Technology
Services
Resources
In this episode, host Noor Naseer delves into the intricacies of supply path optimization (SPO) with Ian Trider, Basis Technologies' VP of Product (DSP).
Trider sheds light on the supply path’s crucial role in impactful programmatic media buying, and on how this strategy enhances transparency and efficiency in the digital advertising supply chain. Listen in to learn how advertisers can benefit from selecting optimal paths to reach their target audiences while minimizing costs and fraud, amongst other hot themes.
Noor Naseer: This episode we're taking a closer look at the supply side. There's a rising consciousness around supply path optimization (SPO) in 2023. While not a new concept, adtech players are presenting new offerings to remove redundant intermediaries and streamline access to supply. Ian Trider, VP of Product for Basis DSP joins me to share his perspective on SPO and some other supply-oriented topics including header bidding, and also qualifying inventory types alongside the IAB Tech Lab, and more. Let's get into this episode with Ian now.
__
NN: This episode this week we are featuring an inside guest here at Basis. It's Ian Trider. He's a VP of product, specifically working on the DSP. Ian, how are you doing?
Ian Trider: Great, Noor. How are you?
NN: I am doing well. I'm noticing a lot of changes internally at Basis and, of course, there's a lot of changes happening in the industry. Very curious for your perspective. You're known both internally and externally as a person who's a truth teller. So today I wanted to focus on things on the supply side, so we'll just jump into one topic and then get into a couple of other ones, starting off with header bidding. Header bidding is not a new topic in the industry—it’s been around for, I want to say, eight plus years, maybe introduced around 2015. For people who aren't too familiar, and they want to get acquainted with what's happening on the supply side, can you give me a definition of what header bidding is?
IT: Header bidding is a glorious messy hack. So, before header bidding, the idea of displaying the ad and asking for the ad were co-mingled. These were not separated in any way. So, there was no concept of being able to ask an ad exchange, do you have an ad for me for this ad slot? What's the bid worth? And then being able to make a decision based on that information. And header bidding solved for that and decoupled the idea of asking for something to show in the ad slot from actually displaying it. And so, in practice, header bidding involves some code that a publisher runs on their web page before their ad server actually starts and does anything, and that handles the part that requests from ad exchanges.
And the glorious hack part is where that client-side code sets some special targeting setting in the ad server. So that even though the ad server doesn't know about header bidding, it makes it possible through some roundabout means to cause the ad server to sort of trick it into valuing that. Suppose it's a bid coming in from an exchange like OpenX, it allows the ad server to know that OpenX is offering a bid at a $1.25 or whatever the case might be, and it enables a holistic auction. It enables full, actual real-time price competition between any exchanges that you're integrating via header bidding between Google's ad exchange, and also, with anything that's booked directly in the ad server.
So, header bidding completely changed everything. It caused a lot of changes for the supply side because the interesting thing there is that suddenly there's very little differentiation to be found between ad exchanges for publishers. I mean, all of the exchanges are getting to compete equally in real time. So, as a publisher why do you care who is offering to fill the impression? All notable ad exchanges are integrated with all the same DSPs. So, it caused a lot of shake up on the supply side.
NN: Are there any potential drawbacks or challenges associated with header bidding? Who would they be for? I think your answer already clarified this to an extent, but just to hear your follow-up take.
IT: Yeah, I mean there's a couple of different ways to look at that. In the technical sense you can look at it, which is it needs to block display of the ads for a short period while this auction happens where all ad exchanges are solicited for the particular opportunity. And that's happening from the user's web browser. So that puts some burden on their device, consumes some bandwidth, but it also means a delay before those ad slots can load which can have some negative effect for the publisher in reducing the number of impressions that come to exist. Although the extent of adoption of this, I think indicates that publishers see the benefits as outweighing the costs. And there's different ways to mitigate that too. There's technology available that allows that to be moved completely away from the client side.
Certainly for ad exchanges it means that they need to find some other way to differentiate themselves. Because when a publisher can ask in real time for “What's your best price, OpenX?” and “What's your best price, PubMatic?” there's no reason for a publisher to not just plug as many of them in as possible. Right? There's very little opportunity cost for the publisher: They can simply plug in more and more and more of these and, in fact, they do. That does lead to an interesting side effect, though, for the buy side, which is if a publisher plugs in six ad exchanges and they're doing that because they want to try and maximize the opportunity that they'll get a valuable bid. But the problem is that means six bid requests going to DSPs. Six bid requests that all of those exchanges have to send out. But six bid requests are going to the DSP for what is actually one single impression opportunity. So, in that regard there's a fair bit of waste involved.
NN: And when you say “waste,” how would you qualify what the metric of waste is?
IT: Bandwidth consumed, servers needed, resources needed—because a DSP has to have servers running that can listen to that bid request traffic. And it's kind of illogical to listen to six bid requests that are for exactly the same opportunity and for which the ability to bid and win on is the same. Because, again, if it's being evaluated in real time, if it's solely a question of “best price wins,” then it really doesn't matter which of those opportunities the DSP bids on. It can win it or not all the same, depending on what's in competition. In that regard, it's wasteful.
NN: You mentioned a little bit earlier, Ian, that there has been incredible adoption of this. So, can you say a little bit more about what adoption has looked like? Would we say that most of the industry has adopted header bidding in some capacity?
IT: Yeah, the best source I know of that is—actually, I think it's discontinued now—Adzerk, which is now Kevel, was publishing a header bidding index where they tracked the adoption of header bidding across—I think it was the Comscore top 500 sites or somebody's top 500 sites. Anyway, the vast majority of major sites are now using header bidding, often with many different exchanges plugged in. So, I would say that at this point this is absolutely the industry norm for how publishers handle integrating with exchanges, publishers of any notable size, that is.
NN: And have there been any updates in the header bidding space? Has there been ongoing conversation about the iterative improvement of header bidding in 2023?
IT: Keep in mind header bidding, like I said, it's an incredible hack. All of this happens without the publisher ad server, and I'm talking about Google Ad Manager because, de facto, that's the only publisher ad server that matters. It's got probably 98% plus market share. This is happening without any explicit functionality in Google Ad Manager to support it. It's a workaround. At the beginning there was not any sort of standard framework for doing this. The way you set up header bidding with each ad exchange was totally customized to that exchange.
Header bidding wrappers came to exist that standardized this, with the most popular being prebid.js, which is an open-source framework that is supported by all major exchanges and a lot of minor ones. And it is now the most common way publishers configure this, by using the prebid.js code and hooking up the adapters for the various exchanges they want. It makes it relatively easy to implement header bidding.
Further advancement includes moving this to the server side. Instead of having the user's web browser make the request to each and every ad exchange, what if we just have the user's web browser make one request to a server, and have that server make the request to all the ad exchanges? And that is facilitated in a couple of different ways. But the open-source sort of role would be with pre-bid server, which is a server-side implementation. It's something that a publisher can host or pay somebody to host that moves all of that to the server side. So, it is indeed the case that the web browser makes one request to the user's web browser and one request to the publisher servers. And then it fans out to however many exchanges they want to go to from there. But the interesting thing about that is before, when it was all client-side, publishers would have a practical upper limit on the number of exchanges that they could plug in. Because there's only so many things you can try and get a user's web browser to do simultaneously, especially if it's on a slow connection. But if you move that server-side it now becomes nearly infinite. So, this is the thing I was mentioning about the duplication of bid requests, the fact that a DSP could see like six requests for every one impression. In that regard, this server-to-server stuff does make it worse because it sort of removes the shackles of what the end user's device can feasibly do.
NN: I want to extend from an introductory conversation around header bidding. Not that anything with you, Ian, ever feels introductory. I know you really get into the weeds on everything that you're knowledgeable about. But header bidding is one aspect of supply path optimization. Maybe we weren't calling it that back then. But now there has been this uprising very recently, in the last two plus months, around this breakage in the programmatic supply path with SPO. Can you give us an introduction to where that started and where there's been this link breakage in what has otherwise been a pretty tight-knit chain that's been around in the programmatic media space for quite some time?
IT: Certainly. What do you mean by “link breakage”?
NN: That we're going directly to publishers for this inventory and taking out some of what would be defined as a middleman.
IT: The interesting thing about that is if you ask five people for a definition of supply path optimization you will get five different definitions. The way that I would define it is about removing unnecessary indirection in the supply chain. And by unnecessary indirection I mean more people involved in the supply chain than is absolutely necessary. This ties back to header bidding because again header bidding made it possible to solicit all of the exchanges simultaneously, so there is no such thing. Nobody has preferred access; now all of the exchanges have equal access to all the same inventory via header bidding. So, at that point it's even a question of if the publisher has six exchanges integrated via header bidding, what's the point in listening to the traffic from all of those?
But beyond that, there's also indirect supply paths. A surprising number of companies in the ecosystem are heavily dependent on reselling supply to some other larger company. And you can see this in what happens from the publisher’s perspective with ads.txt. So, ads.txt of course being the specification that IAB Tech Lab rolled out to try and quash domain spoofing misrepresention—the ability for people to claim that they have inventory for some domain and just lie about it, ads.txt greatly mitigates that. And it requires the publisher to list all of their authorized supply paths. And a sort of tier-one, A-player sort of exchange will not be asking for anything in that file, other than one line representing that publisher's account on that exchange. But once you get below that you'll find that the sort of smaller players will ask for a large number of records to be added, because they want to or rely on reselling the supply. They lack many direct integrations with the DSPs and instead they rely on reselling the impression through larger exchanges.
The amount of bid request duplication that a DSP can see as a result of that actually gets quite extreme, and it also introduces a lot of problems for fraud and other issues, because when an impression gets resold through many different parties it tends to obscure where it came from. Knowing where the traffic came from is absolutely essential to knowing that it's genuine. Moreover, if it passes through a lot of different hands, all of those hands have to get paid right. So, in any supply chain that is highly indirect, the adtech tax is much, much higher.
I see supply path optimization as an exercise in removing redundant, unnecessary, duplicative supply chains—inefficient supply chains. An example of that is, suppose we're directly integrated with an ad exchange. Maybe it's one of the smaller ad exchanges, but we're directly integrated with them. But they also hold an account on three or four of the major ad exchanges and resell all of their traffic through those major exchanges. As we implement supply path optimization, we are directly integrated with that exchange, so we don't accept the traffic from their accounts on the bigger exchanges. There's no point. It's exactly the same traffic we're seeing directly, but with less transparency and more adtech tax because that bigger exchange has to get paid. So that's the angle that we've focused on, although there's definitely multiple different interpretations of what supply path optimization should mean.
NN: I want my next question to be about who's most impacted, but I think if I was listening to your answer carefully or not so carefully, you made it pretty clear that the core impacted parties include anyone who's selling it from a duplicative standpoint. Are there other parties who are also impacted in this shift?
IT: Yeah, well a very aggressive supply path optimization stance could attempt to deal with that scenario where the publisher is integrated with six exchanges via header bidding. This is not even forgetting about the indirect resale of their supply; we're just looking at direct exchange integrations. If you wanted to get really aggressive with the supply path optimization you could try and decide which of those six paths to listen to, because they should theoretically all be equivalent. Some of them might be less expensive and have lower adtech tax than others. This is not a level of supply path optimization that we're doing at the platform level. You start to get into this line where it starts to tamper with what the buyer would like to control over. And, so, we take a little less aggressive stance than that. But you could go that far. If you take it that far then it potentially affects all of the supply side players and has a likelihood of pushing some of the bit players out.
NN: So I want to explore a little bit about what you said further, As far as analyzing or evaluating who to go with if there are six supply partners that you could be working with and you're evaluating the fees and potential cost implications. What does that evaluation process look like for you to minimize duplication and just bring more efficiency to the overarching platform?
IT: One of the other reasons why we don't do that type of “figure out which of those six direct supply paths to listen to” is because fundamentally it's nearly impossible to actually reliably do that. The problem is there's no way to know that the six bid requests that you're receiving are all for the same impression and there's no way to reliably analyze that multiple supply paths are literally sending you the same opportunities. There's lots of subtleties in this: A publisher could work with some exchanges only for certain geographies or for certain ad formats. So there's a lot of complexity in that. And there are some efforts that have been made to try and give some more information that helps with this like a global placement ID, like the idea of a global transaction ID. Neither of those things is anywhere near universal though.
In practice, it would even be considered the question of fees. The fee information is not something that the buy side has. The buy side does not know precisely what the publisher is being charged by the exchange. Certainly, you could go ask the publisher. I'm sure there's plenty who wouldn't mind telling. But it's possible to see that indirectly: For example you could look at the win rate and you could say, “Well logically the exchange with the highest win rate must have, from the perspective of the buyer campaign, the exchange with the highest win rate must have the lowest fee because they're sending the highest effective bid into the publisher's auction.” But there's subtleties in that that interfere with doing that sort of analysis at a platform level. There are subtle technical differences in the way the exchanges operate. For example, some of them tell the DSP upfront what categories and advertisers are blocked. Some of them don't. And so, because of that one that doesn't provide that information, the DSP will bid not knowing that an ad is going to be blocked, and that's something that affects the win rate right. So, there is a surprising amount of subtlety in this. And it starts to get really risky really fast to try and duplicate those direct supply paths.
So, at present we're not attempting to do that. We're just addressing the obviously, clearly indirect supply paths, intermediaries, and other low-value or questionable sellers.
NN: How would you say that that philosophy that we're using at Basis compares to what's happening on a larger scale in the adtech industry? There are multiple adtech players who have really created a lot of noise around the development of products or solutions that are focused on—like there’s this nomenclature around this new SPO release. So how do we distinguish what we're doing from what other folks are doing?
IT: The thing about this is it's not necessarily bad for the buyer for a DSP to more directly integrate with the publisher, but there are a couple of considerations that you need to keep in mind. When there's this separation of concerns where the DSP is clearly working for the buyer and the SSPs are clearly working for the publisher. Up until a couple of weeks ago I was the vice president of RTB platform operations, so I was directly involved in the supply path optimization endeavors or anything to do with supply quality. And it's not a problem you can just block: So you know if there's fraud, or if there's some sort of supply quality concern of another sort, or it's an indirect supply path you can just block it. You don't have to ask anybody. It's real time bidding. A DSP doesn't owe anybody on the supply side anything and it doesn't have to bid on anything. However, if the DSP starts working directly with publishers, now there's human-based complications. Now it's not as easy as just blocking it because there's a publisher that you have a direct business relationship with that's expecting you to be buying their impressions. So, it starts to really muddy the waters in a way. There’re ways that it can be managed but it's something that a buyer would be warranted to make sure that the DSP has carefully considered so that the DSP isn't making choices that aren't aligned with the buyer’s interests.
NN: If you were a media buyer, Ian, is there anything else worth knowing as a buyer as it relates to the progression of work from an SPO perspective from the DSP that you work with?
IT: Yeah, it's not so much about the DSP they work with, but I would express a lot of skepticism about ad exchanges coming and offering SPO. Because the thing is, the nature of SPO in the sense I described of removing indirect supply paths can't just be a matter of cutting a deal with an exchange and giving all your money to them. How does that ensure that that's the most efficient supply path or that it's removing duplicate supply paths? So honestly the amount of subtle data that goes into analyzing this stuff is extreme. And I think that buyers are best served by working with DSPs that have good platform-wise policies, because as a buyer you would absolutely lose your mind trying to do this yourself. First if you can even get the data, because the necessary information for this is not typically in DSP reporting. You can probably get it out of log-level data from DSPs, but then you're going to need an ads.txt crawler and you're going to need a seller's json crawler. And you're going to need to parse supply chain data. And then you're going to have to have a bunch of analysts pore over it for a huge amount of time. And you could do that, or you could just let the DSP take care of it and do it once across all the DSP's customers.
NN: I also wanted to talk to you today, Ian, about defining inventory. Something that you've worked on as a part of IAB Tech Lab has been the redefinition of some supply types and one of them being in-stream. Can you tell us a little bit more about your involvement there and how you've helped define this one supply type?
IT: Look, here's my position: The definition of in-stream has fundamentally not changed. The definition of in-stream is an ad displayed before, in the middle of, or after some video content that the user chose to watch. Now, the problem is that there are some companies that have been really stretching that definition. There's a lot of ad space on the internet that's being called “in-stream pre-roll” that I don't think qualifies as content that the user chose to watch. I mean, when somebody says in-stream or when you talk to a buyer about in-stream, they're probably thinking something like YouTube, ads before some streaming video content that the user meant to watch..
They probably weren't thinking or expecting that it's going to be some floating player that pops up in the bottom corner of the screen on a recipe blog that just shows some unrelated video content that they never asked for and plays an ad before it. There's a lot of inventory being circulated that's being labeled in-stream that fits that definition. So, the adjustments to the definition are intended to clarify and bring us back to what in-stream is intended to mean. And put things like what I described into their own sort of category which is being called, “accompanying content”. Hopefully this should clean up and ensure that when buyers are bidding on in-stream inventory they're actually getting what they expect instead of these other formats that are only in-stream if you squint really hard at them. But it does mean that if people are expecting to get large volumes of in-stream video inventory at $7 CPM it's just not going to happen.
The real in-stream inventory is inherently scarce and valuable and it will trade at those prices. But if that other format I described is of interest or desirable to buyers in the future, it'll be possible to explicitly target that format now. So, the inventory is not going anywhere, it's just being more correctly labeled.
NN: Your focus as of late has been on in-stream, or recently has been on in-stream, but there is probably an issue with the definition of a lot of different inventory types. When did you really start picking up on the fact that there were people putting things into boxes that they didn't belong in? I'm sure it's been happening in the industry for as long as it's been in existence. But when did you decide to really get involved?
IT: This conversation has been going on for a while in the Tech Lab working groups. And I don't want to speak too much about what happens inside there because it's private business until public release of something. Broadly, people lying about things is an ongoing, fundamental concern in real-time bidding because somebody once described it to me as trading bites for money. A publisher sending a bid request series of bites of information. And depending on what you set in there, depending on which bites you send it, affects how much money somebody's going to give you. And so there's a natural incentive to convey the inventory in a way that'll maximize what people are willing to bid on it. The problem is when that's not true it starts to look an awful lot like fraud.
NN: That's the thing about fraud. We haven't talked about fraud in particular today, but I think there are some gray spaces and maybe redefining the terms of what something a client thinks they're buying is, versus what they're actually buying—it falls into that kind of gray space. So, I also wanted to ask you about other types of gray spaces that exist, including MFA or “Made For Advertising” websites. What are your thoughts there? I feel like I'm encountering MFA websites all over the place. They're not great environments for me as the user. But I think a lot of advertisers and marketers are unintentionally buying in those spaces. What do you think the advent of this is? Are clients not being advised? Do they not know? Can you give us a little bit of backstory on just MFAs in general?
IT: First, I'd say it's not a new phenomenon. This has been around forever, and, in fact, is fundamentally nearly the entire online publishing industry—I mean all publishing is made for advertising, if you really think about it. Like, the point is to sell inventory. The point is to bring eyeballs to the site and the site has ad slots on it and you sell ad inventory. But of course, that's not what anybody means by “made for advertising.” They specifically mean sites that are not really intended to deliver real value to the user, that are almost always—like nearly exclusively—visited by paid traffic. So, users arrive there by clicking on paid links or things like that.
The value of a site is largely in its audience. People want to put ads in the New York Times because the New York Times has a certain audience, and that audience is valuable to advertisers. So, one of the problems with these made for advertising sites is they don't have any endemic audience. You can't say, you know, “here are the sort of people who visit the site.” It's just whoever clicked on the link in the recommendation widget. That doesn't mean that is valueless inventory. It's just that it's not got a lot of innate value. I mean, if you sprinkle some third-party audience data over it, it might have some value. But by itself, it's just sort of filler. It's just a sheer sort of tonnage.
Related to this is the problem with this paid traffic sourcing, is that it's a very hard arbitrage. The objective here is you're going to buy eyeballs into the website and get people into the site. You want to make more money off the ads you serve to them than it costs you to get them to the site. That's really hard math to make work. One of the ways some of these sites do it is with the ridiculous slideshows, where they suck you in with some very sort of shocking or interesting thing in the thumbnail in the recommendation widget. And after you've clicked through 20 pages you find out that it was never a part of the slideshow in the first place. But they've now served a ridiculous number of ads to you.
One thing that a lot of sites end up doing is resorting to very questionable sources of traffic. There's sort of a hierarchy of paid traffic sources and up at the top are things like Facebook traffic from Facebook ads and things like that. And then some of the higher-end native ad or like recommendation widget-type sources. But things start to go downhill very fast all the way to the point of just absolute blatant bots, because it is all about that arbitrage of “How do I make more money from serving ads than it costs to bring this person to the page?” And because that math is so tricky, it drives the owners of these sites toward the dodgier and dodgier traffic sources. And as a consequence, these sites tend to have a lot of invalid traffic, or it's certainly not high-value traffic in the best case, but in the worst case outright bots.
NN: How often do you think the supply side or generally speaking has there been an uprising of rejection of these types of sites? Or do sometimes people try to create the opposite case of what you've said? You’re saying there isn't an inherent audience; some people might say, “No, but people are intentionally clicking on them.” And they might be seeing some value. They just have to do some extra labor, and that extra labor implies there's more clicks. So, people might be coming up with their own use case just to go back to the original question. Like has there been this rejection or the opposite?
IT: I'd say there's an increasing rejection of this by the buy side. The sell side, on the other hand, it is important to keep in mind that nearly everybody gets paid a percent on what gets transacted. For an ad exchange, there's a natural incentive to add more and more inventory. More inventory means more money gets transacted, which means they make more money because they get a percent of the spend. And that's true for DSPs as well. But the thing is DSPs are exposed to the entire universe of inventory. It's a very big universe, it's too big of a universe. So, the incentives aren't quite the same. The DSP doesn't need to listen to all this. The DSP is probably looking at their hosting bills and going, “Oh my God how am I gonna pay these bills? They are getting out of control.” So, there isn't the same incentive to tolerate the traffic.
Then, as for buyers, I think it's going to vary. I think you will find buyers that are perfectly happy with it. The buyer incentives aren't always necessarily aligned with the advertisers either. So, it's sort of just a fundamental concern of this industry. I would say the interest in doing something about this is largely coming from the buy side.
NN: That's fair. I am curious what you think is going to happen in the future as these gray spaces continue to mutate—if it's not in-stream tomorrow, it's going to be something else. And if it's not MFA, what do you think is going to be the next mutation?
IT: The bad things always trend toward the novel spaces with less verification and more buyer excitement and higher CPMs; in other words, connected TV. The risks in that environment are substantial. I very strongly encourage people to heavily scrutinize precisely what they're buying and who they're buying it from, but it's the holy grail because it's got high CPM, scarce supply, very little actual verification of what it's served to or where it's served is technically possible, which is just a perfect storm for bad people to exploit. But I think we'll probably see this with other emerging formats, too. I think you can probably expect that problems will come up with audio, even digital out-of-home, perhaps. Digital out-of-home, for example, one simple way is there's this idea that one display of the ad is worth ‘n’ impressions. Where ‘n’ is more than one usually. Where does that number come from? And on a reputable exchange where that comes from is some sort of auditing organization like Geopath, that's actually physically audited and observed and done traffic counts and come up with an estimate of how many viewers that's expected to have. But in the worst case it could just be a number somebody makes up and if you make that number bigger you make more money. So, that's something it'll be necessary to watch out for in the future.
NN: Ian Trider is the VP of Product on the DSP at Basis Technologies. Ian, I know you have a lot to share and there's going to be more conversations to be had, so thanks for the time.
IT: Thanks, Noor.
__
NN: Thanks again to Ian Trider, VP of Product for the DSP at Basis Technologies. I'm planning to feature Ian on a lot more episodes because he's very knowledgeable about intricacies in the adtech space, particularly aspects that not everyone is promoting and publicizing. And really, that's what we're here to do: to learn more about the adtech space and some of the things that are not always being discussed. So, until then, this is Noor Naseer. More Adtech Unfiltered real soon.