There is a silver lining to the challenges wrought by the passage of Apple’s App Tracking Transparency. This new, probabilistic world has created opportunity for startups like Treat, which just announced its $8.5 million seed round led by Greylock. The company uses generative AI to create personalized product photos for commerce brands.
We’re also seeing startups like Fermat, which provides the tooling and network to enable distributed commerce. Founded in 2021, Fermat saw that user-generated content was drawing the most engagement to brands, so it built the mechanism for companies to embed shopping experiences within influencer content. Greylock led the company’s seed round in 2022.
Treat CEO and co-founder Matt Osman, Fermat CEO and co-founder Rishabh Jain, and analyst and investor Eric Seufert joined me on the Greymatter podcast to discuss the post-ATT world and what it means for commerce brands and the companies building tooling to support them. You can listen to the conversation at the link below or wherever you get your podcasts.
EPISODE TRANSCRIPT
Mike Duboe:
Hi everyone, welcome to Greymatter, the podcast from Greylock where we share stories from company builders and business leaders.
I’m Mike Duboe, a partner here. Today I’m excited to welcome Matt Osman, Rishabh Jain and Eric Seufert.
Guys, I’ve been really excited for this conversation. I think for those of us who have worked in growth marketing or e-commerce the past couple of years it’s certainly challenged our assumptions and our learnings along the way.
For some advertisers, this has unfortunately meant that growing a business is no longer viable and has been catastrophic to some. For other marketing friends, we say things have been getting fun again.
And so the goal for this conversation is really to help illuminate what is exactly happening in the world of performance marketing with a specific lens around e-commerce. Ideally, founders and other builders in this space will walk away with an actionable set of lessons.
Our guests today come at this problem from a different set of vantage points. I’ll do quick intros and we’ll ask you guys to do the same.
Eric’s an independent analyst, consultant, and investor at Heracles Capital. He’s really become the go-to follow on all things ad tech. We had the pleasure of meeting back in the day when I was at Stitch Fix and he was at Network working on mobile advertising and programmatic.
Matt Osman is the CEO and co-founder at Treat, a Greylock portfolio company using AI to generate, personalize, and deploy creative for large e-com merchants.
Rishabh Jain is the CEO and co-founder of Fermat, another Greylock company building a distributed commerce network, helping brands drive conversion at the point of discovery.
So guys, thank you again for joining. Before jumping in the conversation, could each of you take a minute to share your background and maybe what led you to dedicate this chapter of your lives to this set of problems? We’ll start with you, Matt.
Matt Osman:
Yeah, so this is my second company. My first company was a natural language processing company and sort of the first wave of AI, but specifically in biotech and biotech document processing. So I kind of got to know the AI lens of this through that. And then I got particularly fascinated in the ways that language becomes, basically, a summoning technology with LLMs and diffusion models.
And there were some very interesting applications in customer acquisition post-ATT. So it’s almost a continuation from the last company from a technological perspective. And yeah, it does feel like building in the age of miracles. I would like there to be one week where there isn’t an industry-changing release or announcement because I need a break, and it’s starting to get quite exhausting.
MD:
All right, Rishabh.
Rishabh Jain:
Yeah, so in addition to the current company, I would say that the lens that I come at this with is from a company called LiveRamp. So I was there for six and a half years and it was basically the largest provider of digital identity. And so for better or for worse, we knew everything there is to know about online tracking.
And so when Apple first made its announcement, we were sitting at LiveRamp saying like, “Holy smokes, Eric Seufert is right.” At the time I didn’t know him. And basically, the writing was so clearly on the wall for me, starting in the summer of 2020. And then for sure, by that December, and I’m sure we’ll go through the history, but that’s when I basically decided to leave LiveRamp. This is like a change that is gonna affect the entire consumer internet, but the first place in today’s topic is gonna be e-commerce, right? And so, yeah, it was just that vantage point of building technology for tracking in the open web.
MD:
And Eric?
Eric Seufert:
Yeah, so first, I appreciate the invite. It’s great to be speaking with such an accomplished and impressive panel today.
I come at this as an operator. So I spent my career in performance marketing for mobile apps, mostly games, kind of grew up in the sort of primordial soup of the app store and kind of got that front row view of the evolution of Facebook and games, Google and all of these powerful technologies that were brought to bear in the service of distribution – distributing apps, getting products, physical goods into people’s hands.
And so I was concerned and troubled when ATT was announced, although I think just in terms of the impact and the preparedness of companies to adapt to that.
In contrast to Rashab’s journey, when that light bulb moment happened for me, I didn’t start a company. I started a venture fund to invest in the brilliant founders, building the tools that would sort of enable companies to adapt to that like, which I’m very pleased to be an investor in.
MD:
So let’s rewind back to around 2016. You know, Facebook ads were hitting their stride for many of us who were starting to deploy them at scale in e-commerce. Shopify was on its rapid ascent kind of on the heels of this. I remember at the time, like the OCPM algorithm and Facebook targeting became so darn good that merchants could really hit scale with very few people in house.
And a few years later, this all kind of just blew up. Many point to Apple’s privacy efforts in catalyzing this wave of change. Eric, you just referenced ATT. Maybe take a step back. Can you explain to us what happened here?
ES:
Yeah, well, so maybe it’s the best place to start here is 2017. So it’s ITP, right? So Apple introduced intelligent tracking prevention in 2017. That’s a Safari browser feature that did a couple of things, but the most prominent sort of privacy safeguard was blocking third-party cookies. So Apple rolls this out and this game of whack-a-mole ensues.
So Facebook, the Facebook Pixel – which is the component, I think, that allowed Shopify to grow so precipitously and allow e-commerce to grow so precipitously – the Facebook pixel was a third-party cookie. That just means it was signed by Facebook, but you had embedded it in your own website. It wasn’t of the same domain as the website.
And what that did was that it just served as a transmission mechanism for conversion artifacts for data about these conversions to flow back. And so when that was blocked in Safari, Facebook said, “Just reimagine the pixel or reposition it as a first-party pixel. So, developers – here, you just have to sign it with your domain. It’s going to have the same sort of functional purpose, but it’s going to be signed as your domain.” So it’s first-party and not blocked by ITP.
So then what Apple did was a, “OK, we see what you did there.”
And so they limited the amount of time that a first-party cookie could send data back to the true owner for. So it just limited that timeline. And so my sense is, what Apple took away from this sort of back and forth exchange was that there is no way to fully prevent this behavior through policy. The way to fully prevent this behavior – which Apple calls tracking, which is kind of mixing the third-party and the first-party context, it’s commingling that data – is to just starve these companies of the data. There’s no way to sort of ask them to abide by this policy unless you remove the data from that sort of transactional frame where it will be used
MD:
Yeah. So, I guess following on that, Rishabh, why did most people not see this coming? You were in the ad tech space at the time. What was going on in the background there?
RJ:
Yeah, so I think that the way that Eric was describing people’s behavior to ITP is actually a pretty good indicator of what happened with ATT.
So essentially what happened was that people were basically saying, “Hey, we’re going to do all these steps to get around this action that you have taken.”
And so the same thing happened with ATT between July and December of 2020, so there was this progression basically in that six month time horizon where we went from people believing that actually, what was happening was the specific identifier called IDFA was being turned off. Then, a clarification was issued saying, “Hey, it’s any identifier,” then an additional clarification was issued that it’s actually a policy statement. And finally, Facebook said, “Hey, you know what, we’re actually going to abide by this” – because for a brief period, they said we’re not even going to abide by this. And then it was pretty clear that they needed to abide by it. And so in December of 2020, that’s when clarity sort of emerged.
However, the big thing that all of adtech recognized at that moment – and it was sort of a strange gentleman’s agreement that if we admit that there’s no solution to this for the next six months, our stock price is going to tank. That’s just true. And I wish there was another sort of collegial way of saying this, but this is actually what was true.
Criteo, the company that I worked at at that time, TradeDev – there are so many companies who have a dependence on this that basically everybody was issuing statements saying,”Hey, we have a solution, or we are working on a solution.” And so that led the agencies to saying, “Hey, don’t worry about it. Our tech partners have a solution.” And that led the brands to saying, “Hey, we’re not worried about it. Our agencies, Facebook, they’re all working on a solution.”
And when you get that comfort that, OK, the people who know what they’re doing are working on a solution, you basically assume that that’s true.
So what was happening is that in the early part of 2021, everybody assumed that, “Hey, we’re working on some magic solution.” Also, we didn’t actually know when implementation was going to happen. I should clarify that. So in January, we actually did not know when implementation was going to happen. And then Apple said, “OK, in 14.5, that’s when implementation is going to happen.”
And that was in the summer. But Apple does this thing that every update doesn’t necessarily get pushed. So 14.5 was not a push update. 14.6 was a push update. So there was another head fake where 14.5, again, people thought, “OK, it’s not as bad as we thought, because 14.5 has happened.”
But actually, most cell phones didn’t update. 14.6 was the push. And then finally, through the end of the year, then it happened. And so it caught people by surprise, because first there was industry storytelling. Then there was this notion of, “Hey, when is the actual impact going to happen? Is it 14.5, 14.6?” And then the last step is there’s a 90-day look-back window issue. So there’s a look-back issue on when the actual measurement is going to happen based on the identifier that’s coming in and the impact you’re targeting. And so that had yet another 90-day delay.
And so by the end of 2021, then it became like all of a sudden people were just like, “Wait a second. This thing is like absolutely demolishing our campaigns.”
The reason you can tell that people did not expect it is there was a precipitous drop in stock prices of publicly-traded companies. And you’re just like, “Wait a second – shouldn’t have this been known?” But, I guess you know, there was a series of events that actually prevented people from fully appreciating what impact was actually going to be felt over what was basically a year.
One of the most interesting things to watch happen over the course of 2021 is just how it played out in the industry.
MD:
Yeah, I mean, for those of us who were sitting on the marketing side of the table, it seems like the response of the ad networks was just to take control away from us. Why was that the case?
MO:
I can give what I think the reason is. So the risk of simplifying too much, what ATT and the things that led up to it really have done is it moved us from a deterministic world to a probabilistic one.
By deterministic, I mean that the post-back link that Eric described gives you as close to perfect mapping as you can get on the consumer internet, where you’re seeing kind of convergence data that are happening on the brand side, and you compare that with performance on the ad network side. And that gives you this great flywheel, and allowed Facebook (now Meta) to be kind of this engine.
Now, because that’s severed functionally, and you know, there’s been this kind of game of whack-a-mole leading up to this, we’ve moved to a probabilistic world where honestly, I think the ad networks have made the call that the only people who can really measure correctly or solve attribution or even attempt to are those with very, very large machine learning staffs who understand this space exceptionally well.
And so I think what you see on the product side is that the move, in-house, a lot of marketing activities. So like Advantage Plus, which is Metas rebranded automation suite, PMAX on the Google side, is an attempt, as you identified, Mike, to take control away from the market to a degree because the ad networks have the resources. And, let’s face it, the economic pain that they need to solve in order to keep their stock prices high. So they are highly motivated and with the resources to do it.
So at a very high level, I think that shift from the kind of world that we’re in, is leading to the product decisions that we see from the networks.
“What ATT and the things that led to it have done is moved us from a deterministic world to a probabilistic one.”
MD:
Yeah. All right. So Matt, you’re kind of bringing us to a topic that I wanted to spend time on. And we’re getting into like the impact section of the conversation. Topic being attribution here. And look, I’m cognizant: you get four marketers together. You could spend hours just geeking out on attribution, so I want to limit this part of the conversation a little bit. But it is an important topic to get into.
So, you know, as you hinted at like ATT moves, it means that the move to probabilistic attribution is going to become necessary. And in that world, you know, there’s much need for large ML and data. The big question here is do smaller merchants have these tools and understanding? They used to just be able to rely on the Facebook platform to be able to handle measurement as well as incrementality testing. What’s happening right now for smaller merchants?
MO:
One of the things that you can obviously see is that there’s a gap in the market that opened for smaller attribution players to emerge, so Triple Well and North Beam are probably the most obvious ones that have been pretty successful and I’m not sure would have been successful pre-ATT.
I think there is a begrudging acceptance now (at least amongst the brands that we speak to) that there’s no one source of truth anymore, and now the kind of big topic is triangulation and using like a multitude of different techniques to try and get to the answer.
RJ:
I think the real question is like, for small and medium sized merchants, what are they actually doing right now?
And actually, I think there’s two big problems that they’re facing. The first problem is that attribution has become much more difficult because there’s signal loss. So when Apple said, “Hey, you have to get a consumer to opt in,” by definition, a certain set of those consumers are opting out. And so you no longer have that signal. So you now need to like Matt has been saying, you need to move to probabilistic models.
But the bigger problem has become you cannot scale a given channel as quickly anymore. So it used to be the case that off of just Facebook alone, you could build a brand up to like a $30 million a year brand. Now, there are brands who are going multi-channel at like $5 million, right?
And this is actually a big, big change in the requirements of attribution and the complexity and sophistication of that attribution. So I talk to merchants all day who are like, “Man, if I can just keep my current ROAS (return on ad spend) and just scale it, I’ll be happy. But the problem is I’m hitting a wall on Facebook, if I scale any further than my ROAS goes down. And so I need to go multi-channel the moment you go multi-channel you need to increase your attribution capabilities.
And so then they can in turn use the platform as long as you have one channel,maybe two channels, but you simply can’t once you have multiple channels because then you’re competing for that attribution and so that’s the moment where it’s become much harder. And, candidly I don’t think that there are good tools for small and medium-sized businesses. That’s why they’re using multiple tools.
I don’t know a single merchant who does not triangulate, like Matt was saying, between their GA, their third-party attribution tool – whether that be Rockerbox, Northbeam, Triple Whale, whatever, you know, pick your, pick your third-party attribution tool at this point – and the in-platform data. And they’re trying to co-measure that with what’s actually written on their Shopify backend or whatever their backend is.
And so everybody’s trying to triangulate across these channels, and there’s also a need for better education at this point because at the risk of getting bashed on Twitter, there’s no shortage of opinions on attribution on Twitter either. And so it’s getting harder and harder for people to understand like, hey, what is the best way to triangulate how I actually do attribution for my brand? And there is actually still a gap on tooling and on education in my opinion.
“I don’t know a single merchant who does not triangulate.”
ES:
Yeah. So, I think what Facebook brought to bear for advertisers was that sort of a priori knowledge of any given person’s predilection to engage with certain types of content. They did personalization for ads, using third-party data to the same degree of efficiency as they do personalization for the in-product feed. And I mean, you can see how well Reels has done just as a result of that technology.
But that also, in certain senses, was problematic. So especially in the DTC landscape, what that enabled was this false belief that, okay, well this incredibly quick path to TAM, meant the TAM was very large.
So to your point, Rishabh, you could scale a DTC brand to like, maybe not $30 million, but like, I don’t know, a million in revenue or something in the first year. And that velocity of growth led you to believe, well, that chart continues to go up into the right for a really long time. So I’m going to raise money on that basis. And the reality was you had already started to saturate; that was your total time. It just explored that very, very quickly, right?
I wrote a piece about this in 2017 called “The Power and Peril of Facebook’s Advertising Platform.” And there were these stories that were starting to merge at that time, like these DTC brands that raised money on the back of very early success because they extrapolated that growth out to year two, year three, year four, when reality they had already exhausted the sort of entirety of that of that TAM.
But to your point about why this is challenging: so, you know if you’re using multiple data sets, I’m using on-platform data set which is just a sort of on-platform clicks basically, and all sort of ad engagement data, and then I’ve got some sort of attribution system that’s making available the purchase data, and then on top of that I’ve got probably some sort of probabilistic model, like an MMM or sort of incrementality measurement tool, that’s trying to tell me, well within the context of this historical spend, how much do we think how much conversion effect we think we can assign to each channel.
That is out of the question for all but the largest advertisers. Having that system? Then having the expertise in house to interpret and integrate? That is just not accessible to the vast majority of advertising clients (that previously didn’t need any of that). Because it was either all deterministic or they were so reliant on Facebook that they didn’t really need to explore any other channel.
MD:
Yeah, Matt. Go ahead.
MO:
So I know I was just going to say, I think Eric’s absolutely right that there’s now this sophistication mismatch that’s like brands doing two million bucks a year having to behave as if they were doing $100 million in times of the sophistication, but with exactly the same headcount and level of funding as before, which causes a lot of issues, as you can imagine.
MD:
So if you’re a smaller brand here, what do you do? And I think as you alluded to earlier, I think there’s some of its in-house capabilities which can be solvable by software, but there’s also a scale of data problem. And I go back to my time at Stitch Fix, we evolved over time as we started to spend more going from a couple million to $100 million-plus a year on ads, moving from a simplistic kind of last touch to some kind of version of multi-touch, which was probably false precision with eventually, incrementality being our North Star, and then we would calibrate last-touch models to incrementality multipliers – which was actually quite useful. But I recognize that it was a privilege of scale. What do you do right now if you’re a subscale brand? Even if you have the capabilities, you just don’t have the data right? What’s the optimal method for a sub-ten million dollar brand?
RJ:
Yeah, my candid point of view on this is if you’re sub-10, try not to diversify channels as the way you get some level of scale. Try to find the one or two channels, because the truth of the matter is that there are certain approaches that are only possible when you do have scale. Not only because of financial constraints, but also just because of data constraints, like you’re saying.
I actually do think that most of the successful brands who I have seen grow very quickly, they have found the one or two channels that they are able to scale and that they’re pretty disciplined about not exploring additional channels. They don’t, they don’t like take the wall that they are hitting for granted. They’re like, okay, I need to, I need to now ask questions around, Hey, how do I actually make the most out of this channel before I start expanding to other channels? And so my best advice, and what I have seen work with other small brands, is actually spend the time to really build out your understanding of the one or two channels that are working for you.
The thing that has become very popular recently as a consequence of this is retail. The channel that people have the highest level of, quote unquote, “certainty” around is just selling through retail. And so the way that this is exhibiting its behavior in practice is that more and more DTC brands, early on, are establishing a retail presence in addition to their DTC presence. And so that’s what I’m starting to see in terms of how you actually approach this problem and what you actually do in practice to get past the 10 to 15 million dollar mark.
“Most of the successful brands who I have seen grow very quickly have found one or two channels they’re able to scale, and then they’re disciplined about not exploring additional channels.”
MD:
For sure. I mean, I think as an investor now, I’m also seeing many brands go into wholesale much earlier in their journey than they would have otherwise, as you alluded to. And also solutions that are almost as easy as it was to spin up a Shopify brand, getting into wholesale can kind of be just as easy, Faire being a good example, right, of kind of one of those flavors.
We’re kind of moving into the third part of the conversation, which is on new solutions that are emerging from all this. And so just to frame it, it feels like in the Facebook heyday, there was an optimal strategy just to go open targeting, let the OCM, OCM algo do its work, just feed a high volume of great creative and great data pipelines, good measurement, and just let things happen. You actually needed very few heads to do things in a best in class way. This is starting to change, right?
And so to me, this signals opportunity, both to be a great kind of savvy marketer, but also if you’re building software, there’s kind of a new paradigm within here. Let’s talk about some of the new methods emerging and where software builders such as yourselves are poised to benefit.
Maybe let’s start with some of the stuff going on in the Shopify world. Matt, I’m calling on you because we’ve talked about this specifically a bunch. There’s clearly a huge effort on them building out their Audiences product now and helping merchants with traffic generation, which is historically a problem they’ve stayed out of. As they evolve to a model that looks, dare I say, marketplace-esque, what are some of the downstream effects and opportunities there?
MO:
So Mike, first off, I think that the solution that Shopify has come up with here is actually pretty elegant from a technical and strategic perspective, in terms of helping their users achieve their goals because obviously they’re highly incentivized as well to try and solve kind of exploding cat problems.
We are starting to hear that Shopify Audiences is beginning to show early promising results. And I know they’re betting on it pretty significantly. It’s come up a fair amount in the recent disclosures and calls with analysts that I know has been covered by Eric and amongst others.
But it kind of does, it strikes the balance of being, I think compliant with ATT, both in spirit and in practice. Maybe there’s a debate about the spirit that’s certainly in practice. And obviously it solves a pretty important problem, which is that Shopify really can’t be using sites of brands that run on Shopify as inventory.
That would not be acceptable to nearly any brand that I know. And what they’ve managed to do is to monetize in a pretty high margin way, for Shopify as a business, first party data they have access to. And I think we know that it’s considered to be a first party based on what Apple said, although it’s sometimes difficult to parse the tea leaves.
MD:
Yeah, so you’re hitting on maybe more of a meta topic on brands better leveraging their first party data. What are other novel approaches that any of you guys are seeing here? And is that something you think people should be even more strategic about than they are right now?
ES:
I mean, so what you’ve seen happen, and to Matt’s point, Shopify kind of exists, basically like a middleman. They allowed you to operate the storefront and then they could process the transactions and they had the transaction data and so that data’s there, in the same way that the iTunes data’s Apple’s, right? Apple doesn’t run the apps that you make purchases in, but it gets first party privileges to the data because they run iTunes. Well, Shopify can say the same thing.
So I think it’s the sort of compliance argument from that angle. I think it’s fine. But you know, it doesn’t Itself operate a storefront on which it can run ads, and to Matt’s point, I think it would be unacceptable to the retailers if it tried to do that on their websites, right? But a lot of companies do. They do operate the storefront –Amazon, Walmart. Walmart said it expects to get more profit from its ad business than from any of its business this year. And Walmart’s advertising revenue went from $2 billion in 2021 to $2.4 billion in 2022. That’s incredible growth.
What I think we’re seeing now is this explosion of what are known as retail media networks, because if I can’t go to the everything store for ads, which was Facebook – Facebook was the everything store for ads. You want to sell anything, you go to Facebook and put your ads there. Well, now you go to the stores for those specific things that are contextually relevant for the stuff you want to sell. Or where the transaction happens. If you can’t facilitate really targeted, direct response display ads, well then you go to the point of purchase and you just try to stick an ad in front of someone’s face right before they’re about to buy something, right? And then just hope they choose you. And, you know, you can make the arguments about whether that’s incremental or not. I think a lot of the people that have dove head-first in these networks, they probably are capable of doing incrementality measurement (and let’s hope that they are) and they’re actually getting incremental benefit from that.
But I think that’s behind the dynamic of what I call the “Everything is an ad network phenomenon.” Everything is becoming an ad network because well, Facebook’s not the everything store now. And hey, that’s an opportunity for me to spin up pretty meaningful super high-margin revenue.
“Now we’re seeing this explosion of retail media networks, because if you can’t go to the everything store for ads (which was Facebook), now you go to the stores for specific things that are contextually relevant for stuff you want to sell.”
MD:
Yeah, this is the topic I want to stay on.
And Eric, you’ve written about this, that is one of the more interesting areas that I think is going on around ad tech right now. And even like, if I date back to – you brought up Walmart as an example – but if I go back to 2017, it was amazing to me that Amazon could just flip the switch on a new business line and like in its first year, be it $2 billion, then $10 billion a second year now, like whatever was $38 billion last year.
And so, you know, there’s one thread of conversation, which is shit, you know, media buying on these plays is very antiquated, and yet they’re still able to get scale. What opportunity is there around that?
ES:
I mean, that’s a great question. I think any marketplace app platform that has a lot of audience data is in a position to capitalize on that now. And they already own the data. Spinning up the ad tech is actually not that complex. Ad tech people like to sort of claim credit for exploring this like incredible innovation, but the reality is it’s mostly commodity tech. The value is the data and how you extract insights from that.
But if you have a lot of first party audience data, you can utilize that right now. And you don’t necessarily even need to build your own ad tech. You could just monetize the audience data. There’s a lot of these so-called retail media networks that don’t actually have any impressions that they sell. They just partner with DSPs to make their audiences available for buyers that are similar, that are exploring customers in that space. I mean, they’re literally just handing data over to a DSP and allowing that to be purchased against.
So, you’ve seen this happen. I mean, Uber just in the last year talked about this new advertising initiative, Instacart has been expanding their ad platform for a really long time. You could look at companies like Ulta Beauty. I mean, they don’t have that many audience profiles, but it’s enough to where if you’re selling beauty products, you wanna activate those in an open programmatic environment. So the list is literally as expansive as every company with a decent size set of audience profiles.
“Any marketplace app platform that has a lot of audience data is in a position to capitalize on that now. And they already own the data.”
MD:
Well, one interesting thing you said is that the technology to actually launch your own kind of retail media network is actually not super complicated. And I’m curious for Matt and Rishabh, if you have a point of view on that as well.
MO:
I have just a quick story about one of our customers interacting with a retail ad network that kind of touches on both the current state of play and also the first party data.
So this company realized based on image data – who is responding to what creatives – that they probably had people who were really into golf, and they didn’t sell a golf product. It was an apparel company that didn’t sell anything involved in golf. So they went to Dick’s Sporting Goods and said,”Ok, we want to advertise against your ad network.”
At the time (this was a few months ago) I think the sophistication of the buying process was a little bit, you know, state dinner, Madison Avenue, 1970s, but it’s getting significantly more sophisticated because there are some kind of third-party things that you can spin up.
MD:
Yeah. All right.
I want to talk about your companies a bit. So Fermat is kind of indicative of a broader paradigm of distributing through content. Maybe to get into Fermat, Rishabh, take a step back and explain the evolution of influencer marketing.
RJ:
Yeah, for sure. So like I was sharing at the start, the reason to start the company was actually because of the problem statement that you can no longer track people from one website to another. And so the question that we started with was, “What does commerce need to look like if you can no longer track people?”
So let’s just pretend that no matter what you do, there’s no tracking. So basically, somehow just the evidence of a transaction has to somehow tell you something about where the person came from. The only way that you could accomplish that is that every store is one to one with every piece of content that you launch. That’s sort of the most extreme version of reality: if you launch an ad with this creative, it points you to this store. If you launch an ad with this other creative, it points you to another store, so on and so forth. And so there’s a unique full store for every piece of creative that you launched. And then that was basically where we started. We said, “Hey, why can’t we do this? Is it actually that hard to do?”
I mean, it turns out it’s not super hard. It’s work, but not super hard. So that’s actually what we ended up doing is we ended up making it possible for a brand to launch a unique store for every piece of content that they launch. And at the time, the thesis was, “Okay, the most engaging content right now that is actually attracting buyers to DTC brands is creator content, UGC content, this type of content.”
And so basically what we did was we said, hey, why don’t we create a mechanism to make it basically trivial to have a store that makes it feel native to that type of content and have the consumer feel like they’re engaging with that content all the way through to the transaction. And importantly on the backend, if the transaction happened, you know with certainty that it came from that particular ad and there’s no need to track in the first place.
So that was the basic idea and that’s sort of like where we’re going. And it turns out that when you do this, they’re tracking benefits, but there’s also conversion benefits because people actually like the fact that they’re in a shopping experience that’s native to the content.
But yeah, the entire reason for the company to be is because once we, especially with a third party cookie going away, once we get into next year, you’re going to have to have systems that are self-contained and closed loop.
That also, by the way, just to very quickly talk on the part on the previous topic, that’s also what makes me super optimistic about retail media networks is because it’s an entirely closed loop system. And so the question is how do you make more closed loop systems available on the open web? And we’re just going to be a participant in that.
MD:
I mean, one issue I’ve had with some assumptions around different, for lack of better term influencer marketing plays I’ve seen, is they neglect the fact that influencers are also subject to the same headwinds that other brands are facing on kind of hitting their full extent of their reach. So there was a world where, actually, organic distribution was a real thing there. Now it’s a very small silver of the distribution that one actually earns as a creator. So, how does that factor into your thinking and building?
RJ:
Yeah, totally. So the two main plays that we end up getting used for with our customers, one is basically UGC style content (or creator content or influencer content), and then the other one is branded content.
So within that influencer world, at this point, everybody basically does either what’s called branded content ads on Facebook, which you know, like the previous version of was whitelisting, and then it’s had many name changes ever since. They do branded content ads or they do Spark ads on TikTok, which is basically the same thing. It’s like you’re paying for reach. And the reason you have to do this is there’s no other way to get scale, basically. The number one problem, if we were to just sort of quote, unquote, rewind to 30 minutes ago, like a brand’s biggest problem right now is, hey, how do I get scale in my business? And if organic reach is also not possible anymore, because of changes in the algorithm and the ad networks needing to monetize more, then you have no choice but to scale your influencer content through these sort of like BCA or Spark ads.
MD:
Yeah. All right, so I want to ask Matt and Eric around this topic of creative and specifically AI and its role in all this stuff as well.
So Matt, you wrote a piece recently that underscores a lot of our core beliefs with Treat. It’s called “The Coming Generative AI Arms Race and E-Com.” So let’s go here for a few minutes.
I think most can agree that image models are going to give designers superpowers. There’s a bunch of new apps being marketed to marketers, given it’s such an obvious use case with a quantifiable ROI of all this stuff. What are your guys’ view on how marketers should sift through the massive influx of tools being marketed to them here?
MO:
My view generally is that with any new technology like generative-AI, you have to think about whether the opportunity is going to lie with the incumbents or with the start-ups. And kind of our thesis is that a lot of tooling around generative-AI is going to end up in incumbent design tools that are going to make the the production part of creative much easier. You saw Adobe’s Firefly suite was launched yesterday, Figma is going to have Stable Diffusion natively. So you’re probably not going to change designers workflows, they’re just going to be able to use that technology and the tools that they already use.
Where I think that generative AI is really going to be powerful is in the ideation phase, and this is borrowing from a framework that Eric wrote about that I think is completely on the money, which is that’s kind of where the bottleneck is in terms of getting high-performing creative, particularly if creative becomes one of the primary targeting levels. You can start to do a kind of targeting by doing rapid creative variant testing. And as the cost of content creation (both images and copy) kind of tends to zero, which is basically what we’re riding right now, the value of knowing what to create and why increases proportionally in the opposite direction. And it gets super, super important.
And so I think we’re about to go into this world of content abundance, and the tools that are most exciting really, and if I were a marketer I’d be evaluating, are the ones that help me understand like what creative concepts and visual elements to be pumping into these systems t just before the prompt level (for folks who have played around with these tools) and not in the kind of the pure play production process, because I think that will just end up in Photoshop and Figma and in the tools that the design and creative teams are already using.
“Our thesis is that a lot of tooling around generative-AI is going to end up in incumbent design tools that are going to make the the production part of creative much easier.”
ES:
Yeah, I think just to expand on that. Canva also announced a suite of AI supported tools.
I think broadly, my sense is that generative AI actually underscores and highlights the value of human ingenuity, human creativity. I wrote a piece about this recently called “Exoskeletons Not Cyborgs,” and I think the beauty of these tools is exactly to Matt’s point, is as content gets commodified and the value approaches zero, you see that the true source of value generation is in the creative input. It’s the thing that separates the human element from the machine element.
So my sense is that marketing teams adopt these kinds of tools most efficiently and effectively as efficiency unlocks for individual members of the team.
There are always a lot of risks with total marketing automation even prior to generative AI. First of all, you can’t just hand over the ad production process to Facebook or to Google. You can’t do that, right? Their incentives are different from yours. And you will see that they will very quickly iterate themselves into Ads that may be totally inconsistent with your brand guidelines, brand standards. You just can’t outsource that you have to own it. Or you have to use a third party tool that integrates with them.
But, the risk… I mean, I’ve just seen a lot of people immediately jump to this sort of logical endpoint of “Generative AI allows us to completely automate everything.” And I think that’s a mistake. I think these things still need to be somewhat siloed and integrated, but with human oversight.
My sense with ad creative, like you said Mike, you’re exactly right. That is probably the most obvious commercial use case for generative AI. And I would just be very, very careful about A) outsourcing that to any sort of third party that also incorporates that into the media buying process because their incentives are not aligned with yours. Now you can use a third party tool for generating creative. I think that makes a lot of sense. But I wouldn’t anchor that into the actual ad network itself.
And then the other thing is I think it’s just going to showcase exactly how valuable creative directors, heads of ad creative. How valuable they were in this process and how competitively advantageous a really good one is.
MO:
Yeah, I completely agree with that. Also, just to piggyback off that and maybe take it one step further is, again, this is gonna sound like such a love-in, but I’m just gonna quote Eric, one of his posts. The opportunity here really is to take the intuition out of the ad creative production process and allow everyone to sort of specialize in what they’re great at.
I’ll give you an example, we built a tool called Lookalike Creative that will basically take every single image that’s ever been run by a brand, including statics, video, including copy, and pull out all of the visual elements using a bunch of tagging and captioning techniques and stuff to identify patterns and correlate that with goals that you care about or demographics that you care about reaching. So it might be that images with beaches tend to perform very well with women 18 to 25 living in the Midwest, for example.
And one of the things that we found is that the model is particularly good at identifying something. It’s just very hard for a human eye to perceive, which was that we were working at a beauty brand. And there was one specific kind of marble bathroom. And it was a specific kind of marble, and it just happened to be really, really eye-catching for the machine that was highly correlated with CTR, which was the thing that they were trying to optimize for. Armed with that, the creative team can kind of do its magic and focus on the stuff that they’re good at. But the pattern matching across, I think this was across like 15,000 creatives. That’s just not something a human really should be either manually tagging or reviewing themselves. That should be outsourced.
So I view us as basically being a little bit at a crossroads where we’re going to allow the growth marketers and the design team to specialize in what they’re great at. And put on the plate of the machine, some of the tasks where machines are just clearly, clearly stronger, potentially layer on things you could think about layering on some kind of data corps as well, that would allow brands in a similar way that an agency has eyes on accounts. You could allow people to opt-in to share visual elements that are working for me with other brands to create some kind of shared intelligence. But the value, I think, is going to be in that specialization.
MD:
Yeah, I think most companies we certainly had in this meeting – it was a weekly creative growth sync where it was delivering insights from all the creatives, these five designers, were producing and saying, “Here’s the stuff that worked.. Here’s what didn’t.” People actually love tying their efforts to business metrics, which was oftentimes not the case beforehand.
I think the optimal amount of humans in the loop is not zero, but you could probably make that process a lot more efficient by what Treat and others are kind of working on here.
MO:
Yeah, absolutely. And I think, yeah, really understanding the why behind visual elements and why they’re load-bearing or not. And then also being able to rapidly create tests at scale that hit multiple different goals with multiple different demographic groups because time was a constraint for content creation. That’s not going to be as real a constraint anymore.
And the other point that Eric made, I think, I think is important, which is that I have heard some kind of marketers say, “Well, you know, well, surely this will just be done by Metta because they have like, you know, the best generative AI people on the planet and this is going to be rolled into the tooling that Metta has. I think that’s a critical error. Because as Eric mentioned, if you’re a DTC brand or an ecom brand, like your brand itself, as in your brand identity is one of your core pieces of IP. Not strictly IP, but one of your core pieces of value. And to outsource that would be pretty foolhardy or very brave, depending on how you thought about it, particularly given the incentive mismatch that’s been identified.
And you kind of see that strange creative race to the bottom that we saw in some mobile gaming creator strategies, like caused by that exact effect. And most of the brands that I know would not allow people to do that.
So there’s this interesting window where I think a lot of folks looking at the market – investors, operators, people who are kind of just generally interested – looked at the last wave of ad tech, and what tended to happen was like a sort of reverse whack-a-mole, or maybe it was kind of whack-a-mole, that a company would be launched, it would have an interesting lens on improving the media buying experience or some aspect of paid media. And what would happen is that the ad networks would incorporate that functionality. They’d build it out. And a lot of very well-funded venture-backed companies got destroyed that way. There’s a unique spot right now, I think, where there are some things that the ad networks can’t build, and wouldn’t be allowed to even if they could. And so it’s slightly – I hate to say it different this time – but it is slightly different this time for some of these tools because of the constraints around ATT and brand identity.
“The optimal amount of humans in the loop is not zero, but you could probably make that process a lot more efficient.”
MD:
Let’s maybe move to one last loaded marketing term and then we can ask a set of closing questions. But this is one personalization, right? So I think marketers have been talking about this for years. I ran performance marketing for a startup that was supposed to be a personalization company. I don’t think we nailed it. There’s a bunch of different reasons why now might be different.
I understand it’s a broad topic, but Eric, I’ll direct it to you to start. What do you feel like? How should people be thinking about personalization given there’s much new promise in this new world of being able to get more ability to get more precision on this. How should folks think about this topic?
ES:
So I approached this from the perspective that we’ve been talking about with respect to how Facebook did that with two different use cases. It was the ads use case and the content use case. So the ads use case, what broke that? Why can’t they personalize the ads in your feed with the same level of efficiency as they could before? Well, because the feedback loop was broken. They don’t get the data that drives that decision-making, was not theirs. It was generated by a third party and it was being sent to them in sort of like an unfettered way and that’s no longer the case. Well, why? And the reason that the personalization in the feed didn’t break is because that’s their data. It’s in a closed loop. What about the retail media networks? It’s a closed loop and that just gives you first party access to the data and you can use it as an input to the personalization.
Well, okay. So me, as a D to C marketer, an e-com marketer. I can no longer send my conversion data back to these platforms, they can’t use it to personalize ads in the feed or wherever the placement exists. But I can do the personalization on my website or in my app. I have all that data. That’s a closed loop to me.
And so what I should be thinking about is, okay, CAC has gone up and there’s no avoiding that. There’s no getting back to where we were. There has been a permanent dislocation to this market. We’re never going back. Targeting will never be the same. So once you accept that, then the CAC for a relevant user, call it a high-quality user, is higher. It’s systematically higher. Well, how do I, how do I attack that problem? I raise my LTV, right?
So I’m no longer personalizing the ads, or I’m no longer personalizing the experience of the ads layer and bringing the person into a universally consistent product experience. I need to have more generic ads, less relevant ads, and that’s out of my control. That’s a privacy thing. And then I need to sort of personalize the in product experience such that I’m driving up AOV, I’m driving up LTV, I’m driving up conversion rate, such that, you know, I’m matching that increased CAQ with increased LTV. And that’s where I think personalization comes into this new dynamic. It’s that I need to ingest that into the product experience, superserve the good users when I acquire them for more money, such that they spend more in my product. So, I’m gonna go ahead and start with the next one.
“There’s no getting back to where we were. There has been a permanent dislocation to this market. We’re never going back. Targeting will never be the same.”
MD:
Yeah, this is a great opportunity to plug two portfolio founders or aren’t here: Builder that does on-site personalization and then PostScript on downstream SMS marketing as well.
So yeah, I definitely believe there is a thesis that brands should be investing a lot more in downstream both conversion and retention.
So, you know, I think it’s obviously been a rich conversation on a bunch of different opportunities that are emerging from this chaos here. The headwinds are still very real for advertisers. But on the side as an investor of being excited about new platforms and business models that can emerge. Is now a time for folks to get more optimistic on ad tech as a category? I think a lot of investors over the years have kind of underestimated how massive some of the outcomes were. I referenced the trade desk, but for every one of those, you have dozens that were crushed by platform risk or something else existential.
How do you guys think about opportunity here as mainly through an investor lens, but also as entrepreneurs on the broader topic of ad tech?
RJ:
Yeah, so I can start. I would think that there’s basically a couple of very interesting opportunities. So the first one, which we haven’t spent a lot of time on, is what’s going to happen when the third party cookie goes away from Chrome. Because basically, ITP took it away from Safari in 2017. And we actually know we can monetarily ascribe the value of that loss. Because every ad tech platform gets $0.30 on the dollar for Safari versus Chrome. So you kind of have a pretty good long set of data to describe how impactful the loss of the cookie is going to be for Chrome. Basically, your ad effectiveness is going to drop 70%. That’s a lot.
And so I would say that the very first thing in terms of opportunities, especially with AI and machine learning models getting better is,” Hey, what are things that are totally cookie-independent in terms of serving relevant ads that you can possibly start to train on right now while the cookies still exist for the next nine months?”
So like, if it were me, I would just be like, “Hey, who’s out there getting as much signal as possible to train the best possible models, such that they’re strategically advantaged at the end of the year when this thing gets turned off?”
The second thing that I would be doing is I would be asking the question – we were talking earlier about like all these retail media networks that are popping up everywhere – it’s actually non-trivial for a brand to plug into one of these. So let’s just say I come to you, you have a retail media network mic and I’m like, “Hey, Mike, I want to plug into the Mike retail media network. And you’re like, “Great, it’ll take you two months and $50,000.” This is like the average conversation that a brand will have with a retail media network. And I’m like, “Man, that’s a lot.” And then I have to go to Eric’s retail media network. And this is extremely irritating.
And so the second opportunity I would think is somebody who makes it possible for these mid-sized brands to actually kind of very easily plug in to these retail media networks, which is very different than just plugging into a miracle and then having your product feed go to a retailer. It’s not the same thing to buy media. And so I would say that those are the two opportunities that I would be thinking about if I were an investor.
ES:
I think this is a new golden era of ad tech. I think we’re going to see many companies that build the tech to fill the gaps that have been left in the wake of ATT and that are going to continue to widen with future sort of privacy disruptions. It’s not just ATT, ATT. ITP is the tip of the spear. Apple’s been ahead on this. But anybody who was reading the tea leaves knew something like this was going to happen. And when it happens with regulation, it’s going to be more wide, far-reaching and probably more disruptive, right?
So I mean, I don’t know if you saw the TikTok hearing, but I mean, that to me is what’s going to presage a national privacy bill. And we have no idea. I mean, it might look like the one that’s on the table, it might be different. But anyway, everyone is going to have to adapt to this. And it’s just going to absolutely change the status quo. And I think many new ad tech companies will emerge to service advertiser needs. There’s been no demand destruction here. That’s the good news. The good news is people still wanna buy stuff, right? And so companies will emerge to facilitate that.
“I think there is an opportunity for mid-sized brands to very easily plug into these media networks.”
MO:
Yeah, I agree. I mean, I think people are just going to have to basically rebuild their entire ad tech stack from the ground up. Because if you think about it, Facebook was doing, we’re singling out Facebook, but I mean, for obvious reasons, but they were doing so many jobs. They were the everything store for ads. They also basically took on an agency role in part of their automation tools.
They really built out their tooling pretty successfully. So I think there’s a huge opportunity caused by this dislocation.
And also, I think that one of the things that if I were an investor and not an operator, I’d be looking at is, what are the second order effects of content being free?
So the thing about the internet, that was extraordinary and broke a lot of people’s brains at the time who were trying to apply like old school business models to what would happen on the internet is that distribution was free. And so all the case studies that you would have like other business school or whatever would be like, oh, it’s like a newspaper, but on the internet. Oh, it’s gonna be like classifieds, but like on the internet. Like we have a webpage now, everyone has a webpage. And then some weirdness really happens when distribution goes asymptotically to zero. And if the same thing is about to happen to to content broadly, images and copy. Some weirdness is about to happen and that’s a good place to make money… sometimes.
MD:
Yeah. All right. So my final question, and it’s really an extension of this, is, you know, we’ve, we’ve kind of been deep in the weeds today, some stuff grim, some exciting, some tactical.
For the closing question, and I’m leaving this intentionally vague. What’s one idea that you’re most optimistic about in the years ahead around all this stuff?
RJ:
Matt, you gotta go, man. I feel so inspired after the way you closed the last one. It’s like you gotta carry us into this one. Ha ha ha ha ha ha ha!
MO:
That was gonna be my closing thing.
RJ, ES, MD:
Ha ha ha ha ha ha
MO:
I didn’t know. I mean, the one thing I’ll just say is like, I’m obviously very into weeds in image generation, more so than copy generation. And I have consistently been amazed at the pace at which things move. There are things that were on our product roadmap where I sat down with our head of machine learning and said, “Yeah, we should think about getting to that, you know, in a few months time.” And then someone will do it on Reddit.
So I think we’re very close to a tactical instantiation of generative AI technology that’s gonna come very soon (much sooner than I thought). I think it’s short-form video – like text-to-short-form video is much, much closer than we thought. I thought it was a year away. I’d give it two months at this pace. The runways demo kind of blew my mind and I’m just dealing with the ramifications.
So I’m broadly exceptionally positive on a bunch of different tools for creation, and now we need to probably have better filters and better upstream tools to work out what to create.
RJ:
Yeah, I would say that I actually got asked a question on Monday. So we do like a meeting every Monday where we try to zoom out a little bit and ask questions about the business. And somebody on my team asked me, “Hey we take an inventory of where all in the business we’re using AI?” And I’m just going to try to keep this relevant for e-com operators, but I’ll just share a personal anecdote. I had never done this before and I’m glad that this teammate of mine asked us. And I was just like, shocked at all of the places where our workflows were already impacted. So our sales team was already using AI for call nodes generation, our marketing team was already using AI. Engineering team was already using AI. It was just everywhere in our workflows. And I could see the pace with which it was improving our operating velocity.
And so one of the things, if I were an e-com operator, that I would be asking is, “Yes, you should be thinking about it in the context of your ad creative like a specific workflow, but really something to really be thinking about is where are all the non-obvious places in my business workflow that it could impact the way that I do my work and like ask the question, hey, should I just be using some of these tools to improve my margins? Because basically, yes, look, CAC has been permanently impacted. And so the next set of questions in order to actually get strong EBITDA as you’re growing is how do you actually improve the overall efficiency of the organization? And it’s worth inspecting and asking, where are all the places that I can create leverage in the organization by using some of these tools? Like, I was shocked when I took inventory on Monday, and so I think it could surprise people as they do this.
ES:
Yeah, I would say that if you think about ad tech, there’s kind of three buckets of use cases: there’s targeting, there’s measurement, and then there’s ad serving. The ad serving, I talked about that, it’s pretty trivial. Targeting, permanently degraded, not much you can do about it. Measurement, where I’m optimistic, is that there will be technologies, tools, the methodologies that are probabilistic that emerge that reach the same level of performance, of efficiency, of precision as we had in the deterministic environment.
I’m hopeful that that will happen. I’m optimistic that that will happen. And that solves a lot of problems. That solves a lot of problems that were masked in the previous environment where everyone thought this was totally deterministic. It wasn’t. And so I think if we actually invest more into those methods and those tools and those technologies, we’ll actually be superior in some ways to then what we had before. Again, targeting? Can’t do anything about it. But the measurement side, I do think I’m very optimistic about the things that can be achieved with that.
MD:
Thank you guys so much. It’s been a great conversation. We appreciate it.
