The Auction Flip: Prebid's Plan To Swap Places With Google

0:08Welcome to The Big Story, a roundtable featuring members of the AdExchanger editorial team. Every week, we bring you an in-depth discussion of key developments in digital marketing and media.

0:23Today's episode is sponsored by Basis, the leading intelligent operating system for autonomous advertising. Its enterprise AI solution transforms campaign briefs into strategies and media plans that integrate directly into omni-channel activation. You can learn more at basis.com.

0:45Fall is conference season. We wrapped Programmatic I own New York two weeks ago. Last week was Advertising Week, and this week was the Pre-Bit Summit. We sent our managing editor, Anthony Vargas, and our senior editor, James Herscher, there to check out this event. And oh, what's their news? Pre-Bit is taking initiative and coming up with a solution that will rework ad auctions in a more equitable way that meets and maybe even goes beyond what Google will be recording to do as the result of its antitrust ruling. I will have Anthony go into the technical details about how the auction setup will change and what that means for all of our listeners whose companies are part of these ad auctions.

1:26Then after the break, we're going to focus on a different area of tech. Our associate editor, Joanna Gerber, went down to Florida for the Treasure AI conference. Treasure AI was a CDP, but don't call them that anymore now. They're going by a term that doesn't yet have a three-letter acronym. Joanna is going to share some of her observations about this conference, including the types of agentic features we're seeing crop up across multiple different types of ad tech and data companies. So stay tuned. I'm Sarah Sluice, our editorial director. And before we start, I want to share what's on my calendar that could be on yours in coming months.

2:03I will be at our Top Women event on November 3rd in New York City in just a few weeks, I'm going to be co-hosting this very glam ad tech event. Yes, those two words can be seen together. So I hope to see many of the top women and their biggest supporters cheering them on at the Chelsea Piers, celebrating all the ambition, the accomplishments of the very talented people in ad tech and media. It's going to be a fun night. And if you have FOMO for missing ProGio New York City, or you simply want more, here's what's up next. We've got CTV World at the end of March. We've got AdExchange or Connect in the middle of May in Las Vegas.

2:38That's going to be a deep dive conference where you learn from your peers, lots of interactive workshops. And I had marketing create a special code just for my listeners, SarahMVL, most valuable listener, that's Sarah with an H, MVL, for a 10 % discount on our early bird rates. So I hope you'll use it because I want to get credit for getting you to sign up for our conference. So with that, Now we're going to get started. There's a lot that went on at the pre-bid conference, but I think the headline that was just our editor's pick is what pre-bid is proposing in terms of how it wants to re-engineer the ad actions, kind of maybe, you know, reorder things, and Google is not going to have quite as much control.

3:25So tell us specifically what that means, Anthony. Yeah, so basically like the elevator pitch on this proposal is to have ad servers bid into pre-bid rather than have pre-bid bid into ad servers, which is how the pre-bid auctions work today. And this is a proposal specifically for pre-bit server, not pre-bit.js. Pre-bit server is kind of like the server-based auction that's more intended for like omni-channel stuff, you know, like CTV, mobile, et cetera. Whereas like pre-bit.js is the version that's used for online web display. And it's like, you know, client browser based. So basically the idea here is that like, you know, from its kind of beginning, pre-bid was originally kind of built to get around the auction constraints imposed by Google Ad Manager.

4:12And the kind of workaround that they did that they came up with was basically to have like pre-bid run like a pre-auction where like all of like the different SSPs would bid on the inventory. And then pre-bid would feed the winning bid from that auction into Google Ad Manager's auction. which is bundled with adx which is where you know the google ad auctions actually take place and that's bundled with uh google's ad server so basically like you know pre-bid would run its auction and then send the winning bid into google's system which you know included its own ad exchange and ad server now by basically like reordering things so that google doesn't get that that uh you know the winning bid sent into its server instead the ad servers have to bid into pre-bid which basically puts Google Ad Manager and therefore AdEx on parity with the rest of the SSPs that are involved in prebids auction, because now they're all just bidding into that prebid server auction.

5:09Okay, so wait, let me just stop to make sure I'm distilling this for everyone. So before prebid auction happens, that result is fed into the GAM auction where the final auction is run. So they're just flipping it. So basically Google has to bid into prebid. So Google becomes prebid. Is that it's it's more like Google is just another SSP in pre-bids auction, just like every other SSP. Just another SSP that was kicking around whether to like have them run in parallel or something. But yeah, like this whole the sequencing matters and actually reminds me of in the during the court during the you know, during like the argument phase.

5:50I think one of the things like that, the DOJ, the big like swing and a misses that they did was they introduced this term, the final auction. logic um this like very important you know all caps three-letter acronym uh to to kind of identify this problem where like google is still in control of the situation pre-bid is just a bid that comes in yeah um but but obviously like nobody you know the google like lawyers took great advantage of the fact that like nobody knew what like oh like talk about you know having andrew casali be like oh you know all about programmatic like what's the fal like final auction logic and he's like what what does that mean another three-letter acronym come on yeah i invented in the courtroom like on the fly like and i'm i'm sitting there like oh man like nobody's gonna get this like yeah so so anthony too i um i think of the headline we kind of distill distill the fact that there's some initiative being taken here that pre-bids proposed auction over overhaul would do it, the court wouldn't separate ad X from GAM.

6:56So what part of this is basically what the court is dictating and what part is kind of the initiative part that there may be taking a step further than might what be strictly dictated by the legal ruling? Yeah. So, I mean, as we all know by now, like, you know, the court did not impose a breakup of Google Ad Manager. There is no GAM breakup. So where this kind of came from, and I was talking to Ollie Burns at Teal and Eugene Dorfman at Post-Industria, who were the co-authors of this proposal. They basically were like, we first put this out there in April earlier this year at Prebids Developer Conference, because like, you know, we were kind of anticipating like, okay, there's going to be like a decision coming down in the Google EdTech antitrust trial.

7:44You know, there's a possibility that GAM will get split up. There's also a possibility that Google will be forced to integrate into pre-bid in some way. So they kind of wanted to lay the groundwork for a way to separate AdX's demand from Google's ad server, give publishers more flexibility about what ad server they want to use because as the DOJ claimed, GAM has like 90 % market saturation for ad servers among publishers. And, you know, it was also like a way to kind of like, you know, make that integration happen because like, you know, they were anticipating like there's probably not going to be like any specific directions for how the court wants Google to integrate with us.

8:29And as it turned out, they were right. And also, you know, there was no court imposed breakup of GAM. So basically they designed this proposal to effectively separate GAM and edX without necessarily having the companies broken up. And it also, you know, there were a couple of provisions included in the remedies that Judge Leonie Brinkima imposed on Google. And Ali Burns kind of like pointed to these three points specifically. It was AdEx demand has to bid into ad auctions independently of Google's ad server. Google has to integrate its ad server with pre-bid independently via an API or some other connection.

9:09And Google has to share bid data, including the winning and losing bids with publishers. So now because like, you know, Google's ad server has to bid into pre-bid just like every other SSP's ad server or other third party ad servers have to, pre-bid gets all of that bid information. They can convey that to publishers. Publishers have that insight into GAM's bidding logic that they wouldn't have gotten otherwise unless they were using GAM as their ad server. and now they have more freedom to plug in their own third-party ad server that can be more customized to their unique needs they don't you know it doesn't have to be the case where 90 of the market is using gam now just to get google's demand seeing seeing the losing bids is important too because um you know you need because like there's this concern that oh google's going to be you know dinging us on like latency that google's going to be like finding these ways like google just prefers its own demand its own systems and now you can see they're like hey like we had better bids coming from somewhere else like why did you go with your own thing that like it kind of makes that very difficult for google to do i also think that anthony's explanation by the way so i was you know in the in this room this javascript working group and i think it is one of these kind of really neat things that is like unique to pre-bid um is like in this room is you know all these people whose like fingerprints are all over this like it's all like javascript developers very hard for people who know each other and there's this crew from gam there's this like delegation from gam who is like it's this big thing it's like gams here they've arrived um and you know and like this group is like arguing about like oh like how is this going to work like what are the remedies and like the gam crew is totally dead silent the whole time like doesn't say that it's all this bickering it's just conversation it's actually like a joke that the guy can't get through his presentation because like the group is just like taking over and talking and the gam crew just like sitting there in like steadfast silence as like the whole group is like wait is that even like what get like who knows what gam will do like we can't like trust them to do that like the court won't impose that and like everyone's like looking at them they're just like not saying a word were they allowed to speak or are they like there's like a silent observer or like google and their legal teams like you can say these two things i mean google's always pretty tight tight-lipped and even like when they're in the room at these like conferences like it's like a joke amongst the attendees are like oh yeah like maybe maybe you'll get like an off-the-cuff like comment from the gam people but Like, you know, you'd have to wait like six months to get an official comment.

11:54I was actually sitting right next to two of the Google guys. It's like in the while there's clearly I've been told, like, don't like you can go there. Like we've said you're going to go there and like observe this. And, you know, because it was like a real conversation. It's like they move. You know, it's not like this is going to parliament or whatever. Like Patrick McCann, who kind of like runs the JavaScript group, is like taking a straw poll in the group. He's like, all right, now we're moving forward with this presentation. Like, oh, and like the other presentation was the president of Prebid's presentation earlier.

12:23It's like, he loses. Sorry. It's like, all right, like Prebid, like they're moving fast. Like, yeah, exactly. So if we were to sum up the feedback to this proposal in a few words, what is the summary? So like when they were kind of discussing the proposal specifically, there were a few just like questions in the room about like, is there a possibility that we could still do a client side you know pre-bid.js auction after the pre-bid server auction that way like you know if it's the case that like a publisher isn't running its partners both in pre-bid server and pre-bid.js if they only have like a couple of ssps that are plugged into pre-bid.js like when do they get like the auction inside you know like that kind of thing and like pre-bid seems very dead set on like the idea of there just being one step that's going to be like you know either it's just this the pre-bid server auction is the final step or like james was alluding to before maybe like you know they run in parallel where like the pre-bid.js auction is happening at the same time as the pre-bid server auction um and yeah they also don't want to be an ad server like they they want like oh like another ad server comes in and like steps in and this is just a module or whatever but they you know in some ways i think it'd be easier if there was like the pre-bid ad server but they don't they really don't want to be like an ad server yeah yeah that's definitely true they want they want this to be like publishers get to choose their ad servers and like they get much more flexibility and they don't have to use so it's pre-bit an ad server then if they publishers choose their ad servers plural uh that's an interesting i don't really know how to answer that question it's still like what they were emphasizing to me was publishers get to choose whatever ad server they want i think that means that pre-bid is not an ad server but i could but what other ad server is there i mean it would be like you know like index exchange like could have its own ad server and it's like well you can use index exchange as that server there's also like um stuff like uh you know the trusted server um iab tech lab thing like that would function as an ad server in this case like hillary slattery was in the room we were talking about she was joking like well we of course we would prefer everybody use trusted server you know like that kind of thing um and then like um you know there's also like this was something that eugene uh from post-industria brought up to me when we were talking about this in advance of the summit he was saying that like you know this kind of lays the like paves the way for um agentic first ad servers to come into the marketplace because like you know like companies are building those but if gam has 90 ad server saturation once publishers it's like okay what are you going after that 10 piece of the pie with these like agentic servers he's like now he's saying it's like oh yeah this actually opens up the door for publishers to be like experimenting with this stuff and like more proactively choosing these emerging solutions okay well i'll be curious to run that one by a publisher because i feel like ripping out the ad server i think is the thing that most publishers are too afraid to do and are maybe too smart to do it would definitely be a big change and like i think like you know the idea of people kind of weaning off of google's tech is scary for a lot of publishers because like it's what they've been used to for a long time but i mean you know there were like plenty of publisher representation in the pre-bid community and you know this is this proposal's in the common uh stage at this point so i'm sure they're getting all the feedback from publisher members and all that good stuff i think the question would be if you're if you're having adx bid into pre-bid server then what is the role for gam is it just placing the ad and marking that it actually ran and it doesn't have any decision making capacity or is it used for the agentic decision making like what's the yeah it's like the way this was kind of framed by pre-bid was that like pre-bid server would become the decisioning layer for the auction so i think gam still does like you know again includes adx so like adx is still running like the google auction the p the the bidders that are bidding through google's auction will still be running through adx and then like gam will still be like grabbing it's like analytics on like bid behavior and like you know like i think you would also have still have like the controls that publishers can put into place through gam reflected in the adx auction portion of things just like the controls they would have put in place for any other SSP would also be reflected in the SSP how the SSP is bidding into the pre-bid auction.

17:01But yeah, there's a lot of like these kind of like, you know, unanswered questions and you know, the tricky kind of like, like figuring things out and like how things are going to look post like this proposal that's for one thing, like, for example, this is the sort of technicalities are beyond me, but there was a concern about how this is sort of set up and companies being able to figure out like the logic of another SSP or like another publisher or something like that. That's a major concern because that was like another thing where it's like, why don't we just do this through prebid.js? And then it's like, well, then, you know, all the publishers business logic would be contained within the browser on the client side where, you know, it would slow down the page and where browser companies would potentially get like a look at it that maybe the publisher doesn't want.

17:54And if you do it server side, this was another point that Hillary Slattery had brought up. Like, you know, you're effectively like limiting the transfer of your data and your decisioning on like, you know, auction pacing and like all these different, you know, terms of art that go into things. You're keeping that limited to just like your trusted partners and not someone kind of someone joke like have we have we considered closed sourcing and there was a bit of like a like a no i mean obviously pre-bit is all about open source software yeah i'm gonna make a prediction here that i do think that pre-bid server might finally gain a lot of momentum and adoption perhaps as part of this change last week last week i was also at a conference over in seattle the ib ib privacy compliance salon and uh steve sullivan home walmart was there talking about how they only do server server side things not the javascript on the page and just all all the lawyers in the room were like very anti letting anyone put pixels on pages like running javascript all of that just gives them hives and they also feel like that's how you lose data as a publisher, it can cause all kinds of privacy issues.

19:12And it definitely felt like the direction that many companies would go because it does protect their own data and also protect them from legal issues. And I think with server, there's been a lot of issues, mainly because you are flying a little bit more blind if you're at a server side integration. But I'm going to predict that the balance of the scales is going to tip more in the server direction. And also it seems like the technology is more mature, flexible. I think that the list of pros and cons is going to be weighing more towards server-side integration. Apple is forcing the web there anyways.

19:51Like the general response to the iOS stuff is like, we got to get on server. That's an intriguing wrinkle as well, that it's more hidden, right? Because that was part of the debate with the privacy thing where people are like, oh, this isn't as transparent. but it's also more protected. I think I remember people, you know, a decade ago, Amazon, everything was server-side and people were like, almost like, oh, they're doing it for sneaky reasons. But now I'm like, well, actually, maybe that was the privacy safe way all along to have it be encrypted via - Yeah, I can say a lot of like sources I talk to sometimes about like, you know, just like privacy related things, people who like, you know, participate in W3C working groups and stuff.

20:32They seem very bullish about server side integrations, IAB Tech Lab, trusted server, things like that. And exactly to James's point, like, you know, like that was another thing that was the iOS 27 thing was something that was coming up in conversations at the pre-bid summit. And where it was like, actually, Justin Wall at Attitude, he was just like, yeah, this is just like another reason why we just need to move everything server side. He was just like, yeah, like Apple's just given us all a big flashing warning sign, like get off the client side. So you can both protect your own data and perhaps protect what data you're collecting from others.

21:09It's a little bit of a paradox there. So thank you, Anthony and James, for being on the ground there. We're going to take a break and talk about what we learn about the land of CDPs and what treasure lies beneath.

21:30I'm Sarah Sluice, editorial director of AdExchanger. And with me today is Michael Olson, EVP of client development at Basis, which is supporting our podcast this month. We're going to talk about something that doesn't get enough attention, paying invoices. What is the cost of slow financial reconciliation and who is the most impacted? Hi, Sarah. Thanks for having me. The cost is time, cash flow, and trust. When financial reconciliation takes too long, payments get held up and finance teams, whether they're in agencies or ad brands, are just chasing down numbers. People are spending a ridiculous amount of time figuring out why two systems do not match.

22:16Agencies probably feel it the most because they're stuck in the middle carrying the float. They have the client asking questions on one side and media partners on the other. The longer it takes, the more time everyone spends fixing the past instead of focusing on performance and health and growth of the business. So you outlined a very frustrating problem. I would not want to be that person chasing down invoices. Is this primarily a technology problem, a workflow problem, or a people problem? It's all of the above, but I'd lean towards mostly a workflow problem. The industry definitely doesn't have a shortage of technology or data.

22:57We have a shortage of systems that actually work together. And when they don't, people end up filling in all the gaps manually. They end up downloading spreadsheets, comparing reports, sending emails, chasing down discrepancies. It's a mess. Technology is obviously part of that answer, but another piece of software by itself just simply doesn't fix it. It's imperative for technology to actualize omni-channel data monthly and push to reconciliation systems. As an industry, we have to connect the tech stacks interoperably, end-to-end, and then allow the technology to do the heavy lifting to align the data between the contracts, media plans, media transactions, and the payments.

23:43Okay, so it sounds like technology and workflow can make this just a bit easier. What does automated reconciliation look like? A lot of automation talk right now. And if I'm thinking about the future agentic state, ideally, reconciliation just happens. AI is constantly looking at delivery, spend, invoices, payments, and then flagging issues as they happen. Instead of somebody finding them three weeks later, and in a lot of cases, the AI should be able to fix those issues without a person ever touching them with the correct furnaces in place. People should only have to get involved when something unusual happens, which is the human oversight.

24:26That gets away from this huge month-end exercise that costs a lot of time and resources and much closer to knowing where you stand financially at any given time. This is the ideal state basis is going towards for our connection with Prisma by MediaOcean, a platform used by numerous agencies for financial reconciliation. Okay, so it sounds like by connecting those two different platforms, we can and adding agentic AI, we can make it a lot easier and people won't have to chase down invoices all the time like they used to. So thank you so much for sharing with us what that would look like. And thanks to Basis for supporting our podcast.

25:04Thanks for having me, Sarah.

25:12And we are back. So, Joanna, we sent you down to Florida last week. You escaped advertising week. And you were at the conference of a company called Treasure AI, a company that I know as a CDP or a customer data platform. But, you know, whenever there's a category created, like, after a very short period of time, what happens next is everyone wants to flee the category to the point where, like, we'll write a story about CDPs and we'll get emails after. and people will be like, we're not a CDP anymore. Like, we're this. I started this job so recently, and you guys were like, here are the SSPs you need to know.

25:52And within three months, every single one of them I spoke to was like, well, actually, we're not an SSP, mind you. Yeah, there'll be, I can't say how many correction emails I've gotten. Just call them on ad network. Not this. Don't call us this. So they're calling themselves an AEP or an agentic experience platform. So they're just coining another three-letter acronym. So I'm going to give them a little shade for that. even though it's something all the companies do. So, but tell me, what does a CDP turned AEP do for marketers? Okay, first I will say in their defense, they have not called it an AEP.

26:24They have not tried to acronymize it. That was all you, Sarah. On their website, it does, Joanna. Oh, shoot, I should check my sources. Although they do still call themselves a CDP on their website. So I'll give them credit for not fully abandoning one. They're just trying to be both. Well, regardless of what they call themselves. And everything eventually becomes an acronym in this industry anyway. If it's like a three-word phrase, even if you don't acronymize it, it will be acronym. We'll add a third thing to it to make it a three-letter acronym. Okay, enough, Joanna. Sorry. Yeah, enough. That was all on me.

26:56So basically an agencing experience platform, also known as a, I don't know, CDP on steroids. Basically, it just means that throughout every single step of the customer journey, there are agents involved. They can be queried in natural language. and all of a customer's data from a ton of different sources is in one single place, which we can get into this a little bit more later. But I think that is something that is becoming less of an anomaly and a selling point and more of an expectation among customers. I think for a while, people weren't really using centralized data as a selling point and being like, oh, are you tired of bouncing between 16 dashboards?

27:31We have good news for you. And now, you know, we have MCPs. We can unify data really easily. We can have different platforms talk to each other using AI. And as a result, I have a feeling that customer expectations are evolving very rapidly and they're going to start expecting that at face value. So maybe every platform should have an agenda experience. Oh, I might have my CDP, my CRM, maybe a DMP, you know, rattling around somewhere. So like what is the CDP more for that paid plus owned data? Is that kind of the shorthand? Yeah, it seems to be primarily their own data that they've accrued over time.

28:11Much of it coming from, yeah, their own like CRMs and things as well. When I talked to Nestle, they said most of the data came from a CRM. So, okay, great. So you talked to Nestle Mexico, which apparently just owns every single product in a grocery store in Mexico when they're like leading up to the list. It doesn't appear that way. Oh, a lot of products in your grocery store. So is there kind of a good shorthand or way of thinking? like what this type of tech can do for a company like that, a CD or a CPG with a CDP. Yeah. So for Nestle Mexico, something I thought was really interesting, which you might have noticed if you read the story, because I literally made it the headline, is that normally brands are like, oh, we want as much data as we can possibly have on our customers.

28:59But Nestle was like, you know, for us, that was an asset, but it was also a massive challenge because we had so much purchase data and so much consumer data, but we didn't have a way of tying it together between different brands within the overall Nestle umbrella. So, you know, somebody might be buying sweet and condensed milk and that same person might be buying dog food and they had no way of necessarily bringing that data about that person into one central place. They only had dog food brand data and condensed milk data. So I think something like Treasure AI was a huge asset for them because it helped them understand the customers more specifically and understand ties they weren't seeing before between brands where you know they were like oh a lot of people who buy chocolate are also buying condensed milk because they like sweets but a lot of those people are specifically looking for a quick sweet treat so now we can target that type of audience with the same types of ads not for the specific product but for the specific moment they're looking to shop in and I thought that was pretty interesting okay it kind of feels to me also like a tale as old as time because it brings me back to the days of like i don't know like seeing a recipe for let's assume cool whip and graham crackers and like jello are owned by the same company and then they're like here's like the cool jello graham cracker pie that you could make and they like have them like but but they're gonna call it a salad not a pie that's the old school way of doing it but there's definitely product overlap you know in ways that you know recipes that use them all hopefully not a recipe that combines dog food and sweet and condensed milk.

30:29But maybe you want your... I mean, I don't know. I feel like my parents' dog would really like that. And, you know, in terms of the thesis of the conference, kind of the talking points, the directions brands are going in terms of how they want to use their data and how they're actually also using them, what's the pulse check there? I think something that is being talked about more loudly than it was for the past couple of years is human accountability for AI. That is kind of Treasure AI's whole thing. Rafa Flores, who's their chief product officer, talked a lot about accountability and how it falls on the people who are creating the tech and people who are training the tech, not just the tech itself.

31:13We've had this conversation before. We've seen it in the news where an AI agent goes rogue and everyone says, okay, who's to blame for this? And I saw a stat a few months ago that I thought was very interesting that said that a, you know, I should have pulled up the stat. I don't remember exactly what number, but it was a significant portion of consumers will blame, maybe not blame, but they will lose respect or lose trust in a brand if they get inaccurate information about a brand, even if they get that inaccurate information from a third-party agent like Google's AI overviews or from, you know, ChatGPT.

31:45So it's more important than ever to make sure you're training AI accurately because it's going to have a negative impact on the brand, even if they aren't legally accountable for it. and you know on the accountability point like something we're just hearing a lot more at this was like mentioned at our programmatic io event last week uh we also recently uh published a column on this by uh evgeny popov which if you haven't read it head on over to ad exchanger.com and check that out it's basically about you know the idea that insurance underwriters are becoming more involved in ensuring these ai solutions and so it's like you know when you were to ask like Like, who's responsible for when things go wrong?

32:24And it's like, there needs to be like an escrow account somewhere to like help pay everybody off so that, you know, if something goes wrong or if there's make goods or clawbacks or anything related to like agentic AI, you know, we have something in place. So that's becoming like a bigger industry concern right now. For sure. And like, you know, this isn't our industry, but we've obviously seen cases in the news about lawyers using AI and then citing cases that don't exist. And obviously no one goes, oh, we'll have to hold open AI accountable for that. Like the lawyer is also going to get in trouble personally.

32:56So human accountability is massively important because beyond ethical reasons, it's also going to come back and bite you in the butt. Yeah, that and we're already hearing it. What does that have to do with like the sweetened condensed milk and the dog food? Like, is that just like about hallucinating connections or something that might go back to consumers? like what do you think is the through point for a brand example Joanna or is it sure well I think that was a little bit more kind of standard AI in a way that people are pretty comfortable with but I think this becomes more important when we talk about generative AI and you know new copy being created new creative being developed and among a lot of tech vendors I've spoken to they tend to give brands an option of how much oversight they want so you know some brands are comfortable saying we trust this company you make a new asset and we will automatically send it out but a lot of those brands are choosing to still have you know human eyes on the new copy or the new creative you know whatever new ad version it is before they send it out so that they can flag errors and i talked to a tech vendor recently um where they said that recently one of their clients was i believe a skincare brand and the ai accidentally started writing this ad for hair care and they went hey that's that's actually not what we sell you need to retrain your ai on our product so that this doesn't happen again and obviously if they hadn't had oversight there that would have ended very poorly yeah okay so i think that yeah definitely when you get to product descriptions of brand guidelines a big big issue so i want to i want to kind of close with one other trend that um that you observed during the conference in terms of it seems like instead of having an agent people kind of like to have a whole a whole team of agents different different names different purposes you got like a whole a whole small workforce of people with different roles so like where else have you seen this trend and like what why is this is this like a human thing or does like the ai actually work better when you have like 10 people instead of trying to have one or sorry 10 people 10 agents versus one agent trying to do 10 things so yeah um at the conference treasurer ai unveiled something that they're now calling digital workers which are exactly what they sound like they're little ai employees and they each have a different job title and role and last month or it's actually the very very end of august optimizely optimizely launched almost the exact same thing i actually covered it once again at exchanger.com take a look and the interesting thing there and the way that optimizely framed it being kind of unique and different is that it's less about having multiple agents that each do their own task so obviously that's important but it's also that for example sarah you might have a claude instance and allison might have a claude instance and both of those might be trained on your information and your data and have access to your email and your calendar but as of right now there is no agent that has access to your and Allison's data at the same time.

36:07So when the agents can communicate directly to each other, they have a much wider range of access without obviously infringing on any individual's privacy. And that's one thing that these companies have been touting. I think it's kind of weird because the agents also talk to each other. And Treasury AI did a real-time demo on stage that involved the agents just making small talk with each other. like it didn't further the plot at all and one of them said something crazy or he's like oh i keep the conversation going and the coffee interested or something like that so ai is still not good at small talk but i hot take don't think it needs to be yeah i i also will give credit to a company doing an imperfect live demo versus like a a perfect one that's so rehearsed that like they're not even trying anything a little absolutely i like to i like to be on the edge of my seat so give me something i mean out of all the things to go in perfectly ai's messing up some small talk is is not exactly the tech blip that it could have been yeah exactly i want an authentic agent here um that makes some some mistakes and is is not perfect at small talk okay um so what what what What do we think we're going to see next out of either the AI infused product trends or just maybe in the CDP land or the AEP land?

37:33Joanna, what did you leave thinking we're going to see more of? I said this in my predictions video before ProGio last month, and I really stand by this, which is that I think we're over the period of pure AI hype and we're leaning into where humans need to have hands on the keyboards and be paying attention. because the kind of thesis that Treasure AI was really leaning into the whole conference was that AI needs human supervision to the point where like they handed out tote bags that said that, which I might not be carrying with me on the subway, but I do appreciate the sentiment. And I think that, you know, now that AI is capable of so much, at the same time, it's a lot easier to see what it's incapable of and maybe what it's quote unquote technically capable of, but not doing as good of a job as a human can.

38:18And I'm thinking about, you know, visual creative and AI-generated food art. And we've written about that in the newsletter. So I think the next step of AI is going to be kind of, you know, the honeymoon phase is over and now we're starting to see the flaws. And the question is, will the relationship last? That is the perfect note to end on, Joanna. I'm going to carry that through the rest of my week with my AI interactions. And we'll see you all next week.

38:50This episode is sponsored by Basis, the leading intelligent operating system for autonomous advertising. Its enterprise AI solution transforms campaign briefs into strategies and media plans that integrate directly into omni-channel activation. Learn more at basis.com.