Loading summary
Transcript40 lines
- [00:09]
A
Welcome to the Big Story, a roundtable featuring members of the Ad Exchanger editorial team. Every week we bring you an in depth discussion of key developments in digital marketing and media. This 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 omnichannel activation. AI is causing massive change in our world and in our neck of the woods digital advertising. If you think several steps ahead from where we are in agentic AI, there is a great possibility for things to go right or terribly wrong. This week we have a guest who is looking at the chessboard several moves ahead and has a nuanced take of what AI may do to our industry. James Deaker, who recently spoke at our programmatic AI conference and post the yield doctor videos on YouTube, wrote a column for us about risk. What happens when autonomous systems begin making commercial decisions without clear ownership, governance or accountability? We are going to dive into his POV as the founder and head of Corkea Media. I think I got that right. Corkea Consulting and with the goal of making all of you more informed and with a more clearly sketched out picture of digital advertising's future. Now that AI has entered the chat, I'm Sarah Sluice, editorial director of AdExchanger, and with me today I also have Joanna Gerber, our AI reporter who's going to add her perspective to this discussion.
- [01:56]
B
Reporting for duty. No pun intended.
- [01:59]
A
And before we get started, we got one bird, Korkia. We've got an early bird deadline to sign up for Programmatic IO New York City that's ending this week, so you can sign up and stack our Pod 10 discount. On top of that, the conference will be September 28th and 29th at the Marriott Marquis and with our sharpest minds as speakers, I think James will kind of give a good, good preview of the caliber of speakers we bring in. So now let's check in. James, really thoughtful, nuanced take. I will just say like we get so many column submissions about AI and a lot of them are like so warmed over, like kind of seeing the same surface level things over and over again. Yours was not that. So paying you a compliment by way of an insult other people. But anyway, what inspired you to write this column?
- [02:52]
C
First of all, Sarah and Joanna, thank you for having me. What inspired me to write? I actually, I don't. I think you remember I spoke at one of your conferences at Admonsters six months ago and at that time spoke about how I saw AI evolving and bucketed things into three different groups. Where I saw an evolution, where I saw true revolution, and in areas where it was too early to tell. And one of the themes running through that was using like self driving cars as an analogy. And the picture I had for too early to tell was potholes in the road, the things that we just didn't know that the cars would be tripping over. And so that had been on my mind a lot. And so as I looked at some of the challenges of the things that we don't know over the last six months, I see that we've been significantly underestimating what I'd call the long tail risk. So one of the themes running through my talk was this idea that a lot of things will get easier with AI. Hopefully if we get it all right, there should be a lot less fat finger mistakes that a lot of sort of the problems that, you know, errors that people make should go away. And so in the day to day we should see less risk. But I don't think we've been looking effectively at the worst case scenarios and even the things that could take a whole company down and take them down really quickly. And so I actually started the talk with an analogy from a guy called Ron Howard, not the actor, the Stanford professor who formed decision analysis. And the lesson was one of those formative things in my life. I was a PhD student in his class and you had to do a multi choice test, and everybody knows how to do a multi choice test. So we were all acting like we knew how to do it. But he had a twist. Instead of just answering the questions and putting a mark by A, B, C, D, which of you thought was the right answer, you had to assign a probability to whether you were right or whether you were wrong. And so if you were 100% sure that the answer was A, and it was A, you put a hundred percent, you put zero by all the others and you'd get one point for that. But he had this asymmetric component to teach us a lesson. So if you put a 0% and that etchy ended up being the answer, you would get negative infinity on that question and therefore you would get negative infinity on that test, and therefore you would get negative infinity on your whole course grade. So this brought home to me the idea that if you didn't understand the new framework that you were operating in, and there were many people in that class who hadn't internalized this and put down 0% on things right, and got a negative infinity, they hadn't learned that they were operating in a different environment and they hadn't learned the lessons from long tail risk that there was potentially significant consequence. And so I did the analogy for right now we are all continuing within digital media and digital advertising to act as if the world has not changed. Like we all know how to do those multichoice questions. And so we're all ticking the boxes and putting a hundred percent and nobody's thinking about what does that negative infinity look like? And negative affinity could be the company you work for going out of business. It could be bringing down others in the industry. So that's a long winded answer to your very short question.
- [06:37]
A
Well, and James, you have a background at a lot of different publishers and platforms kind of helping people market, sell, inventory, yield. So from that publisher point of view, I'm wondering like what could go wrong? Like what would be the worst case for a publisher who's trying to sell their inventory?
- [06:58]
C
So I think in the talk I laid out about five different worst cases. I'll start with the one that I know very well, which I think is overlooked right now, which is the pricing collusion example. And the reasons overlooked is not a lot of people and a lot of publishers spend a lot of time thinking about pricing and pricing law. But, but at least in the US a lot of the law has been in place for a really long time. The Robinson Patman act goes back to 1936 and it's usually put in place to protect against companies monopolizing or acting together to become effectively a monopoly and move prices in a way that hurts consumers. And so what you don't want to do is have publishers acting together, making rules up behind closed doors and saying, you know, we all think that a 300 by 250. We know the market price is $5, but we're all going to act together to sell it at $10. And then advertisers would be the ones who would be effectively, you know, screwed by that. And right now, in a pre AI world, the way that the Justice Department or any other organization would investigate something like that is they'd look for paper trail. Did companies have meetings? Was there a clear action where these publishers got together and said we're going to set prices in a way that's potentially going to hu advertisers or consumers and there's a paper trail. In an AI world, there are a number of different scenarios where you could have exactly the same effect that would be driven by agents without a person making an explicit decision. And so there's what I called implicit and explicit decisioning there. The implicit one could be the AI agents don't necessarily talk to each other to collude, but they somehow realize that it's in their interest to start pushing prices in a certain direction. And because they're all based off the same underlying code and logic, we start to see prices for the whole industry drift in a way that starts to hurt advertisers, agencies, consumers. And that would be essentially implicit price discrimination. We don't have clear law around how that would get handled now, but if it was handled as it would, there would be a clear violation. The more explicit case is as we start to put protocols in place, whether it's the MCP or whether it's agent to agent, you're going to have more direct communication between agents that are both buying and selling, potentially selling agents communicating with each other. And you can see a situation where they may explicitly realize that it's in their interest to start moving price in one direction or another. And again, anybody looking at the effects of that, it would be similar to if there was a pricing discrimination occurring. But in these particular cases, the leaders of the companies, the executives in charge, may not have even realized that they were violating the law and why that's important. And I talked about that a little bit when I spoke at the conference is there is now a growing evidence of law that companies are still going to be responsible for their AI agents. There was an example from Air Canada where their chat bot on their website was giving a policy to a consumer that was actually at odds with what was represented on the rest of the website. The judge in that case in Canada said a reasonable consumer is not expected to go at the look at the rest of the terms and conditions on the website if the chatbot is saying here is the rules. And so in that case, the, the company was on the hook for the actions of its AI agent. We haven't got our heads around that yet. And so part of what I was talking about was how do we start to put governance in place to, to handle those kinds of constraints.
- [10:59]
A
Yeah, I, I like the point about price inclusion. You bring up the Robinson Patent Patman act, which we are seeing come up a lot more. I think there's a lot of anticipation it's going to be used for algorithmic pricing and dynamic pricing type litigation because we're seeing kind of, you know, personalized pricing that is not great when you're the one that's being charged the higher price. I want to bring up another old example. It's kind of funny to look back and realize this example is like more than 10 years old, but there was this like $23 million textbook for sale on Amazon. And basically they were just had this algorithm that was looking at a competitor's price and then raising it by a penny. I think every time that their price was raised and before you knew it, there was a $23 million textbook. You know, that would be great for the seller, you know, if someone accidentally bought that $23 million textbook or whatever. Right. Not so great for the buyer. So, you know, but maybe I could see the opposite being an issue of somehow you're selling a car for, you know, 10 cents instead of, you know, 50,000 do thousand dollars or whatever. So. So maybe tell me a little bit about that. I mean, in some ways it's kind of old school AI but could it get weirder? As we move into the world of agentic.
- [12:19]
C
But could it get weirder?
- [12:21]
B
I mean, yes, always get weirder.
- [12:23]
C
You know, the answer to that has to be yes, it could get weirder. I, I think we're going to see a lot of untoward consequences. Some of them will be by design in terms of bad actors starting to do things against companies, whether they're creating artificial traffic. And some of them are just going to be use cases that we hadn't foreseen years. And I'll draw another analogy from another industry. There was examples in San Francisco when a group of people all chose to call Waymos to the same cul de sac, knowing that that would create a traffic jam, which it did. And Waymo had to put rules and procedures in place within their own algorithms to protect against that. And so we're going to have, we're going to see a lot of situations that, where companies are going to have to change the way they operate or the rules for the, for the AI. I think one of the things that is going to be a lot different for many of us who are used to dealing with enterprise software is that it's not clear what a rollback means. And so one of the things I talked about is the need for an audit trail and the ability to go back to the state you're in. If we were in enterprise software and we received a new version of that software and there were significant problems with it, you would go to your IT department and say we need to roll back to the previous version. In an AI first world, that is not going to be possible in the same way because of the way these systems are designed. And that in itself is going to mean that we have to work out how to create stable states along the way that we can get back to. Since we probably won't be able to, quote, roll back the AI, we may be able to change the guardrails, we may be able to do a number of things. And as we go through this, we're going to have to work out what it means to be creating an audit log. Maybe it's a back end series of data streams, but I got a question from the audience who was talking about when agencies bring campaigns live, they often take screenshots. Everything along the way. I think there's going to be the equivalent of virtual screenshots of a lot of different stuff so that we all have insight into what's happening on both sides, because there are going to be a lot of your weird stuff that's going to happen, and we're going to have to have an understanding of why, how, and exactly what was going on.
- [15:02]
A
Okay, so more documentation is one possible solution for this. We have so much more to talk about. We're going to take a quick break and then return. I'm Sarah Sluice, editorial director at Ad Exchanger, and I'm with Katie McAdams, the chief marketing and Commercial Officer at Basis. Welcome, Katie.
- [15:26]
D
Thanks for having me.
- [15:27]
A
So Basis has found that media teams today are juggling an average of nine different platforms to run a standard digital campaign. Which makes my head spin because I know when I like switch browser tabs or switch products, I'm like, wait, what was I here for again? So how does this impact their ability? Ability to be successful as an advertiser? Sure.
- [15:50]
D
So it's a great question. What we find in our research is that our industry is losing anywhere from 80 to $100 billion annually in value leakage from errors, inefficiency and siloed campaigns sitting in all of those different platforms that you're talking about. And advertising is just becoming more and more fragmented, whether it's across teams, channels, tools, finance systems, and now different AI solutions. So that's a lot of context switching for one team in one.
- [16:18]
A
Wow. So what would a connected advertising system look like as an alternative to those nine platforms?
- [16:25]
D
What we find in talking to agencies and brands is that that journey really needs to start with consolidation. And by consolidation, I mean getting all of your media contracts, your campaign plans, your invoices, and your client communications into one place so that you have a single source of truth. And once you have that foundation in place, then something important really starts to happen. You actually have data that is clean and reliable so that your AI can function with it more meaningfully and more predictably. So the brands and agencies that get to this state fastest are not going to be the ones who are bolting on the most AI tools. They're going to be the ones who are able to build that operational foundation first.
- [17:03]
A
I like this point that centralization isn't just about me as the media planner, but also about having more unified data that will then help me with AI, which I'm glad you brought up AI. So tell us a little bit more about how AI is being added on to this connected advertising system.
- [17:22]
D
Sure. So this is where having that solid foundation in place is going to actually help AI become more of a multiplier for your organization and your teams. As an example, Basis has Compass, which is our agentic AI planning tool. It lives right inside our platform and it solves the problem of media teams spending hours and even sometimes days synthesizing media briefs, building frameworks, building media plans, and then creating client ready presentations before a campaign even launches. So Compass actually takes that brief and generates a complete omnichannel strategy across programmatic, search, social and Direct in minutes instead of weeks. That strategy then becomes connected and pushed into their media plan, which can then be activated on through the BASIS platform, across programmatic, search, social and direct media buys. And what we find is that agencies and brands who are using Basis overall are seeing 30 to 40% operational efficiency gains when they operationalize all of those workflows into one place. And that really creates an expansion of capabilities with teams being able to gain back time to focus on strategy, creativity and growth.
- [18:29]
A
So we have more efficiency through centralization, which then enables more use of AI, which is even more efficient. So really interesting to talk to you Katie and thank you to Basis for supporting our podcast.
- [18:42]
D
Thank you.
- [18:50]
A
We are back and we're going to talk even more with James about what can go wrong or right with AI. Do you want to. Want to kick it off with a question?
- [18:59]
B
Yeah, absolutely. So James, I'm glad you're talking about kind of the importance of AI governance because I think you're right. That's absolutely something that's overlooked, especially with all the excitement around automation and the idea that humans can just be fully hands off when spoiler alert. They can't. But famously, along with AI, humans can also be flawed. So I was curious, kind of how you think that we as an industry should decide who is in charge of this ownership and governance. Should this just be, you know, every company owns their own individual products or should there be kind of maybe a larger body overseeing these tools on a bigger level.
- [19:36]
C
Oh, wow. There's a. There's a lot in that question. So, yes, humans are flawed, but I think one thing is that's not a reason to let AI just do its own thing. If nothing else, it's clear that we approach problems right now in very different ways than AI. And so that's a check. Whether it's a good human chick or a bad chicken, there is a check in doing that. I think one of the things that's important to think about these interactions is the role of different groups. And by that what I mean is there's going to be expectations of the person who is working directly with the AI, like asking the questions, running the campaigns as an individual. There are going to be a different series of expectations for groups within the company, whether it's the equivalent of what would now be IT governance or your legal oversight, possibly groups that don't exist, but there is going to be a group who's going to have a different series of expectations. And whether you think of that group as being a committee or a function that doesn't exist, clearly there was going to be a different group within that company. To your point about the industry's role, I think that is a conversation I think we need to get going much quicker as an industry organization in that there is clearly a need for more industry leadership. Of course every company is going to have a slightly different take as to what is important, but it's clear that we as an industry have benefited from the fact that most of our technology is homegrown. And so we have set up our own tech stacks in a way that work for us, although in some cases, obviously there's a lot of pain along the way. But why that's important is things like privacy and how you support GDPR and California ctpa. Those aren't well understood within the AI frameworks right now. And so there are some very smart people in our industry. If you talk to the people at the IAB tech lab, they're throwing around scenarios like we may need to create our own version of a agent or an infrastructure just for the ad tech and the digital media industry because we can't rely on the general models. And so that question of how these groups are all going to fit together is, I think, still open and evolving. The fact that we have different protocols that are being pushed by different groups is understandable at this stage in the evolution. I think it would probably be negative if there was only one, but if you look at the fact that the underlying MCP is industry wide. There's now adcp, the IAB have their own versions. There's a number of different protocols that are being promoted. That in itself isn't necessarily unhealthy because I think you need some innovation there. Clearly we do need protocols and we probably need more about the governance, because a protocol is not governance. Protocol is just a language for how different agents can talk to each other. The rules around what these agents can and should not be able to do. When I raised the pricing example that I just shared with you all, there were a lot of people who hadn't thought about that use case and actually were designing ways for the agents to act more independently on price, who also hadn't thought about the fact that they may even be in violation of pricing law by doing that. And at the moment, nobody's throwing the flag on this. And I will tell you, first of all, thank you again for having me on, but what I'm talking about is not sexy. This is not what people want to talk about at conferences. And I was essentially the only person saying governance controls here are worst case scenarios, because most companies want to be talking about all the great things they're doing to make the industry better and safer and move faster and better for consumers. As soon as you say things like governance and control, half the audience sort of turns off and goes to sleep and the other put it in a bucket of that's important, but sounds difficult. I'm not going to deal with it today. And so the question how we get more urgency around this as an industry is very much top of mind for me.
- [24:26]
B
Yeah, well, I think that also plays into the news that came out on June 2, which is that the Trump administration just had Trump sign an executive order. Well, or maybe he autonomously decided to sign that executive order, basically requesting that companies that are generating new AI tools allow the US Government to look over them and kind of give some oversight before those tools become available to the public. And in the past, Trump had actually refused to sign a very similar executive order. And now again, I think it is at the discretion of these companies. But I don't know. Do you have any thoughts on that and maybe what the role of the government should be in that and what sort of details they should be looking at before these tools are released to the public? Obviously that's beyond just ad tech and that's about AI on a larger scale.
- [25:14]
C
I think one of the great things about America compared to other business economies is the fact that it's generally pro innovation. If you look at some other parts of the world, there is a lot of regulation that mean that innovation has slowed down. And so generally I'm probably leading businesses run fast. That said, we are dealing with technology that in many cases the companies who are building these don't adequately understand themselves and don't understand the risks involved. And to be blunt, I feel that companies need to be in position where they bear the consequences for their misaction. And we may be in an environment right now where people are building so fast, with the prospect of making money so fast that they're not worried about problems that can exist two, three, four years down the road. And anytime you're in that kind of environment, you need some external force, whether it's the government, whether it's industry bodies, whether it's legislative processes to say, hold on, some of the things that you're doing can have untoward consequences that you're not taking account of right now. I'll leave it at that because I'm not going to get into the politics of who should be doing what. But clearly there do need to be controls. And in some cases companies don't have incentives to build those controls for themselves if they think that they can cash out before the problems come back to bite them.
- [26:54]
A
I have, and I have so many thoughts here to add. I'll maybe just add a few of them. I think, I think what comes to mind for me is the open RTB spec, which I think is an industry standard. So I think one to speak to the government point. Does the government know like how to make the open RTV spec fraud proof? No, that's the kind of thing that falls on the industry. And there's been so much fraud over the years related to the fact that you can put whatever you want in this spec. And that's what led to the creation of ads txt, all stuff I covered a lot many years ago. And talking to some of the people who built these specs, they just had no idea how it would be misused. They weren't really thinking about how it could be adopted and adapted for the industry. And I think in some ways, you know, they were building for one thing and ended up being used for another. But I feel like we're at a similar point here with AI where it really will benefit the industry overall if it can think five or six steps ahead the best it can, instead of only thinking one step ahead and then having to deal with like years of cleanup and fixing and it's just, just it's just so much harder. So, so I hope James, that you inspire some people to think about some, some different ways that this could be misused. Because I, if I could bet on one thing, it's that 1 million percent there are going to be bad actors testing every single, you know, possible line of defense. They're going to be ordering the waymos to the cul de sac. Like that's just way the way this industry works. And so ordering the weight loss to
- [28:47]
B
the cul de sac is such like funny
- [28:52]
A
it's going to happen. So I'm curious about your thoughts about that. And then also, I mean there are a few industry initiatives are trying to kick out kickoff discussions about what kind of AI standards would look like. And I wonder if you have any, any thoughts about how they're, you know, addressing ownership and governance and all those things.
- [29:14]
C
Look, I think the thing to you're talking industry wide. I want to start with how companies could approach it first because I don't think any company. That's a bold statement. There's possibly some of the biggest companies, most mid sized companies in our space have not come to terms with the fact that they need to operate differently. They're used to having an IT group, a legal group, an ad operations group and they're all separate. And the kind of questions that we're talking about today as to where to control the risk don't exist in most companies. Those that are getting told by their board or by their CEO. You need to experiment with AI running a series of different pilots within the company, trying to automate as much as they can, as quickly, quickly as they can. But many of them don't have a good testing strategy for determining when something has been a pilot or an experiment. It should get promoted to production and what the controls are that you need around that. And so one of the things I talked about was companies need to get more real about what does it mean to be running these experiments, what does success look like when you actually have done a pilot with AI? How do you then broaden that within the organization with the right controls in place? So that's sort of one whole work stream around how do you use it and then separately how do you set up the group internally who is asking the kinds of questions I was asking in the talk around what could go wrong? Because to be blunt, that's an expensive role or series of roles to hire for. It's very specialized. It is going to be a cost center, at least in the short term. It's not going to be one where you point out and say that's revenue generating. And so it's the kind of controls that are not going to get put in place. And so given that I'm asserting companies are going to struggle with their own governance to start with and that's going to be a problem with that actually increases the need for more industry, whether it's through industry bodies or other mechanisms to take the lead on what are best practices, how should these communication protocols be used, what you know, all of these kinds of things. Because the biggest companies will do it right. If you're a Google and a meta or Amazon, you have hundred people focused on these types of things, maybe more, and you're going to get it right. The ones I'm most worried about are the ones in the open web, the publishers who, you know, they're focused on generating good content, they're not focused on AI risk. And that's where I see the problems being. And that's actually where I think the industry bodies have the most opportunity to rally different groups together.
- [32:19]
A
I want to maybe zoom out to get your thought on the media industry more broadly. I feel like media is kind of flailing, to say it nicely. It's been a really challenging time. I think kind of the end of programmatic page views and scale. Last week we were talking about the change of ownership with Vox Media and buzzfeed and others. So what's your take on kind of the media seller ecosystem, digital media platforms, like what's ahead and what's ahead, knowing what you know about where AI is going?
- [33:01]
C
Well, I mean the kind of companies you're talking about that tend to be more news oriented are not optimistic. When I think about parts of our industry that are growing and certainly where I work with most clients, it tends to be areas like retail media networks that are clearly growing. It tends to be ctv, otv, maybe even digital out of home. Like it's clear what where the formats are and it's clear where people are spending their time on frankly devices and apparently apps that are less news centric. And so I, I worry that people are wringing their hands saying look at news falling apart. And it is in many respects a downward trend at the moment. And what for a long time has held it up has been search has been, you know, sort of stronger than display advertising. And now search is coming under pressure with AI scraping a lot of the content and not even having that pull in the search revenue for a lot of these publishers. So the number of companies that face those kinds of headwinds I'm not optimistic, unfortunately about the traditional, what I'd call traditional media companies now. At the end of the day, people are spending more and more time online on their devices and the business models of advertising is a proven business model. So digital advertising I'm optimistic for. Traditional news. Digital media I'm not optimistic for.
- [34:43]
A
Okay, so a healthy environment for commerce platforms showing ads or other type of transactional platform showing ads. Not as great when you're, you know, here for the content, seeing ads on top of that. I think, I think that's great. I think that's the truth. That's sad, sad to acknowledge, you know, but, but there and hello, Ad exchanger, you got a new wrinkle for, for B2B in that environment. But no need to go into that there right now. James, any other parting thoughts to give our listeners?
- [35:18]
C
Well, I will say, look, these companies that are under pressure, whether they're traditional media companies who are struggling with how to implement AI, one of the things I would caution people to be careful of is cutting your way by assuming, cutting people by assuming that AI is going to replace them. And one of the things I talked about is one of the biggest challenges that we hadn't really got into is the idea that the experienced operator for what I'm going to call the short to medium term, meaning the next two years, is actually more valuable than they've ever been because they're the kind of person who can recognize where there are problems. When the AI starts to go off the rails, they're the one who's going to pick it up in a way that somebody who's new to the industry wouldn't because they've seen all the corner cases, they can pattern match off different experiences. And so that's where I think the most opportunity is for companies to show leadership is actually creating good governance structures internally and having those experienced operators be the ones who are frankly in charge or have responsibility for how those systems are set up as opposed to trying to cut those experienced operators to find efficiencies and that I really do think, you know, any publisher can have that within their decision making authority to lean into those kinds of things or back out and just accept that it's something that's happening to them.
- [36:55]
B
So you're telling me that layoffs aren't always the answer?
- [37:01]
C
Definitely not.
- [37:02]
A
I love that point of view and I happen to agree with you as well, James. So I think that's a great note to end on and one with a little bit more optimism amid a kind of mix mixed mixed forecast to be.
- [37:18]
C
Well, thank you guys. It's been my. Thank you for having me. It's been a real pleasure.
- [37:21]
A
Thanks, Jane.
- [37:22]
B
Thanks for joining.
- [37:27]
A
This episode was 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 omnichannel activation. Learn more@basis.com.