Commerce

Podcast: RecSys and internet commerce (with Michael Komasinski)

By Eric Benjamin Seufert

On this week’s episode of the podcast, I am joined by Michael Komasinski, the CEO of Criteo, to explore the rapidly evolving landscape of agentic commerce and the critical role of recommendation systems in the AI era. We delve into how Criteo is positioning itself as a commerce intelligence layer for AI assistants and the technical distinctions between large language models and purpose-built recommendation engines. Among other things, we discuss:

  • Criteo’s recently announced advertising partnership with OpenAI
  • Whether agentic commerce will transition from assisted shopping to fully autonomous purchase decisions without human oversight
  • How recommendation systems based on purchase data outperform large language models in providing accurate product discovery
  • If retailers will eventually trust AI agents to manage complex fulfillment and brand trust in conversational environments
  • Why the integration of semantic language models and high-volume reward algorithms defines the future of digital commerce
  • Criteo GO, Criteo’s automated advertising platform
  • How the partnership between specialized advertising technology and generative AI platforms will reshape the global discovery layer

Thanks to the sponsors of this week’s episode of the Mobile Dev Memo podcast:

  • ⁠⁠INCRMNTAL⁠⁠⁠. True attribution measures incrementality, always on.
  • Xsolla⁠. With the Xsolla Web Shop, you can create a direct storefront, cut fees down to as low as 5%, and keep players engaged with bundles, rewards, and analytics.
  • ⁠Branch⁠. Branch is an AI-powered MMP, connecting every paid, owned, and organic touchpoint so growth teams can see exactly where to put their dollars to bring users in the door and keep them coming back
  • Voyantis. Voyantis uses predictive AI to transform customer value into high-impact signals that boost ROAS across Google, Meta, and more.

Interested in sponsoring the Mobile Dev Memo podcast? Contact Mobile Dev Memo advertising.

The Mobile Dev Memo podcast is available on:

Transcript

Eric Seufert: Welcome to the Mobile Dev Memo podcast. I’m your host, Eric Seufert, and I’m joined today by Michael Komasinski. Michael, welcome to the podcast.

Michael Komasinski: Hey, Eric, thanks for having me on.

ES: Well, it’s good timing, because Criteo’s been in the news recently, so I’m happy to have you on during this period. Before we dive into our our conversation, which will unpack all things agentic commerce, can you please introduce yourself to the audience?

MK: Yeah, sure. So Michael Komasinski, CEO of Criteo, been in this role for just a little over a year, and what a year it’s been. From sort of pulling back from cookie deprecation to, you know, changes in the retail media landscape, the emergence of agentic commerce, and now as of this week on Monday, ads in OpenAI or in ChatGPT. So it just keeps going faster and faster, and it’s awesome to be in this space right now.

ES: And prior to Criteo, you were the CEO of Dentsu.

MK: Yeah, so I spent really kind of 10 years between Merkle and Dentsu, and you might know Dentsu acquired Merkle, but yeah, I was the CEO of Dentsu Americas and then also ran global data platforms for the group, and before that was the global CEO of Merkle where I’d spent eight years between Europe and the Americas and then eventually running the whole company. So it was a great run, a lot of great colleagues back there, and now they are an important partner of ours. So everything goes round in this industry.

ES: And then kind of a couple hops before that, you you’d spent quite a quite a stint at Nielsen. Is that correct?

MK: Yeah, yeah. I was at Razorfish, Nielsen back in the day, which was a great grounding in kind of data analytics. I always like to say like I Nielsen’s kind of where I grew up as as an executive. Just a very rigorous company, really great on sort of product development and just the way that they run the place year to year. I was there during the take private when Dave Calhoun took over. So it was a really transformative time in the company back then.

ES: You’ve touched kind of a number of of different, call it, vantage points across the digital advertising landscape and and kind of the advertising landscape more broadly. So your perspective is is fairly multifaceted, I would say.

MK: Yes, it is. It is. But you know, as you do, you kind of find your way to the right place in a career, and I’m really happy with where I’ve ended up here. I believe in today’s age, being platform side at a platform like Criteo that has the data assets that we do sits in these kind of fast currents like performance and retail media, I think we can accomplish a lot here. And so yeah, I think I’ve yet again sort of found my way to the right place at the right time.

ES: Well, the time is the era of what what people are calling agentic commerce, and I’m I’m really happy to to have you on to talk us through your perspective on that. So Criteo recently launched the agentic commerce recommendations service for AI shopping assistants. Can you kind of just talk me through that? What is that? You know, what’s the kind of commercial model? How does it feature in this in this landscape?

MK: So you’re right, we launched that a couple of weeks ago, and and you can think of it as like a commerce intelligence layer for AI assistants. And what it does is it powers product recommendations inside of conversational environments with high-fidelity purchase-oriented signals. And what’s interesting, as we’ve brought it to market, we do side-by-side testing offline against LLM environments, and we see on average a 60% uplift in product relevance and accuracy. The reason we launched it is we really wanted to demonstrate the value of our access to real-time purchase data and our AI models to turn that into relevant recommendations. And I think it’s something that you’ve written about as well in terms of the underlying models that power these different systems, right? So LLMs by definition are semantic models and they’re great at language, but RexSys models need to be built on data loops and reward algorithms at high volume and scale. We very much believe in that, and the Criteo backbone is really built on a RexSys platform. And so this service is really a way to sort of give a front end to that in a way that can be tested in a partner environment and allows us to, you know, sort of prove out the efficacy of that data set. So it’s super interesting, right? And we believe the future of great commerce recommendations in discovery platforms is going to be powered by a hybrid integration of semantic platforms and RexSys platforms. And so again, this is our sort of way to push that agenda, put our capability out there, make it testable, accessible, and it’s early, so we’ve not focused as much on the monetization path yet. But eventually that could be SaaS, that could be pay-per-query, and more importantly, it could support both ad and non-ad use cases. But right now it’s just about proving effectiveness, validating use cases, and driving partner engagement. So monetization will follow all that.

ES: Got it. Because then that was going to be my next question is how do you get paid for this? I guess the phrase agentic commerce, I think is is it’s almost like AI in a way, because it can be it can be so vague and it can mean so many different things that it might it can be unhelpful in some cases. So so when some people talk about agentic commerce, and this is kind of I’ve pushed back pretty forcefully on this idea taking shape, they’re talking about I’ve got an agent and it’s just buying stuff for me. Like in the background. Some stuff’s going to show up on my doorstep tomorrow, and I’ll be pleased that it did, but I will have no knowledge that that was purchased on my behalf. There’s just some agent, you know, fulfilling what it it it anticipating my needs and fulfilling them with my credit card, right? Then there’s there’s so that’s kind of the call that like the the robot buying stuff model. And then there’s the model that OpenAI had had sort of implemented which is the instant checkout model where they’re going to allow retailers to slot into your your chat and just based on what you’re discussing, it’ll suggest products that are relevant and then, you know, you’ll check out right there. That they they sort of just it was it was information reported yesterday that they’re probably walking that back. They’re going to do that more in the underlying apps that they’re onboarding onto their apps platform. And I had pushed back on that idea too just because I said that’s not going to work as well as ads does. And it sounds like what you’re what you’re and let me know if I’m, you know, and I’ve read the press release and I understand, you know, how you’re positioning this, but just kind of, you know, feel free to like flush it out more. This sounds like this is RexSys for the the chatbot environment. But but it sounds like this really could be RexSys in in a number of different environments where there’s some sort of agent being invoked. Is that is that am I kind of understand that correctly? So there’s there’s there’s advertisers, there’s there’s retailers who would like to have their stuff recommended where it’s relevant and where it’s like additive to to some interaction, and that’s what you’re facilitating. Is that is that the right interpretation?

MK: Yeah, no, I think you I think you said it well. Um, yeah, it’s basically where the fidelity of product recommendation is an important use case to the overall experience, right? That’s probably the simplest way to describe it. But look, I’d love to go back on a couple of the points that you started with there that I think are super interesting and and a good way to like sort of get foundational. I completely agree with you on the autonomous versus agentic nomenclature. We also are not big believers in the autonomous use cases at Criteo, and that shapes then kind of our worldview of how we think the ecosystem evolves, right? Because we see agentic platforms really more as augmented or assisted, right? Like the word agentic gets kind of used it gets used in confusing ways. Sometimes people mean agentic as autonomous, sometimes people mean agentic as in augmented or assisted. We definitely subscribe to the latter definition. So like you, we are skeptical of the autonomy use cases. As a side note on that, you you might like this, so it’s become one of my favorite sort of dinner party questions like at work events. I’ll go around the table and I’ll ask people sort of what they will or won’t do when it comes to autonomous commerce. And I’ll start very deliberately with that autonomy word. And two funny things kind of happen. One, everybody automatically defaults to the toilet paper reordering question, right? It’s unbelievable how consistent this is. And then everybody always ignores the autonomous rule of the game, and they immediately go to augmented and assisted and start talking about how they would use agentic to do these things as long as they get to approve the final purchase or the final step or whatever. I’m like, wait, you’re not playing the game the way it was set up. But nonetheless, it is remarkably consistent when you ask people their sort of personal, you know, preferences around that. But I agree with you, I’m not a big believer in the autonomous workflow. And again, that that shapes our roadmap, which is how do we surface great products so that people get a better experience because they’re still going to have agency in that experience and also that there will be customer journeys that will continue to touch on multiple formats. They’ll land in e-commerce, they’ll land on publishers to do deeper research on cited sources, and they’ll ultimately do shopping and basket building in brand contextual environments. And so all those things kind of shape our roadmap and how we are moving the business forward.

ES: I like that rhetorical trap there. I’m going to use that because, yeah, it’s pretty easy to drift outside of the autonomous use case, because I think in theory people love the idea of the Jetsons-made robot just buying you a bunch of stuff, but yeah, the that there’s sort of like a bright red line past toilet paper and paper towels and air freshener refills.

MK: It all sounds good until your small closet in your New York City apartment is overflowing with stuff that you haven’t used yet, right? Or you know, a new car shows up on your front porch or your your driveway and so I like that rhetorical trap. But I think also like the way this is positioned, I mean, you get to be friends with everybody, right? Another kind of point I made in this piece that I wrote, Agentic Commerce is a Mirage, was that, you know, if you don’t get Amazon on board here, it’s it’s a non-starter. And I mean, people push back on that, but I I still think it’s true. I mean, it’s 25% of US e-com. The but in that sense, if you’re sort of like RexSys for everything and in these contexts where traditionally, you know, RexSys hadn’t been applied like the contextual use case in a chatbot, well then Amazon could be your friend, theoretically anybody could be your friend because they want to have their stuff recommended. Is that right? Am I thinking about that right?

MK: Yeah, that is true. That is true. And as you think about where where multiple people can be friends, you know, Amazon’s obviously got a great commerce graph, but so does Criteo and they are distinct, actually. So our graph has intelligence about people’s interactions with products in different environments and different contexts than what Amazon has in their graph. And so those two data sets are the two most powerful commerce data sets in the industry. And yeah, depending on sort of the surface that you’re trying to uplevel, they absolutely could sit side by side.

ES: Talk to me about this kind of playmaker role if we would move into this agentic purchasing paradigm. How do you maintain the playmaker role if if there be agents shopping versus versus the human doing the shopping? Does that change at all? Is that a meaningful difference or or no?

MK: Well, I mean, I think it I think it depends on the level of autonomy, but in the worldview that we prescribe, and I think you could see this in the change that ChatGPT made like you said this week to pulling back on the instant checkout platform, humans want to remain in control of a lot of the purchasing decisions. And I think a lot of that’s based on the fact that retailers continue to own fulfillment, trust, brand contextual shopping environments. And so with that, there are going to be multiple touchpoints in customer journeys. And so, you know, jumping ahead a little to, you know, ChatGPT’s focus on ads and being the discovery surface, ChatGPT scales discovery, Criteo optimizes downstream cross-channel conversion. I think it’s kind of the simplest way I’ve been able to sort of describe the the the partnership there. And just to maybe sort of go more broadly, even as AI assistants like narrow options, somebody still has to rank products by likelihood to convert. Somebody still has to measure outcomes across channels. Somebody still has to verify purchases and optimize towards ROAS. And and that’s our lane, right? We help determine what actually sells and we help drive and optimize and measure that cross-channel.

ES: You mentioned that I write about RexSys a lot, and I think one thing that people don’t really appreciate is that ads is applied RexSys, right? I mean, it’s just a very specific application of RexSys, really. And that’s where a lot of the AI research in the the sort of cutting-edge research that’s happening now and being monetized, this is what drives me insane is, you know, you read these articles that AI is it’s there’s this bubble and it’s not really driving any sort of material gains for these companies that are deploying all this money against research. And I’m just like, do you not do you not understand the ad space? I mean, this is every this is everyone is pointing their their research bazooka at this because this is where all of the frontier research is yielding these incredible gains. And it’s just it’s so frustrating because it’s like, you just just listen to these earnings calls, where do you think these gains are coming from? They say it. It’s really frustrating. But but talk to me about so this kind of idea is like, well, we’ve got, you know, some some big set of candidate things that we could show to this person, right, and then we’ve got to whittle that down. Talk to me about the where maybe the line blurs or where RexSys becomes ads targeting, right? Because I think there’s a specific moment where where that distills into ads targeting. This is ads targeting. We’re targeting ads now, but fundamentally this is like a RexSys algorithm or, you know, the what we’re the the system that we’ve built does that at scale. Talk to me about where the baton gets handed off, right, and and when that becomes ads targeting.

MK: Yeah, well, to be clear, for for us, there’s like 80% overlap between these things, because that’s how Criteo drives performance outcomes. Right? So it’s really a great big prediction engine that’s looking for the optimal impressions to serve across whatever channels we’re pointed at to drive the campaign outcome that that’s in question. And so, you know, we’ve been sort of doing this for 20 years. It’s a little known fact, actually, that company was founded 20 years ago and believe it or not, when Criteo first started, it was a recommendation engine for DVDs, believe it or not. And and they couldn’t figure out a great revenue model for that 20 years ago and quickly pivoted towards the retargeting use case on the open web back in 2005, 2006. And that’s what the company has its origin. So back then that was all machine learning, but you could imagine over 20 years like that prediction engine has just gotten bigger and more powerful. We learned how to normalize SKUs, learned how to build up shopper graph intelligence, and so again, the recommendation system or the the commerce recommendation service is just a like a front end on the RexSys system that underlies the whole company. And you know, I mean, I guess if you look if you really get specific and look at the similarities and differences, they both rely on intent signals, both use predictive models, they both aim to match demand with supply, but maybe the differences would be traditional ad targeting, non-Criteo, would be like audience-based. Agentic recommendations are more like product-level decisioning inside of a live interaction. I think that’s a really key difference. And, you know, Criteo’s always been a lot less about placing an ad impression as it has been trying to drive to an outcome.

ES: Got it. And and so, you know, these things converge kind of depending on the context, right, and, you know, obviously like there’s tremendous amount of overlap. And a lot of the original research, like if you go to, you know, YouTube developed the two-towers framework, for instance, and that was 2016, right, so that was that predates transformers. But that was just for video recommendations. And then a lot of the ad stuff got was built on that, which is just the kind of dual encoder architecture which has, you know, become a lot more complex now. But so we’ve got Criteo is kind of call it RexSys for for all, right? I mean, is is ACRs the acronym you’re using? Agentic Commerce Recommendation Service? Or is that a thing? Is that a thing?

MK: No, it’s funny, we haven’t acronymmed it. It actually still gets spoken to in its full glory, but I’m sure it’ll find an acronym soon and maybe it will be ACRS. I don’t know.

ES: Maybe that’s a that’s the difference. That’s the difference between RexSys and ads targeting. If if you articulate the whole phrase, the whole name, it’s RexSys, if you acronym it, it’s ads targeting. But so let’s say that ACRS, you call it like it’s RexSys for anybody. Anybody can get the benefit of that in their own environment, right, in their own in their own product, in their own in their own product context and get these these products recommended to them. Is that kind of how you’re viewing this? Is that the how you slot it into the landscape?

MK: Yeah, yeah. So like let’s take a couple of examples, right? So like if you were to take travel, like you could ask an assistant for a long weekend in Barcelona, like it could summarize options, but like which flights get surfaced, right, which hotel, which package is most likely to convert? How is any of that based off of actual price availability or real demand signals? Like that’s the same optimization problem that we solve in retail. Or take like restaurants, right? If you asked for the best Italian place nearby, the assistant can list options just by retrieving things across the open web, but like ranking them based on likelihood to book, real-time availability, historical behavior patterns, like that’s decisioning and that’s RexSys, not sort of semantic or or or, you know, real real augmented retrieval augmented generation. And so on and so forth, right? And it all comes back to having high-quality structured data, real performance signals, and then being able to optimize against outcomes at a high scale so that you get data loop learning. And yeah, just to sum it up, like if there’s a decisioning structure behind in structured data behind it, like this model can apply to bring relevancy and personalization to any user experience.

ES: Can you share examples of companies that that you’re working with already, like just to just to give a sense of like where where this is being applied?

MK: So we’ve kept a couple of them confidential just just because that’s the way that partners have asked us to do for right now, but we are in testing with two large platforms. Those tests keep getting progressively more sophisticated and I think continue to prove the uplift that we’ve shown in our own offline testing. And I think that that partner list will expand, you know, it could be commerce platforms that are starting to get into, you know, targeted product recommendations and ads like in the say like the post-purchase space like checkout that needs a service to like create the right recommendation out of a product catalog or a feed. So we’ve got I think a few different paths, but right now it’s the two large partners that we’re testing with and and we’ve kept those unnamed for now.

ES: Where does this scale to in terms of like product categories beyond I mean e-com is kind of the obvious I mean it’s obvious one, right, but where else does this scale to? Like what other product categories? What are the verticals?

MK: Yeah, so I mean, I think it could go in a lot of directions, right? Like think about like financial services. If somebody is looking for say like best rewards credit card for travel, right, like that’s a that’s a structured comparison problem and you could optimize based on profile, intent, probability of approval or activation, and that engine could play a part in that. So yeah, I think there’s a couple of different examples, but right now it’s probably more directed at partners that own sort of high-volume use cases where product is essential.

ES: Got it. And then but I mean are you what about the modality? I mean is are you thinking like web kind of is the starting point or it could be in consumer apps because I mean I guess it’s yeah, it wouldn’t be we wouldn’t be tied into any specific format there. Is the commonality here because I mean consumer adoption of of your chatbots is is obviously been, you know, really rapid, right, and I think that has become kind of a new functional norm in terms of doing research, right? I mean that’s just that’s kind of what you do now. Whereas before it would have been just a query-based search, now it’s probably more of like a conversational-based chatbot experience and I think you’ll see a lot of those those use cases proliferate to be vertical specific, right? So like I you could imagine when you’re buying a car going forward, it’s going to be through some sort of conversational interface because like here are my requirements and, you know, give me give me some options, right? Is that kind of the the interface that you’re looking to target? I mean that it’s it’s it’s really is meant to be this sort of semantic chat-based interface that is best accommodated.

MK: Yeah, exactly. And the starting point clearly is anything more tied to our current catalog and shopper graph, right, which is going to be more in the kind of fast-moving consumer goods categories. It could scale to automotive and things like that architecturally, but that’s obviously not where we’ve got a lot of our IP currently, just based on the legacy of the business and where we have those sort of high volume of advertising situations. Right, so but the idea being like kind of the broader idea is that like the way I do shopping discovery will be conversational going forward, essentially, across any category, autos as an example, but grocery, you’re going to keep and you’re going to keep coming back to interfaces that give you high-quality recommendations that are complete, accurate, and have intelligence about around likelihood and suitability. When you don’t get that, you’ll move on to some other experience or service. And so that’s, you know, we’re designing this to to uplevel, you know, those types of platforms and where they’re competing for sort of best-in-class user experience, that’s really what makes it win.

ES: All right, so kind of shifting gears, talk to me about Commerce Go. What is that? Can you walk me through that?

MK: Commerce Go is our new self-service platform that launches at the end of the quarter, so just in a couple of weeks. And what’s different about this is it’s our it’s our first true self-service product for full-funnel cross-channel performance. And really what we’ve done is we’ve made the application like really, really simple to focus on the SMB marketplace. And so it’s self-registration, super simple, and then it’s actually like five clicks to campaign is is sort of the tagline that we help to position it with. A small medium advertiser can be up and running like literally within 15 minutes. Once they get registered, they can pull in creative direct from their domain, we’ll auto-tag or check for their tagging situation, and we can actually do that automatically for them. They put in a budget, a campaign objective, and they can be off and running like that fast. So I think it takes advantage of a couple of different trends in the marketplace. One, the kind of democratization of SMB advertising. It also takes advantage of the sort of compression or convergence between performance and brand. And we think there’s a real need out there for that SMB segment to get access to cross-channel performance, which this enables. So it’s open web, social, and eventually we’ll have other channels included in it as well. It’s been great. We’ve actually been transitioning our small clients to this for the last six months. And what’s new here in a couple of weeks is like a full launch for net new and self-reg. But in the conversions that we’ve been doing the last six months, we’ve had great results and we’ve got 3 or 4,000 clients live on on that platform now, and they get like 20% higher ROAS, well over a third of them are taking advantage of the cross-channel setup to include social, they churn less, spend more, um, they get better performance. It’s a really great product. So we’re excited to get to the the the full net new launch here pretty soon and then continue to build on that with additional channel expansion over the course of the year.

ES: How how do you manage the measurement for that?

MK: So that can be through whatever tag management system they’re using today. So we can detect like Google Analytics tags as an example and it can just integrate like right into that. And so we’ve got sort of an auto-tagging capability and an ability to detect that. So what it does is it takes sort of that technical friction away from a non-technical buyer or whatever extra cycle times that that would typically introduce into coming onto a new platform. So a lot of that’s automated behind the scenes.

ES: And and like how do you see that like so for an SMB, like how do you see that slotting into the kind of existing suite of like automation tools that they already might be using, right? So you think about like kind of Advantage Plus and you think about PMax. Like is this just kind of sits alongside those and covers just the the programmatic inventory? Is that is that kind of how they would approach those?

MK: Well, I think what’s different about this, and yeah, admittedly there are other products in the market, but they’re almost all dedicated to whatever walled garden they came from. And so what’s unique about Go is the cross-channel setup and the ability to hold constant performance across. So we think that for marketers that are coming onto platforms for the first time or some that are looking for maybe a simpler cross-channel setup, we think that Go will be attractive to them. So I think the segment is still expanding and although we’re certainly not first to market with this, it is unique in the cross-channel nature of it.

ES: Right. Can you give me like a picture of the upcoming roadmap? Like what are you working on? I mean this obviously is coming end of quarter, so this sort of the self-serve is on the roadmap, but what else? Like what do you have in store for 2026?

MK: Yeah, sure. There’s a lot. I always break it down into three buckets. Talk about agentic, talk a little bit about performance media, and then retail media. So in agentic, we’ve not talked about this yet, so maybe we come onto it I guess, but would be scaling our OpenAI integration that we announced on Monday. Continue to expand the partner use cases for the Recco service. We are in pretty advanced testing with a few of our retailers on conversational shopping experiences, shopping bots, conversational ads, things like that. And then we have been deprecating user interfaces across our whole product portfolio in favor of MCPE-enabled front ends, whether that’s audience, campaign management, etc. So the MCPE-ification, if you will, of our whole suite is is a big agenda item. And then in performance media, we’re going to be moving up funnel and launching a discovery ad product in the second half of the year, hopefully a beta in the first half, and continuing to look at new supply paths that we need stronger access to like CTV. And then in retail media, it’s really about scaling auction-based display, shoppable video ads. We just launched conquesting and continuing to bring our retailers new monetization paths and and scaling the demand side. So a great new package rolling out for Commerce Max, the demand-side tool for retail media, new insights package, better cross-retailer support, better measurement. So it’s busy. Like there’s a lot going on in the company right now. And yeah, we’re excited about where it’s headed.

ES: I I just do want to get to the OpenAI topic. I mean, I know it’s it’s kind of it’s news, right? So I know maybe there’s not a ton you can share, but like before that, can you touch on just like the MCPE-ification? Like what what motivates that?

MK: It’s to make the company more scalable and and to reduce the reliance on the managed service component. It’s as simple as that, right? I think managed service just doesn’t allow you to scale a platform as as quickly as as you would otherwise. And so we think, you know, automating those things or or in some case just making them easier to access and and and not getting hung up in a lot of front-end UI and workflow. The trend in sort of how all that in like workflow in media planning and activation is towards essentially agent-based or prompt-based instruction and and and interaction. So I would say we’re just in some ways modernizing the front end of all these tools, but what we’ll get is a lower cost to serve and certainly make them more scalable. And and that’s the direction that the market’s moving.

ES: Got it. Yeah, and I think it’d be a great place to end just to talk about your recently announced partnership with OpenAI. I mean, I know it’s it was big news on Monday, we’re recording this on Friday, so a couple days to digest. But like for for people who haven’t heard about it, maybe you could just kind of, you know, provide some some sort of broad strokes outline of of the partnership and what you’ll be doing together.

MK: Yeah, sure. So Monday was great. Really excited to be OpenAI’s first advertising technology partner. And, repeating a few things I guess, but the integration’s live, and what it enables is brands can come through Criteo’s, you know, ad platform and get access to this new contextual ad unit that ChatGPT is opening up for their free and go versions. And the inbound from clients this week has been incredible. Like just level of interest like nothing I’ve ever seen. And I think it stems from people know that this is a powerful surface. You know, our our study show that the traffic that comes from LLM platforms like ChatGPT, like they convert at one and a half times the rate of other referral channels. And so it’s a no-brainer to get into now a scaled access program through a pretty broad-based platform like Criteo and start to test how that impacts, you know, traffic levels. But we’re really excited. We think it really starts to scale this sort of new discovery layer. And we share ChatGPT’s values and principles around like it’s got to be grounded in experiences that are additive, relevant, built on user trust, and I think this first unit that they’ve rolled out is very true to those principles, but there’s like a lot of learning to do and certainly the product is going to scale and evolve pretty rapidly over the course of the year.

ES: Yeah, and I just I just checked and as opposed to pretty much the entire market, Criteo’s up today. Today was a kind of a down day broadly, but you know, it’s it’s obviously been warmly received, the partnership among you know other things.

MK: It has. I mean, yeah, it was a good week in in what is kind of an odd market. But it also speaks to like broader dynamics. I mean, the sort of TAM or the market for discoverability is expanding, and because of how OpenAI is approaching this, you know, companies like Criteo now have access to that TAM that we didn’t have before. So it’s purely incremental to play in the discovery layer in a way that we couldn’t, you know, even two weeks ago. So, you know, if market participants are considering that factor when they evaluate us, then, you know, they’ve got at least part of it right.

ES: Michael, this was fantastic. I really appreciate you taking time on a Friday going almost kind of weekend here, coming up on 4 o’clock. So I appreciate your time and I appreciate you sharing your roadmap with the Mobile Dev Memo community.

MK: Thank you, Eric. We’re big fans of Mobile Dev and the podcast, so really honored to be here with you today. Thanks again.

ES: Yeah, cheers. And appreciate you making this happen. Thanks so much.

MK: Thanks.

Comments: