Earlier this year, SPAC transactions brought open internet ad manager Taboola (NASDAQ:TBLA) and mobile app monetization platform ironsource (NYSE:IS) to the public markets. The ION team behind the Taboola deal announced a further combination in the space in June that would list smart-TV ad manager Innovid via ION 2 (NYSE:IACB).
Each of these deals aims to boost a player that has developed a unique business model that stands to benefit from changing tides in digital ad consumption in different ways. In AdTheorent’s case, it comes to a market with tools for efficient digital ad targeting that do not rely on personal data or cookies at a time where privacy concerns are at the forefront of the technology discussion and cookies are on their way out.
Google (NASDAQ:GOOGL) and Apple (NASDAQ:AAPL) earlier this year announced moves that drift their sizeable portions of digital ad real estate away from cookies. SPACInsider caught up with AdTheorent CEO Jim Lawson to talk about what all of these changes mean for the industry and what companies are doing right and wrong to adjust.
The Q&A transcript below was edited and abridged for clarity and space.
SPACInsider: What I found interesting looking through your materials was that you really have to understand the tectonic shifts that are going on in the digital advertising space right now in order to understand AdTheorent’s place within it. And so, just to start off with that in mind, do you think you could give just a quick lay of the land in terms of the digital ad space as you see it from your vantage point?
Jim Lawson: I think you’re spot on with that question because I do think that the ecosystem is going through a number of changes. The advantage that we have in the marketplace is largely driven by two things: one is the power of machine learning, data science-powered ad optimizations in the programmatic context; the other is the shift in perspective on what is okay for ad targeting and what is okay for the use of data for ad targeting. What are the privacy expectations of consumers? Brands don’t want to be on the wrong side of that conversation and brands don’t want to be utilizing personalized or individualized data in a way that is not consistent with the current expectations of consumers. And, they don’t want to alienate their consumers by using data in a way that’s crossing the line.
I think what you’re seeing is a recognition that Facebook – they know everything about you. And, we learned a lot about that in the last couple years. They know everything about you and they’re not afraid to use it. In the open internet, where we advertise, a lot of the historic methods for ad targeting were cookie-based, which obviously is behavioral advertising. The essence of behavioral advertising is: you visit a given property, and then a piece of code is dropped on your device and then you’re essentially tracked on the internet. Or, audience-based advertising, which essentially consists of licensing profile data, which has IDs appended to it. And then DSPs would leverage those IDs — whether a cookie ID, or a device ID that they license or otherwise obtain –and they would leverage those IDs to target ads in a programmatic environment.
The reason why they do that is because it’s the easiest way to do it, and, frankly, it was the only way to do it for a long time. AdTheorent came around in 2012, attempting to solve for that.
SPACInsider: As you mentioned there, the displacement of cookies feels like it’s going to be a major determinant of the size of your available market. Google and Apple seem to have already basically thrown down their gauntlet in the cookie space. So, do you have a specific prognosis for where you think that is headed in terms of what share of the digital ad market will be made up by cookie-based ads, maybe a year or five years from now?
JL: Well, one thing I would say is I don’t think that our market share entirely depends on the deprecation of cookies because even now – let’s say for the next year and a half, cookies are available. We can utilize cookies. We can utilize them in a different way. If a customer wants to do a re-targeting campaign, we can do that, but we don’t rely upon it for targeting. We just don’t view it as special. The reason why they delayed this transition is because regarding the consortiums and industry groups that were coming together to replace cookies with more aggregated information, I think a lot of the industry leaders realized that that was just going to take longer. And, if they were to pull the ripcord too quickly, there could be chaos in terms of the attribution and measurement capabilities related to the API connections that the industry groups were creating to essentially communicate conversion activity back to advertisers. Without that work being done in the industry, the transition away from cookies would be quite chaotic and I think it wouldn’t be good for Google — I mean, they have a big advertising business. It wouldn’t be good for the open internet. I don’t think consumers would be happy if there were paywalls put on content all across the internet because advertising was essentially shut down because of the inability to make a responsible transfer from one world (the cookie world) to another.
So, I think it’s prudent that they slowed that down. I think we at AdTheorent are ready at any moment for the cookie to go away. We have a number of advantages because of the way that we use data and don’t use data. We optimize our campaigns based on a number of different statistics that come through a bid request and the bid streams in programmatic. It could be things like: the keywords in the URL, it could be the size of the ad unit, it could be the age of the phone or the type of the phone, it could be the publisher category, all these other factors. And when conversions happen, our machine learning platform assigns predictive scores based on the likelihood of a future conversion occurring based on what we’ve seen in the past. We don’t need to know who any user is, we don’t need to know that a specific user visits these websites or lives in this town. We just look at a moment in time, a person accesses a piece of digital content. There is a microsecond where AdTheorent and other DSPs are deciding whether to bid on that digital real estate. And, what we do is we say: “The bid request has this data in it. What do our models tell us? Is this going to be valuable for a bank trying to sell a credit card or is it going to be valuable for a quick service restaurant trying to get somebody to buy a hamburger?” And our models are there to tell us that in real-time and microseconds. And, it’s just a better way of targeting.
SPACInsider: In that way, it seems like your AI and machine-learning based ad systems were almost tailor-made to hop into the market at a time when cookies were exiting stage left and brands would be looking for efficient ways of doing segment-based campaigns. So, did you prophesize this or is this somewhat more of a result of developing a tool that just happened to be right for the job as the market changed.
JL: That’s a great question. I would love to be able to say that I prophesized anything. It’s more a combination of factors. Our business started in 2012 around mobile. So, we were confronted with the problem of how do you advertise in digital without cookies? It wasn’t necessarily because we thought cookies were going to be out of favor in 10 years. It was more so that they just weren’t available in mobile at that time. When we started out, we realized the effectiveness of machine learning in that use case, and then we immediately pivoted our business after a year or two to be an omnichannel provider so we could bring the power of machine-learning optimizations, not just to mobile, but the desktop and to all the screens.
So, it’s combination of factors. But, for us, we knew early on that there needed to be a way to target without all the taint of behavioral advertising. Behavioral advertising can be used effectively and responsibly in some cases, but in other cases it can cross the line. Things like re-targeting a person who visits a website for a very sensitive healthcare condition is against the rules under the NAI (Network Advertising Initiative) Code. I don’t know whether that’s happening at a large scale or not, but it’s not happening at AdTheorent. And, if we do any kind of re-targeting for a customer, we’re always sensitive to those types of concerns, but, again, that’s not our special sauce. We believe that also with re-targeting, you’re really not expanding your audience. You’re advertising to consumers who have already indicated an interest in your product at some level. So, you’re really just operating within a much smaller universe. Predictive prospecting, as we call it, is where we can cast a wider net and allow the machine to kind of help us refine that. So, we think it’s just more effective.
SPACInsider: It also seems like basically anybody who’s done any online shopping has had the effect where they buy something and then that product then like stalks them in banner ads for months afterwards. But, you’ve already bought it, you’re not going to buy it again.
JL: Exactly right. There’s an inefficiency there, like, who benefits from that? I guess when they just really hit you over the head with something that you already bought it’s kind of annoying. It’s not adding value. And, you know, I think it’s really a good example of the limitation of that method.
SPACInsider: And, in terms of how your engine works, the presentation noted that it ingests about 200 data points on each impression and those are all essentially public data points rather than data on the individual. You mentioned a few of what those data points are in the presentation. One that stood out to me was the local temperature at the time of the conversion. Obviously, the machine-learning does this stuff on its own but I’m sure you’ve seen some insights as well so what does something like temperature tell you?
JL: That’s the beauty of machine learning. You have no idea what factors, atmospheric factors or otherwise, may or may not contribute to conversion lift. People can’t figure it out, there’s just too much data involved. We’re sometimes very surprised when we look at a campaign after the fact and we see the data on the conversions, and, for whatever reason, in Denver on a Tuesday there may have been more conversion activity for a particular thing. That’s something that we wouldn’t be able to know in real-time if we were looking for it, but the machine learning system can detect it in real-time. It takes conversion activity for the models to learn. So, in the beginning of our campaigns there is a learning period. In the very beginning of the campaign, before there are enough conversions, we optimize towards where the clicks are coming from. And then when we have enough conversion events, we optimize towards those data points where the conversions are coming from — so it’s kind of a learning process until you have enough conversion data.
SPACInsider: Interesting. And are you able to share any other data points that you collect that may seem idiosyncratic but maybe you found produce some insights.
JL: You know, there’s so many of them on there. I’m not sure which ones are outliers or idiosyncratic, but there’s a lot of different variables that you might not think of like: the placement of the ad on the page, operating system of the phone, the age of the phone, whether it’s a Samsung or Apple device or all those different factors; whether it’s the app environment or the web environment. We also have points of interest baked into our platform so we know where all the public libraries in Seattle are, and we know where all of the Home Depots in Brooklyn are. And, we’re able to leverage that point of interest data to make location data more contextual and valuable. We may realize that there’s a lot of conversion activity coming from Starbucks locations in a given town. They can be intuitive like a Starbucks and they can be maybe less so like the waiting room at a Midas car repair place or something along those lines. You might not know where those conversions are coming from, but the point of interest database makes the geography a little more digestible.
SPACInsider: It also appears that your revenue on ad placements on smart TVs and gaming consoles is now outpacing broader revenue growth and this seems like a market that is like a surface that’s just being scratched upon. What do you see in terms of the increases in Smart TV adoption and how the digital ad market will develop on that real estate?
JL: We see the CTV opportunity as being explosive. Our year-over-year growth is 300% on CTV. Our investment in that has been minimal to date, other than making sure that we have the appropriate scale and access to the CTV inventory, but our bread and butter as a business is value-added capabilities. And, we’re just in the beginning stages of that. One of the things that we can do that we’re proud of is that we’re investing in more in the future is post-view conversion measurement. So, when you watch something on your connected TV, if you then take an action, we are able to essentially tie that back to that viewership and then be able to use that in our machine learning. So, we can make television viewing more of a feedback loop so that our performance models can work in that context as well.
SPACInsider: What do you think could be the next big thing that could shake up this space even more? What are you tracking?
JL: I do think that we are leading what will be an evolution of ad tech towards more of a data science and machine learning-powered targeting world. You see, the rhetoric exists on a lot of websites, and you see it in a lot of marketing materials, but you don’t see it in practice very often. I think that the reason for that is it’s very difficult. It’s very difficult to operationalize in an efficient manner. The use of data science in microsecond to target, it’s just not easy to do. It’s taken us a very long time, we have a great team of data scientists and engineers that have helped us build this in-house since 2012. We’re not the product of a number of acquisitions or tag-ons — we’re just the same group of folks and we’ve had the same CTO, who’s extremely gifted, since 2012. I think the industry is headed in that direction.
A lot of companies are in the process of working to figure out what we call backfill IDs or IDs to take the place of cookies. We believe we already have that answer. We’ll support the backfill IDs, because we think the open internet benefits from having IDs that are that are privacy-friendly and used responsibly. But, there are questions about scale and the level of adoption that some of those other ideas will have, so we will support it, but we don’t depend on it.