00:00Sous-titrage Société Radio-Canada
00:06Hello Todd.
00:08Hi, thank you for having me.
00:10Thank you for being here with us at Cannes Lions Festival.
00:14Criteo is one of the few companies to have tested Privacy Sandbox,
00:20the tool from Google.
00:23What feedback can you share about this tool
00:27and why do you think that so few companies have experimented this tool from Google?
00:34Oh, well, I mean, there's a lot wrapped up in that question.
00:38Let me take you back in history a little bit.
00:41We had partnered with Google for almost five years now
00:46and really the intent behind Privacy Sandbox was to secure a more data safe
00:53and privacy safe way for consumers on the internet and brands who wanted to reach them
01:01to interact in a way that wasn't invasive.
01:04Okay, so the intent of Privacy Sandbox is really important.
01:08The challenge of putting a new privacy framework in place in an ecosystem
01:14where there are a variety of players, each one with slightly different interests,
01:19turned out to be pretty challenging.
01:21So our partnership with Google was a very healthy one.
01:26Obviously, a lot was learned during the course of those five years.
01:30There were some fits and starts, some starts and stops.
01:34Ultimately, a decision for third-party cookies to remain.
01:40What we learned was that a couple of really important things.
01:45One was that it is possible to use different approaches and methods
01:52to provide relevant product recommendations, what's core to our business,
01:57to consumers without a cookie being the endpoint or the key for targeting.
02:03That was very important.
02:05The second thing it did for us is it sped up our advancements in AI.
02:12Before AI became a big discussion, of course, it's all the rage here at Cannes this year,
02:18especially agentic, which I think we'll talk about maybe.
02:22But what Privacy Sandbox did for us is it allowed us to advance our use of AI
02:28specifically from machine learning to deep learning and inference and ways of making a great product recommendation
02:37matched to you or myself as a browser of the internet or on an app
02:43without having all the same data signals that might have come through a cookie.
02:48So from those two perspectives, it was an incredibly great journey.
02:53We also learned that, to your second question, that it was really difficult to balance the interests of parties that needed to win
03:04for this new framework to be put in place.
03:07In particular, there are two things colliding.
03:12One is for brands, especially brands that emphasize performance, meaning measurable sales lift.
03:20A product was sold, an incremental product opportunity was created.
03:25That's really, really hard to create on some of the fuzzy logic and signals that come out of deep learning.
03:32So that was one really important note.
03:35The other one was that not enough of a market was created for dollars to flow through to publishers
03:43that would be receiving advertising if Privacy Sandbox was working for a range of different marketers.
03:50So we saw both of those things.
03:52And the Google team, to its credit, really stuck with it, tried different things.
04:00And we were happy to work with them because the intent behind Privacy Sandbox was ultimately a good one.
04:07So you are saying that you are going to continue to collaborate with Google on these aspects,
04:14even if the third-party cookies are still…
04:17Oh, yeah. I mean, Google's a great partner to us.
04:20I would say what's uncertain is how to take the learnings of that period.
04:26A lot was invested during that period.
04:28I'm not sure everyone you talked to would say the same thing that we would think at Criteo about our investment.
04:34We did indeed get a lot out of it.
04:36It really helped shape our multi-pronged addressability strategy.
04:41And we'll continue to invest on that level.
04:44I think one of the things to keep in mind is when you're trying to do something with great intent,
04:51this is very product-focused, sometimes it's really good to be specific about the problem that you're trying to solve
04:59in a way that you can bring the industry along with you.
05:03So my urging to the Privacy Sandbox team was just with them today, was let's think really hard about what we learned in the last five years,
05:14and let's maybe pick one or two things that would be advancing the privacy cause,
05:20but really solving a problem that was a near and present opportunity or risk.
05:27Also, to be more specific, I'd love to see Google open up YouTube and put Privacy Sandbox in front of it for targeting,
05:36and specifically protected audiences.
05:39Another thing that's a little bit further out that I think is a really interesting consideration is Agentic.
05:46So if you think about how Gemini is now somewhat clawing back a lost share of search,
05:55and with I.O. last week, a lot came out that benefited Gemini, the DPI mode.
06:03There's a real interesting problem, which is how the memory of you and I as a user of Gemini
06:11could be used for delivering a perfect product recommendation.
06:16That's a different view on privacy than we saw with cookies.
06:20And do you think that Gemini, AI, Agentic could be integrated into Privacy Sandbox?
06:28Conceptually, a lot of the things that we were working on with Privacy Sandbox,
06:33including trusted execution environments and at least some early work to define private modeling is quite relevant.
06:43Now, whether or not we can and should pursue those things is something we're just at the beginning of the conversation about,
06:49but I think when you have a team that is as focused as that team is on making the web a more private place,
06:58then it's just a matter of finding the right spot to apply that passion.
07:04And so we're doing our best to be a bit more specific this time and I gave you the two examples.
07:10So we'll see.
07:11Let's skip this Atlantic subject.
07:14I think it's interesting regarding open web possibilities.
07:19What do you think that it would change in terms of targeting people on the open web, you know, through Gemini, through Agents, things like that?
07:31What will be the future with this kind of technology?
07:34Well, so the future is upon us. It's happening before our eyes.
07:40And there are a couple of things that I would observe are defining it right now.
07:46For the open web or for any domain that, you know, an agent can get to and index or call when a reasoning model is being used,
07:58you have a very different consumer experience because you may never reach the place where the information was originally being harvested to answer a prompt.
08:10So one of the things that needs to be worked out between content creators, big or small, and the agents which are providing an answer to a prompt,
08:21again, we think about product recommendations first, but there are lots of other things that people like myself are using OpenAI or Claude,
08:33OpenAI GPT or Claude for any of the other models.
08:38So you've got that piece of it, which is how is content going to be discovered, monetized, and delivered to a user?
08:48That's one side of it.
08:49That's one side of it. The other side of it is most companies that we work with, whether they're content creators,
08:55whether they're retailers or brands, are all thinking about how do I get a piece of this experience?
09:02How do I stand up an agent of my own so that I take the attention that I have and maximize it or extend it?
09:11And so I think you're going to see a lot more personal agent kind of interactions on retailer sites, certainly on publishers.
09:23I think Time Magazine was a really good example in the U.S. of someone who began to put agent flows together.
09:30I think you're going to see a lot more of that. But there are two very different worlds.
09:33One is taking the consumer somewhere else and giving the answer.
09:38One is maybe doing a better job of deepening the consumer experience on the site itself.
09:45And we have to see ultimately what we prefer. We're the ones deciding.
09:51Like where we decide we trust the answer most, where we find it to be most entertaining to do our work,
09:59whether it's our shopping or discovering the facts about something.
10:04What I see is that consumers are voting now. Is it search I'm choosing? Is it an open website I'm using? Is it GPT?
10:12And do you think that these AI agents are... Is it the end, in a way, of the open web?
10:22Is it going to create these world gardens from the big tech? And as Créteo, how can you exist in this environment?
10:34Well, those are two very big questions. But I have... I definitely have my opinion on this.
10:41So, first of all, the first answer is no. I mean, still content... And I sort of gave you an example.
10:48We're already watching behavior. While search traffic may be going down, you know, kind of people who are natively loyal
10:57to different open web points of presence or apps, they tend to stay there.
11:02So, like, we're not in a place where an agent is trusted for a perfect answer every time.
11:09And it's going to take time for us to get there.
11:12You can think about the last two years between, say, GPT-3 and GPT-5, which is coming.
11:19A lot of advancement. But still, we seem quite far away from the trust that's needed for a perfect answer every time.
11:29So, I mean, then you have to go with our own behavior. Like, I like to read things. I like to validate things.
11:37I consume things that a model's never going to give me, no matter how trusted, no matter how effective.
11:46There are certain things about my cognition that need to be fed by content, by things I'm looking at, by photos, by interactive experiences.
11:56So, it's very hard for me to believe that that's going to go away. In fact, I think they will coexist with one another.
12:03And so, we're trying to take a more horizontal view of Criteo. Just to shift to your second question, which is, you know, five years ago, we were still mostly a retargeting company.
12:18Mostly bottom of the funnel. We had a little bit of retail media just emerging. Today, we're a full funnel, cross channel, you know, shopper journey company.
12:28We're going all the way from a discovery ad, a product being discovered without intent, where we're leaned back in a social feed, maybe on CTV, maybe on the open web.
12:41And as soon as we're able to get the right fit in Agentic. So, we look at things very differently today than we did five years ago.
12:49And Agentic just really expands our view of the world, where retail media was brand new. You can think of it like this.
12:57When I started at Criteo five years ago, retail media was just a green shoot of success. Now, it's actually quite massive.
13:05I think Agentic will follow a similar pattern, maybe a little bit faster. But if you look across social, CTV, retail media, open web, and maybe add Agentic to it, it means that our aperture for cross channel just gets bigger.
13:23And so, it's exciting to me as a product person, because the way that you might choose to discover a product in an agent versus on social in the feed versus on a retailer website is quite different.
13:38And that means we sort of have to meet you or me as a consumer in a different way. And there's a lot of value we can add to brands, to retailers, to publishers in that regard.
13:52So, it's exciting. Thank you very much, Dad. You're welcome.
Commentaires