15:07.919 So I think this is a word that could come from the mouth of publisher because there are less the uh that's true. 00:15:08.000 --> 00:15:08.960 The buyers wouldn't say that. 00:15:09.200 --> 00:15:26.559 But the buyer wouldn't say that, and when it comes to buyer, if they're willing to spend, it's like I have a great campaign for McDonald's or for a brand new iPhone and I want to publish in New York, but actually my budget goes in India and I'm not selling there. 00:15:26.720 --> 00:15:28.639 Is it a good decision to make? 00:15:28.799 --> 00:15:29.759 I don't think so. 00:15:30.159 --> 00:15:30.399 Right? 00:15:30.559 --> 00:15:37.840 So I think from the from the buy spot buy side perspective, you need to measure cross-device attribution. 00:15:38.639 --> 00:15:51.759 And by the way, some of the budget are being avoided from ID list and uh publishing on premium iPhone user because of that problem. 00:15:52.480 --> 00:15:58.480 So you either spend the money unwisely or you avoid it completely because you're not measured. 00:15:58.559 --> 00:16:04.960 And we say if you have a good identity spine, then you can get two words. 00:16:05.120 --> 00:16:06.799 There is a way, you need you need to work. 00:16:07.360 --> 00:16:08.240 You need to do some work. 00:16:08.639 --> 00:16:08.879 Okay. 00:16:09.279 --> 00:16:10.960 Well, look at the publisher now. 00:16:11.120 --> 00:16:15.679 I spoke about how publishers can deal with a little bit of inaccuracy. 00:16:16.000 --> 00:16:24.639 But you know, do they understand what they need to be thinking about before they start to measure value? 00:16:24.799 --> 00:16:31.440 It's not just the scale, you know, it's engineering um a solution, engineering their resources. 00:16:31.600 --> 00:16:39.200 So they're looking at an auction and saying, yeah, more bidders, better bidding, this is great, better use of inventory? 00:16:39.360 --> 00:16:40.080 I don't know. 00:16:40.559 --> 00:16:43.039 So it's not always more is better. 00:16:43.200 --> 00:16:48.879 We see that many publishers reduce the amount of bidders, like in pre-bidded called adopter. 00:16:49.200 --> 00:16:56.480 So they reduce and they still keep the same value or reducing the ad TXT of the allowed publisher. 00:16:56.720 --> 00:17:11.039 The publisher needs to think of the future needs that he's not seeing right now and the flexibilities because what we're seeing is that many publishers they start on the web and they see great success, but then they go into in-app. 00:17:11.440 --> 00:17:18.480 Does your vendor can keep running the same service seamlessly on in-app or when it moves to CTV? 00:17:18.640 --> 00:17:25.200 Privity is evolving, yeah, Amazon APS is evolving, and all of those markets need identity. 00:17:25.440 --> 00:17:45.839 So you you think of your identity vendor as a solution that thinks for you, the doing the work of easy of integration, whether it's web, whether even sometimes uh an aggregator expand and saying, I'm gonna get my my traffic on the server side. 00:17:45.920 --> 00:17:54.880 So does your provider, does your identity partner can help you server to server without being on the page, code on page or on the SDK? 00:17:55.039 --> 00:17:57.599 Those are the questions that you need to think. 00:17:57.839 --> 00:18:05.200 Looking for the long run, I'm now on I have my content, it's on the web, but I'm gonna expand, I'm gonna be in ENAP, I'm gonna be on CTV. 00:18:05.359 --> 00:18:08.799 I want my identity vendor to be as flexible as myself. 00:18:09.039 --> 00:18:14.880 What are you evolving the company in the direction of to be able to do exactly that, to be that flexible? 00:18:15.200 --> 00:18:23.920 We see a lot of trends on adopting AI solution as part of your ability to move faster. 00:18:24.319 --> 00:18:30.240 Even a simple task as translating the site to a taxonomy with LLM becomes more accurate. 00:18:31.039 --> 00:18:36.000 We see ourselves more going into the in-app world. 00:18:36.480 --> 00:18:43.039 So far, there were solution on server-to-server API, which is easy, but we see the needs for SDK. 00:18:43.359 --> 00:18:46.799 And there are multiple

mediation platforms that need the solution. 00:18:47.440 --> 00:18:56.960 And also when looking about mobile, because the AI is hurting the web publisher and it's not affecting the in-app. 00:18:57.200 --> 00:19:08.240 So, what we're seeing is that there is a lot of demand for uh identity and suite of solution like onboarding and activation and identity, and also CTV. 00:19:08.319 --> 00:19:26.880 So we see ourselves as expanding our buy-side solution, yeah, but still maintaining a very um hard KPIs on accuracy when it comes to publisher on flexibility, judging you by performance and revenue. 00:19:27.440 --> 00:19:30.559 They have a lot of pressure with AI to bring the best. 00:19:30.720 --> 00:19:36.720 So if you need to do the best, you need to be equally investing on all channels. 00:19:36.960 --> 00:19:43.519 And we see ourselves strongly on CTV, strongly on mobile, and expanding to GEOS. 00:19:44.079 --> 00:19:46.960 The idea is scaling with intelligence. 00:19:47.200 --> 00:19:53.200 One part of that is AI, which we addressed, but it's also scaling overall to meet the needs of the market. 00:19:53.759 --> 00:19:58.960 We understood that Germany is uh a very important market. 00:20:00.079 --> 00:20:08.960 Even though the publishers are more conservative, there is a demand for identity and there is a demand for understanding how to overcome idealists. 00:20:09.119 --> 00:20:17.119 We even hired uh um a general manager and open a branch, local branch here in Germany, the same as in France. 00:20:17.200 --> 00:20:36.799 Yeah, because we believe that you have to be in the market in order to influence, and that's why we're coming here to spread the news, to share our success stories within Germany, because we believe the German market is a few years behind with adopting technology, but it's getting there. 00:20:37.200 --> 00:20:40.559 The publishers who've been on my show today telling me about that. 00:20:40.640 --> 00:20:44.400 But I'm interested what questions they're asking in Chen IQ. 00:20:45.200 --> 00:20:53.759 So they're they're asking how to maintain privacy without hurting their performance. 00:20:53.920 --> 00:21:01.359 Uh, there is a process of reconsent where anytime you add another vendor to your CMP, you have to do a reconsent. 00:21:01.519 --> 00:21:01.680 Yeah. 00:21:01.920 --> 00:21:03.839 And that's one challenge that they ask. 00:21:04.079 --> 00:21:09.039 The other challenge is what are the demand sides that are working for you? 00:21:09.279 --> 00:21:12.000 Okay, the Europe, the the US market is huge. 00:21:12.160 --> 00:21:14.160 Are they the same in in Germany? 00:21:14.319 --> 00:21:29.119 But we come to realize that we need to penetrate the local SSP, and that's what we're doing with local SSP, because many publishers here in Germany have a split of maybe 50%, which are there, those buyers do not exist in North America. 00:21:29.440 --> 00:21:31.839 CTV is always intriguing for me. 00:21:32.000 --> 00:21:33.039 What does that look like? 00:21:33.759 --> 00:21:39.440 So CTV is challenging because this market is yet late to the game of identity. 00:21:40.240 --> 00:21:46.960 But there is a need to understand exactly what does it mean a household share device, a Wall Mountain device. 00:21:47.119 --> 00:21:50.480 They're holding their phone while they're watching and they're installing stuff. 00:21:50.640 --> 00:22:16.160 So for us, is working with uh um attribution companies, providing them identity solutions in the form Of a device draft that has also the ability to connect Ruco device, Samsung device, all those uh channels that have their own platform, their own technology, but the user is expect exp exposure more to CTV. 00:22:16.480 --> 00:22:18.799 Now it's kind of a house level match. 00:22:19.119 --> 00:22:28.640 But we see a trend where CTV is moving also to um devices like iPad and phones where people take consume their content there. 00:22:28.880 --> 00:22:29.279 Yes. 00:22:29.519 --> 00:22:38.160 But we're trying also to accommodate the need for those platforms, and it's more about I think attribution on a house level. 00:22:38.880 --> 00:22:48.400 And that's where we found ourselves uh trying to understand how can we educate the buyers on CTV to ingest the signals. 00:22:48.720 --> 00:22:51.680 Yeah, let's just let's just talk about it plainly. 00:22:51.839 --> 00:22:55.359 The change and the challenge that AI presents. 00:22:55.680 --> 00:23:06.720 Because when you're thinking about it from a public perspective, your customer could be your biggest customer, maybe AI bots. 00:23:08.000 --> 00:23:22.000 Yes, so luckily when we analyze the signals and deciding whether the criteria to form a household and then a person has multi-dimension criteria. 00:23:22.240 --> 00:23:35.839 So obviously, a bot is being detected, not always, but it's challenged whether it's someone scamming the site, and now you are trying to identify a bot that is doing something for the person. 00:23:36.240 --> 00:23:43.440 I think still we see that publisher getting hit and the the amount of page views are shrinking. 00:23:43.839 --> 00:23:55.039 So it's even more challenging to yield better performance on the on the share that's still left for the real person that's visiting the site. 00:23:55.200 --> 00:24:06.160 We identify the user agent when it's coming on a bot, we identify their IP address, we identify a lot of signals when it's not a person, but it's not completely sealed. 00:24:06.640 --> 00:24:12.559 So to be honest, it's like an ongoing battle once AI take over more and more. 00:24:12.799 --> 00:24:26.880 But then you ask yourself, how about providing for those giant also identity to say, okay, you know that person, but I can know other stuff that the user actually still surfers, still going in the website. 00:24:26.960 --> 00:24:33.759 So we think that our next job would be getting identity to those big AI chats. 00:24:34.079 --> 00:24:42.880 We see what the user are

doing on multiple sites, multiple vertical, and I think we could also add supplemental information even to AI. 00:24:43.200 --> 00:24:50.240 And to your point, if you think about it, you know, the value of the gold now, the human gold, will be much more valuable. 00:24:50.400 --> 00:25:04.559 And probably I wouldn't say the pressure, but the opportunity for buyers as well as publishers to get that right, to understand the difference between matching and matching accurately. 00:25:06.000 --> 00:25:10.799 So matching and matching accurately, I think it raised the stakes, I think. 00:25:11.200 --> 00:25:18.319 You have an auction right now and you need to provide something rather than checking what is the accuracy of what you've provided. 00:25:18.559 --> 00:25:42.240 And we we said that at the early stage, publisher has a lot of pressure to provide value because if they're shrinking and they still need to show growth, I'm not sure yet that all of the publisher are aware of the accuracy issue, but once chargeback is coming, it's going all the way from the top, from the advertiser to the DSP, SSP, and back to the publisher. 00:25:42.559 --> 00:25:42.880 Okay. 00:25:43.039 --> 00:25:48.319 An identity vendor tells me they have enormous scale. 00:25:48.640 --> 00:25:57.519 What's the first question that I need to ask in order to determine if that scale, if that more is better? 00:25:58.640 --> 00:26:08.079 Evaluate and measure your own first party data against what the identity vendor is offering, because then you'll gain control versus what they're telling you. 00:26:08.319 --> 00:26:10.799 That's testing that testing that 5%. 00:26:11.359 --> 00:26:25.279 Yes, take your 5% of your first party data that you know their email, you know their identity, they already visited, blend them without telling them, and send it over and see what the results or best 5%. 00:26:26.240 --> 00:26:28.400 But what's really the number? 00:26:28.559 --> 00:26:34.160 What's really the cost of inaccurate identity? 00:26:34.640 --> 00:26:46.880 So hurting the experience for the long run, being exposed for chargeback, the ability to stand for your content and the audience that you are selling. 00:26:46.960 --> 00:26:48.960 Because at the end of the day, you are giving it free. 00:26:49.359 --> 00:26:55.599 Understand what is the accuracy, understand what are the sources of this data that is being fed into your streams. 00:26:56.079 --> 00:26:56.319 Okay. 00:26:56.480 --> 00:26:58.000 Well, it all comes down to trust. 00:26:58.160 --> 00:26:58.880 We talked about it. 00:26:59.039 --> 00:27:00.000 Measurement, trust. 00:27:00.160 --> 00:27:01.279 We talked about the metrics. 00:27:01.359 --> 00:27:14.640 So in identity, trust has to be what trust is something you earn by showing a real value for long-term relationships. 00:27:14.799 --> 00:27:21.279 You're making sure to stay high on the numbers because at the end of the day, it's an it's a it's a number of profits for the publisher. 00:27:21.599 --> 00:27:26.319 But adding a solution for advertiser and buy side is kind of creating a holistic. 00:27:27.039 --> 00:27:44.000 So keep on the guard for the changes, but don't lose accuracy, don't lose reliability, because you need to earn the trust that you're not bluffing, you stand for what you're providing, you're not rounding corner. 00:27:44.799 --> 00:27:53.680 If you want to keep for the long run, keep straight, keep technology, and keep research always on top. 00:27:54.319 --> 00:27:56.160 And that's for any technology vendor. 00:27:56.319 --> 00:27:56.960 Keep research. 00:27:57.200 --> 00:27:57.599 I love it. 00:27:57.759 --> 00:27:58.880 But it was a great interview. 00:27:59.119 --> 00:28:01.279 May I thank you for it, thank you so much. 00:28:01.519 --> 00:28:02.319 Thank you, Peggy. 00:28:02.400 --> 00:28:03.519 It was great.