Podcast: Everything is an Ad [Techonomics]
Jake: This week we have Andrew Yates joining us. He’s the co-founder of Promoted.ai and previously was at Pinterest and Facebook. Andrew, welcome to the show.
Andrew: Hi. Thanks for having me.
Jake: Something we do with every guest as they come on is we ask them, how did you get to where you are now? So can you take us through that journey?
Andrew: Yeah. Hmm. It all started when I left graduate school many years ago.
Jake: I think this is the best intro we’ve had so far. So keep going, keep going.
Arun: Yeah.
Andrew: It all started when I married my wife in graduate school. In fact, actually, that’s a real story.
Jake: Yeah, Let’s do it. Now we know.
Andrew: Yeah, no, really.
Arun: It actually started when your mom met your dad if you think about it.
Originally Deciding to Do Ad Tech 10 Years Ago
Andrew: If you think about it. Yeah. And then fast forward so many years, and then my wife got a job at Pinterest. She won some hackathon. I was doing my PhD. I decided I’d prefer to be married instead of doing a PhD in a different state. And so I left my program, and it was time to choose what I’m going to do with my career. And I thought, Hmm. What do people do here in San Francisco? It was ads. That’s what they do in the San Francisco Bay area.
Google is an ads company, Facebook is an ads company. So I knew I wanted to do a startup and I thought that the biggest industry to be a part of in technology is ads. And so I was going to become an expert in that and start a company in ads. But at that time, it felt like I should probably acquire some experience in ads that would be useful so I wouldn’t just be yet another nobody wash-out kind of PhD student doing yet another ad tech startup, especially at that time, that was the fashion. If you didn’t know what you were going to do, you started an Ad-y sort of whatever startup and hope for the best.
Arun: And just to set the scene for other people. What year span are we talking about?
Andrew: This was the early 2010s.
Jake: Outing the age of the guests.
Andrew: Yeah, This was ‘12, ‘13. I joined a startup called Red Hot Labs and they were doing ad tech stuff. Right. And it was like optimizing on top of Facebook ads, and I saw this was like, geez, this is never going to work not because this is not a good startup, it’s a great startup, but because Facebook is Facebook.
And if you wanna learn how to optimize Facebook ads, you should go work at Facebook. So I did, and I joined Facebook and I joined the ads ranking team. I worked on the click models and conversion models and all like the “Practical Lessons of Predicting Clicks at Facebook” system. And then I joined the marketing science team at Facebook and did more of a research-oriented role around how to use information that Facebook had to better predict conversions and incrementality and how to model that, how to use data more effectively.
And then I wanted to get more into the engineering production side, as opposed to simply the research ML side. And I joined Pinterest, eventually was leading the ads marketplace engineering team.
Jake: Hmm.
Andrew: And then I started Promoted and that’s how I got here. There are a few other steps along the way.
Jake: We can dig into those. Yeah.
Arun: What’s interesting here is what was it like working on like a foundational team at Facebook, because, I feel like what you were doing there is like working on the search team at Google. It’s the thing that funds all things. That’s what basically the ads team is at Facebook.
Andrew: Yes. Oh, it sucked.
Arun: Yeah. That’s also what I hear about searching Google by the way.
Jake: The answers come out.
Andrew: Yeah, look okay. So I’m gonna skip back a little bit. I’ll get back into that because people are like, tell us more if it sucks. It was great. There you go.
Arun: Yeah, yeah, yeah.
Andrew: Going back to the startup thing, I was at Pinterest doing in-house stuff for a while and hey, it’s a cushy life, but are you always exactly proudest of everything that you’ve built? Could be a lot better, and I think Pinterest felt that way too. Although from organizational perspectives, maybe there are reasons for why it’s not better. This is how Promoted got started: sometimes it’s not just talent and opportunity. There are also a lot of organizational reasons for why systems are built the way that they’re built.
No Good Ad Tech Startups Anymore
Anyways, I got back into the startup dream. I realized that it was time to think about it. It was a woman who worked at Pinterest in business development, who took me out to lunch, and said that there are no good startups in ad tech. And do you know of any, and there weren’t any good startups in ad tech. She said all the good ones had been bought by Google and Facebook.
And by the way, the one that I’d been working at several years ago, that was bought by Google. And so I realized, “Hey, look, there’s a market opportunity here for doing this type of technology really really well.” And this ties back to what it was like working at Facebook, by the way, which is there are fewer and fewer people who’ve actually started the system.
Working on a Foundation Team in Facebook Ads Engineering
And I didn’t start at it as well, by the way, when I joined Facebook and this is why I thought it kind of sucked, was what they really needed was to throw a bunch of bodies at refreshing models. And that’s the worst job. I don’t know why people think machine learning is so fun. I guess they think it’s smart or more interesting than other types of software, but a lot of it is a lot of grinding metrics and that’s how I described it at the time: “The Ads Score Mines.”
Facebook did a really fantastic job of designing these incentives so that individual teams can just focus on whatever metric they’re doing and grind on it.
And the gist is that if you can imagine, like in Middle Earth, the elves had come and they had designed this great system and Facebook did a really fantastic job of designing these incentives so that individual teams can just focus on whatever metric they’re doing and grind on it. And it doesn’t matter what their other teams are doing.
They had a very strong top-down leadership. People need to do their work and not focus on what other people are doing and not argue about it. Just make those metrics grind up and do whatever you’re doing. The bad news is the act of doing that wasn’t a whole lot of fun because it’s just like refreshing models.
They’d have another team go build out the next version of the system. The team that I worked on was more of just like grinding out metrics. We would be increasing total Facebook value by 1% every couple of weeks, but the way we were doing it was just basically adding more data into models, retraining them, grinding, and that kind of stuff.
It wasn’t really a whole lot of fun, but it was really eye-opening and it was really interesting to see what it feels like when you’ve created a working machine and what it feels like to keep that 1%, 1%, 1%, 1%, 1%. And thinking about how Facebook has grown since 2012 through 2016 or so, which is the time I was there in 2017, 18. Yeah. I forget. I’ll think about that later. That’s how they built it.
But anyways, that was my experience on the Facebook engineering side, and I met some really, really talented engineers. They’re great. A lot of them went to Cruise in fact, now, actually, once they’ve built the thing, then operating it is a little bit of a different situation.
A Passion for Designing a Better Ads System
And I was really passionate about designing the system and making it better as opposed to operating it. And that’s why at Facebook, I joined the marketing science team and I was able to do novel research around, basically, different ways of sending data to Facebook to make your ads campaign work better.
And this is, by the way, a preview of how Promoted is designed in our thinking of it, which is when you’re doing targeting on Facebook effectively, you’re telling Facebook some useful amount of information, but it’s a binary thing. Or like you give them some sort of label and they train something on that label, but you could just give Facebook the information you used to generate those targeting audiences
And they could just directly model it in combination with what they have and get a better result. like every single insertion, as opposed to like at the audience level. And so that was the sort of work that I was working on at Facebook. So that was really interesting. And then, yeah, at Pinterest, I felt like I was getting a little too far away from productionization and engineering because ultimately I did want to know how to build this myself.
And so I joined Pinterest and took some of my experiences at Facebook and helped build some of the Pinterest systems.
Arun: You know, It’s funny. In my work in self-driving, I think a lot of times people come up to me and they think it’s all the same. It’s like this incredible machine learning problem. It’s like amazing. And it is when you first solve it, because that’s when you’re trying to solve this really really interesting problem.
And it’s okay, like we got it. So that it’ll recognize stoplights now. And it’ll recognize stoplights under these conditions and everything. And then you find out that you have to make it like 30% better. And then just grinding out that 30%, especially when you’re getting to the last few percentage points, it’s just backbreaking and not fun work.
Andrew: But it’s very valuable.
Arun: Totally, totally, totally. It’s valuable. It’s good, it’s a good job, all that stuff, but it is like mentally, it’s a grind. It’s really tough.
Jake: It’s going from that, like zero to one to the one to 100 and 100, 2000 type of like data-driven, like mass scale company is like just a completely different ballgame is in terms of an engineer, right? It’s like the way that you even frame whatever your objectives on a quarter-by-quarter basis is not rooted necessarily in metrics until you get to that point where you are just grinding the machine down to increase revenue on a compounding like points of a percentage basis on a quarterly basis.
Designing Incentives is Hard and Takes Leadership
Andrew: One thing I think we are taking for granted here is that the system exists in such a way that you can grind and reliably get these smaller improvements that build on each other. And this is another inspiration of Promoted, which originally was called “Algorithmic Auctions” because the idea of it is you need to set up an incentive system so that everyone is behaving.
In their own local interest, their own greedy local interest that creates the whole system to be better, and that makes the whole system to be better. And that doesn’t just happen. We take it for granted that it just happens, but no, it actually takes a tremendous amount of leadership in design, which especially for these economic systems or ad systems, many companies see the grinding part, they see the metrics part, but they don’t really understand or appreciate the leadership and design part.
And so they end up with a sort of working system, but mostly a lot of them end up with a mediocre system in a bunch of internal teams that keep fighting with each other forever. And every launch review is this huge collection of different metrics. And it’s a horse trading show forever. You don’t end up with a Facebook. You end up with… I love Pinterest, but they’re not a trillion dollars.
Jake: Yeah. Maybe that was a good time to dive into what Promoted does, kind of the evolution of it from that initial name to what it is now, and then maybe we can back into how that got created.
Optimizing All Commercial Media As Ads
Andrew: Yeah. We optimize marketplaces. And the way we do that is we’re using performance ad tech and delivering everything that’s commercial media, like an ad. So that means we’re doing unified optimization for marketplaces, for search and feed and ads and promotions. And our ultimate play here is to network together the properties we are optimizing and creating an ad network, a fully native first-party ad — -everywhere.
Jake: And so let’s ground that in an example, so I’m like a shop owner or a marketplace owner, how would I use Promoted?
Andrew: Yeah. Let’s use an existing customer. How about Hipcamp? If you’ve used Hipcamp, by the way, it’s really cool.
Jake: I’m gonna take a trip in a week actually using Hipcamp. So I’m excited about it.
Andrew: Oh, nice.
Arun: What is Hipcamp?
Jake: Airbnb for,
Andrew: I don’t know how to describe it, but “Airbnb for camping.” Yeah.
Jake: Sorry, Andrew. I know, I didn’t wanna do it either. I was like, it’s on the tip of the tongue.
Andrew: I know I was trying not to do it, but that brings it home. But anyway, Airbnb for camping. And what do you see when you open up Hipcamp? We control all of that in the sense that we are running the decision algorithm for what shows up in search and feed everywhere on Hipcamp. And by the way, if you open Hipcamp, there are no ads.
If you open Hipcamp, there are no ads.
They don’t even do merchandising. Really. They don’t have promotions, but you can still use the same ideas. What’s going to convert? What’s a good match? And then all the measurement side. So a lot of what we are doing today is on the back end and on the measurement and data side, where we added all of the rigor around ads, measurements that you would have on, let’s say Facebook newsfeed or Pinterest ads.
And then we do that for everything. And now you have a measurement of everything and you do things like: it’s showing this listing in a certain feed. Is it incremental? Meaning if I show you this, it’s gonna cause you to do something. That kind of thinking comes from the ads world.
Promoted Started as Only Ads For Marketplaces
Promoted actually started in just ads. What we found out was, wow, people are doing regular e-commerce searches and there are a zillion competitors out there. And most people are building in-house. Our number one competitor is building in-house. Generally, they’re just looking for whether or not this looks like a “good” item versus “ads thinking:” is this an incrementally valuable item.
So what we’ve added to Hipcamp was a whole bunch of logging around their own data about measuring everything if it were an ad and then to hold every single item accountable to deliver value back to Hipcamp and model that and then use that for deciding what to show in search and feed.
Jake: So what are the signals that play into that then? How do you get the feedback? Cause and an ad writes a click or impression or something like that. Is it just pretty much the same type of info signal back into the model to help reinforce the promotion?
How Promoted Handles Data
Andrew: Yes, we’re doing real-time streaming. We actually do our own eventing. That’s another thing that we learned: early on, we’re gonna be the “backend for your backend.” I think we actually literally had that quote somewhere on the website, and we found out really fast that no, because whoever you are and however you’re logging data, it’s garbage. It’s trash, terrible. Stop it.
Like you gotta know mobile, like there’s no click on mobile. All right. There are no click events. No, if you ever do mobile engineering, it does not click capture. No, that doesn’t exist. There are like navigates and you have different views that come in and out of visibility.
You can’t just send huge fat requests on a mobile device.
There’s no click and then same like impression, there’s no impression event that happens. No: it’s visibility and you get a stream of them and then you need to batch it and you need to send it over through a wireless network without getting throttled. And you can’t just send huge fat requests on a mobile device.
Web Thinking On Mobile Fails
And I think a lot of lazy sort of web programming where people do these sort of easy mode, drag and drop JavaScript sort of integrations, which is also becoming more challenging because people aren’t excited about that kind of load or privacy issues. But none of that works on mobile at all. And some people don’t find that out until they need it to work and you cannot go retroactively back and fix it if it’s not ever logged correctly in the first place.
Jake: Yeah, the data’s gone.
Andrew: Anyway, mobile eventing. It’s not there. You can’t fix it. It’s not logged directly.
Past Engagement is the Best Future Predictor
So we do from the mobile side: eventing and web too, and then we have our whole streaming data infrastructure where we’re joining all of these events together and aggregating them. And then that’s coming back to the models and the models mostly are dominated by what people are clicking on and engaging with.
Duh, that’s your objective, so probably the best signal for that is what did people do in the past. Mostly, not all entirely, but that’s a good signal. We also ingest whatever other signals you have. We are not providing any third party data. We don’t need it. And that’s the other thing we’re learning from Facebook is that people think: I have this really great data, blah, blah, blah. That data sucks. As soon as we get on the first-party data, it’s just garbage in comparison to what we can collect ourselves for so many different reasons. And if it is valuable, that was my whole research point. It’s finding the money. Prove it, send it.
Arun: The first thing is the, what I heard from all of that was stop logging. And then I guess the next question I would have is: how do you enforce rules on logging? Do you give people sort of… How does Promoted make that better? Do you give them an API or an SDK, or how does that work?
Make Money not Metrics
Andrew: We have open-source SDKs. We’ll integrate it for you too. And then we provide metrics at no additional cost. Our theme is that we have a value called “Make Money, Not Metrics,” and the reason we have such a value is because
Arun: — -I love that — -
Andrew: you can keep building reporting forever. If you’re not accountable for actually doing something with a report, it’s just this might be useful to someone else to do some other thing that you’re not actually responsible for accomplishing you can build and log and create reports…
Jake: Coolest visualizations of all time.
Andrew: The coolest visualization of all. And it’s wow, that’s cool. And then it’s: “did you ever use this to make any definitive judgment on anything in the business ever?” And it’s: I felt like I did, but no, you didn’t. And that’s how we think of logging: we want it to be correct because we need it to be able to deliver the incremental value on the search and feed and ads optimization side.
We want people to have fantastic eventing and measurement infrastructure. Not because that’s what we sell, but because you can’t do anything else interesting until you’ve correctly measured what you’re trying to accomplish.
So we just provide all of that metrics, infrastructure at no additional cost. We cover it by charging a minimum rate per month, covering the infrastructure piece. But the idea of it is that we really want people to have fantastic eventing and measurement infrastructure. Not because that’s what we sell, but because you can’t do anything else interesting until you’ve correctly measured what you’re trying to accomplish.
So if you don’t even have that, everything else is garbage. So please just get that out of the way as fast as possible. And we’re very serious about it. We’re like: Hey look, we’ll do whatever eventing. We’ll do all the consulting. We just gotta get the data in place because everything else derives from that, let’s get that in place first. And that’s how we provide it.
Jake: Makes sense. Arun, you’re also dealing with logs these days with model prime. what do you see as some of the takeaways or maybe some of the differences from what you’re doing over at model prime?
Arun: So for model prime, we deal with robotics logs. And I see a lot of the same challenges, like people not knowing how to log their data properly. People using older techniques or no technique at all to make sure that they have observability. And when you start dealing with high volume and whatever of data, it’s not just a question of you having to log properly in order to find interesting things not just to find things, but to find interesting things.
And that’s really what it seems like you have to isolate out right? And so do we. And like I see a lot of parallels and I see a lot of parallels with a lot of times, it’s not the core competency in org. If you’re trying to build Hipcamp as an example, your core competency is building this experience for your users. It’s probably not on the logging side and probably not on the ad side. Right?
Ads and Accountability to Prove Value
Jake: Or relevance models, right? Let’s be clear, it’s not just about ads, right? It’s about leveraging those or this is what I’m giving Andrew leveraging those signals in an ad type way to like inform your relevance models of, okay, if I’m looking at my feed is like this thing, giving me the outcome that I desire. And if it’s not then like, how do I rank that lower or higher?
Arun: Totally. Sorry. We’ve talked about ads so much. I’m getting everything back to it now, when in fact you’re really
Jake: I think my favorite part of this. Yeah,
Andrew: No, that’s, that’s our philosophy. Everything is an ad. Everything is an ad, because “ads” has such a bad connotation. People are really hesitant to say, “I work in ads.” Yeah. Cuz they get it. There are all kinds of scummy behavior around answering: “is this valuable or not?” and it’s frustrating.
But nevertheless, these ads are the biggest businesses in technology. So I think this created this big blind spot of a lot of really interesting ideas around how to think about ads: Commercial media — - that’s accountable for producing some kind of value. And fundamentally, we think of that, ads, as the fundamental thing that we’re doing: commercial media.
An ad is a way to layer in this optimization of how much would you pay to be delivered in a competitive way. Whereas removing the ad auction piece of it is simply more of a flat take rate or the business makes a fraction of the sale. But it’s still fundamentally the same kind of thinking in our philosophy.
Bigger is Better for Promoted
Jake: Hipcamp is a company that is likely on the smaller side in terms of its engineering and product work. I’m not sure I’m just making an assessment based on where they sit in the market. Does Promoted have a general segment that it feels like it operates strongly in? Compared to maybe some of the behemoths or maybe some of the startups that have just been created.
Andrew: oh, The bigger, the better Hipcamp is one of our smaller customers.
Jake: Okay.
Andrew: and the reason for that is not so much that we can’t help smaller customers, but more that bigger customers are much more valuable. And what we do is a lot. It’s typically you go out and you go to SoftBank: “Hey SoftBank, we need a hundred million dollars to go build an engineering team, seriously.”
And okay, we’re gonna open in the Seattle office. We’re gonna open a Mountain View office. And from being there doing it, we know: “Nah, just gimme 10 great people and we’ll provide it at a 10th of the cost.” And actually, it’s: — I’ll go into a little bit of go-to-market challenges here — — But we like working with top marketplaces for a couple of reasons.
One is this where we can shine. This is where we can show they’ve already done the low-hanging fruit stuff. If it’s something that your in-house team can do in six months, and it’s really straightforward, what do you need us for? right?
Jake: Yeah.
Promoted Targets Top Marketplaces Because Those Are The Hardest to Optimize
Andrew: Go do it yourself. Or there’s all, there are bazillion other vendors out there that are like, you know, entry mode for entry people.
Jake: Or whatever. Yeah.
Andrew: Yeah. And by the way, ML in a black box is so easy. Anything that’s so easy again, it’s so easy, you can do it yourself. And if you are a technology company and we go after main technology companies, because that helps us build an amazing product, because now we’re not solving really basic problems for people who are just clueless and don’t prioritize it.
We’re solving the core problems of top technology companies. We go after Andreessen Horowitz’s Top 100 Marketplaces in general, two-sided marketplaces with non-fungible inventory and ephemeral inventory, like the problems that are so hard that there doesn’t exist any kind of good substitute you can throw in some sort of Google Recommend or AWS Personalize.
They just don’t work very well for these types of marketplaces for a variety of different reasons. I mean, They work okay. And we expect that people have already set up their own Elastic instance or they are running Algolia or something like that. Like they’ve already done best practices, but this goes back to the Facebook story from the beginning of this interview:
Compounding Growth in Marketplace Search Improvements
1%, 1%, 1%, 1%… it matters. It’s not just the “1%,” it’s that you’re fundamentally improving the thing that people are doing, why they’re there. And it, over time, becomes this uncanny experience that somehow magically, you got the thing you were looking for and you train users to expect that, and you become like this default where people just go define whatever it is that was on their mind without even thinking about it too carefully.
It’s magical, but it takes that sort of discipline to just incrementally keep pounding it over and over and over again. So anyways, we can build that kind of system when there’s the volume. And there’s also a foundation of that they have the basics in place. So they’re not just relying on us for doing the basics.
The Best Properties
And the other part of it is our ultimate vision is to build a commercial media empire and a commercial media empire needs to include the absolute best properties. No one wants to be a part of a fourth-tier ad network with anyone coming and going. That’s garbage that exists all over the place.
And you know what those ads are, they’re bad. They’re not good. You don’t want them. But our goal is to seed a network, so your network is seeded by the absolute best, most unique properties of inventory. That’s nowhere else. That’s really engaging that you can actually buy it.
Bootstrapping a DR Ad Network from Marketplaces
That’s actually another thing, for DR (Direct Response Advertisement): How are you gonna actually transact? And for marketplaces, they own all of the fulfillment, right? So it’s not just like drop shippers or someone who’s trying to arbitrage the system. So that is the other part of it why we’re going after the two-sided marketplaces is because it allows us to aggregate this really fantastic unique set of demand for attention, and then build out our network that way.
But like, how do you start an ad network? How do you build a network without a network? How do you sell a network without a network? That’s not a very attractive proposition, right? Hey, join my ad network. Who else is in it? Nobody. So how about you have a network, your internal marketplace is a network. Let’s make that work. And then when we have enough of them and they’re good and you want to be a part of it like other companies you respect, then we want to take all of those together and then start distributing them amongst each other and also like include some other properties like social media and publishers.
Jake: How does that work from like a sharing perspective from a marketplace to a marketplace, like eventually, how do you envision that happening? I imagine that internal network yes. That data’s important and that data’s important and potentially proprietary to that like a specific set of users for Hipcamp, for instance, it’s a bunch of people who want to go camping, it’s that’s great.
No Data Sharing
Another company might want to have that as well. But like in terms of data ownership, who owns that data? How are you building that network? How are you gonna then try to build this connection between two companies who might otherwise be competitors?
Andrew: One is that we don’t want anyone feeling like they’re helping their competitors and it’s not that useful. I know that typical ad tech is let’s get a huge ball of data aggregated across everywhere. That’s Google and Facebook’s model. And other examples, like Amazon, and Microsoft, if our plan is let’s just take the core competency of Facebook and outcompete them. That’s a bad plan. that’s it? Don’t you think? No, our plan is.. yeah, go ahead.
Arun: It’s funny because you say that because when we’re pitching sort of like, our data in for a platform and a lot of times people would ask us: Hey, are you going to replace snowflake or AWS? No, no. I’m not gonna fight battles that I can’t win. Okay?
Andrew: Even if even if you pick off a couple of early victories, how big can this be? Because it’s like they expect… Yeah. And also the, if you haven’t noticed the world is becoming more and more hostile to that kind of business plan
Jake: no, you don’t say
Andrew: Have you noticed? You have.
Arun: For a while, you didn’t need a business plan. It was just like now I think the fact that they’re hostile to any kind of business plan is a huge step up from where we were in 2021, in terms of VC funding.
First-Party Data and Buying Experience
Andrew: Yeah, exactly. Anyways, Hipcamp’s data is most relevant on Hipcamp. It’s more of you want the real-time aspect. Our perspective is you want first-party data. It’s most important to have true first-party data that you can trust and relevant to where the users are. So the idea is can we take, for example, like GoPuff is a customer of ours and Reddit is not, but we’d like them to be, so the gist of it is the dream of it would be, can we take a listing of GoPuff and put it directly onto Reddit? Literally, let’s say we’re optimizing both Reddit and GoPuff. I know Reddit’s the way they want their system to be is: How about you load all of our inventory? It is like ad formats and the ad format we provide and then you click on it and then it takes you to the other website and then the website tells you to download the app.
And the app says, okay, now you can find with it. And every single one of those steps is just like conversion rate, like halves, if not more, what if you could just literally call your search back into the retrieval system and it just fetches items from someone that’s not you? And then you can just buy it directly there, wherever you’re at.
And so like you have to have this sort of integration. So the way going back to your original question of how we think of this integration is that you don’t need to have some sort of data sharing pool at all. What you need is this interface, the standardized interface of here’s all of the available inventory.
And here’s how you display it and here are all of the metrics associated with it. And when it’s relevant to make a connection, namely, when someone makes a conversion transaction, you can make that connection back, which is, of course, you have that. If you make the purchase, you actually transact, then we know who you are, cause you bought the thing.
So there’s not an identity problem anymore. But at the same time, in terms of who gets to see this, we have the first-party data of how it is this being delivered already. We don’t have as much like we’re solving targeting in the sense of like real-time signals about how people are engaging with a little bit of exploration and past experiences of other things from this marketplace.
And then the other side of it, I know everyone thinks about it like targeting, but we think of it more of like experience, which is having the old world of trying to retarget people and then having them click through and leave to go someplace else. And then you capture those users and they are your users forever to this idea if you don’t wanna rent your users from some other platform.
Focus on Increasing Orders with Seemless Integrations
Jake: The metric you use is what customer acquisition cost? LTV or whatever,
Andrew: yeah, I think this is a big mistake that may have made more sense years ago, but I think people have over pivoted onto this sort of thinking way too much. I think the real problem people have is how do I get more people to buy my thing? If you have a valuable service, including the fulfillment part, you’re not gonna get disintermediated.
You may have a more fundamental or a fundamentally different problem. If you have no valuable service that could just be sold anywhere, but for like marketplaces in particular where they have the fulfillment and the integrity owning this inventory, it doesn’t really exist outside of their ecosystem.
Your problem is you just need to have more orders. You need to have more buyers and more sellers, and do it faster. So our philosophy on this is to try and make it as seamless of an experience as possible. And do that in a way where literally you just take our back ends, which everything’s wrapped, right?
So the ability to be able to do the metrics and the optimization that we do and the ad system, like we do local ads, native ads. Now you have a consistent wrapper across all the different marketplaces. And then the vision is if I’m another, let’s say Outchool and Teachers Pay Teachers.
Those are two of our customers. Let’s say I’m a teacher on Outschool, and our school makes a call to Promoted for what Outschool things I’m gonna show, and we mix in a few Teachers Pay Teachers items to advertise to the Outschool teachers. This is extremely relevant, and it’s a seamless sort of transaction.
It’s not: Hey, here’s a banner ad about this other marketplace. I think that’s more of like a domain match. If it’s more like social media then, which is like more, Hey, you just have a huge surplus of attention. And now you’re just trying to capture some of it.
Yeah. There are some targeting issues and some like niche issues, but those exist anyways. No matter how these platforms are trying to advertise, it’s more of how do I close that gap as much as possible to someone that’s interested to, they can actually get the thing they’re looking for. And we see this as trying to make it as native as possible because a lot of the drop-off comes in the experience of how people interact with ads, not so much the targeting, and the optimization, although that’s also important.
Jake: Yeah. So it’s almost like the funnel, the piece of the funnel that you’re most focused on is making sure that once someone is there looking at the things that they feel like those things are relevant to them enough to like basically log a conversion. And if they don’t have that ability, then you, the model isn’t working correctly, if they do then the model is working correctly.
Do I have that right? And if I do then, how does Promoted then work with customers to identify how well they’re performing and benchmarking against potentially other competitors that you might run up against?
Andrew: Yeah this is getting back down to earth because we don’t have a network. We don’t have a network today. We’re not selling a network. We don’t have a partnership with Reddit.
Jake: You had me sold so hard that I got to ask you that question. Now you have to bring me back down to earth.
Andrew: I know I drag down to earth. Right now we don’t optimize a network of networks. We optimize a single network. So we optimize your network, and your marketplace. Yep. And the same technology we would use for doing the network with the networks, but come back down to earth. It’s A/B test, by the way, our competitors are always building in-house. At least at the type of companies that we’re selling to they will have data scientists and it can get quite testy,
Jake: I was gonna say, I bet you that’s fun. Those conversations.
Andrew: Yeah. It’s interesting.
Jake: conversations on a fundamental data level.
Arun: Yeah.
Andrew: It gets to be, we have to win the trust of these customers, and we have to earn it and they will make us pay. Like if you are lying or faking it, this sort of fake it till you make it startup strategy. I feel like what people don’t realize is that this sort of advice: How to start a startup. Everyone knows them, at least in this ecosystem for like Andreessen Horowitz’s top 100 marketplaces. So if you’re faking it, they will find out and they will hate you. They wanna find you out, cuz they wanna know you’re smarter. I do this too. If you know what engineers really love to do? Find out fakers and be like, you’re faking it.
I can just do this for free. Here’s the trick. I can do it myself. So the way we test it is you bring in all of that sort of energy of we’re gonna find you out and you actually do a great job, right? If we’re gonna explain everything to you, how we’re doing, you can set up your own experimentation.
However, you want to do it. We have experimentation, but we actually continue deprioritizing, although fancy dashboards and stuff because internal engineering teams are like, no. no, you are not telling us what the experiment results are. We’re telling you what your experiment results are. And we gotta yeah, okay, go for it.
Right. And you win their trust. Cuz it works. And that’s what we keep and that’s what we focus on. It’s like that 1%, 1%, 1% sort of mentality of it pounded into you at boot camp at Facebook, we do that for our customers and every day it’s like whatever we produce. Yeah. Okay. That was yesterday.
What do we do today? What are we doing today? Make sure it works. Make sure it works right now. Make sure it works better tomorrow and prove that it works and they’ll run their own experiments. And eventually, you went over the engineering team because it’s useful because it works.
Arun: It’s this thing like you can do, you can build your team. You can do almost anything in-house, but doing it and doing it well are two entirely different things. And I always think a lot of these things that companies have had to do in the past like you had to build your own ad tech team. You had to build your own whatever team. As industries mature, and as the technology matures, you can buy more of those solutions either as a technical component or as a service. You know, then you don’t have to be an expert. And I think it actually accelerates. It’s one of the reasons why certain markets accelerate faster.
It’s because you have componentization as an option, you just don’t have to build it yourself. It’s one of the things that actually slows the company down a lot. So they have to build everything themselves to basically make everything work.
Andrew: I’d love to talk about this a little bit, because what you’re saying sounds very reasonable. It’s Hey, we’re gonna build Stripe of ads or something like this. The big problem with this is that what a lot of companies actually want to build is equity in themselves. I think what a lot of companies really want is to go public last year.
That’s their dream, operating themselves. That wasn’t exactly the goal. It’s more like, how do we become a very valuable company that’s valued as a technology company? How do we achieve this, how do we structure our business so that it looks like in a spreadsheet that it has a hundred X multiple on revenue with no consideration for operational costs?
That means you do wanna try and build it in-house, or at least you definitely wanna look that way. If you go and raise hundreds of millions of dollars from venture capitalists, your goal is to spend that, to build some sort of asset, or at least you better look like that.
Jake: something that you can yep. Something.
Andrew: so it’s, it’s been an interesting go-to-market and it’s been very interesting the last couple of years in particular, because the cycle has gone pretty quickly.
It went from, we started the beginning of COVID just from like early prototypes. So it went from like the apocalypse like you won’t even have toilet paper and the stock market crash to tech bubble to not so much a tech bubble. And that was fast. Right? And so there was like this idea of okay, build an engineering team.
If anything, I seem to be working with spending capital as fast as possible to build as much capture as much as you can. The problem from an engineering perspective is that doesn’t really incentivize engineers to build great systems that work well and frequently it doesn’t which makes sense.
Some companies pull it off, and others don’t. So I think now in this market environment, people are realizing that we’re actually accountable for running this business for the long term. We don’t just sell it to someone else. And somehow these magical engineers are just gonna like magic it into existence because look how expensive they are.
They actually have to work together and make it work and cooperate and like each other in some way, or at least be cordial and cooperative and deliver. And that dynamic I think, is just now starting to be realized. So anyways, I wanna point out this idea. Commoditizing or like making different parts as a service.
It’s not always what people actually want to buy. They don’t always want to deliver value in their core business today, as efficiently as possible. They may be trying to build something else, like their own careers or equity or what appears to be equity in their own company.
Jake: That’s interesting. So basically you’re saying that the incentives are just misaligned, so like their incentives are maybe on a more personal basis or maybe on a more like, Company oriented basis where in reality, like the thing that would solve their problem is this commoditization of a part of their business that they could just buy to help deliver more value faster.
Andrew: It’s like misaligning to what? Yeah, exactly. Misalign to what, right? Yeah. It’s like if you’re thinking, oh, the business asset that exists today should be effectively run and eventually profitable. That seems like
Jake: You’re on your way to your..
Andrew: that, is that a coin?
Jake: house. Yeah. Yeah.
Andrew: It’s okay then how big is your business? And like how much, how valuable is it if you have to mark to market at where you’re at today, is that good? Maybe it’s not, maybe it’s oh no, actually the only way that this valuation makes any sense is if this becomes the next Amazon or the next Uber, and if it doesn’t even look like it’s gonna become that, then there’s not even a hope that this is ever gonna be achievable.
So that’s very rational. And by the way the same thing for like in-house engineering teams. It’s if your job is as you’re running your career in many senses. So it’s if you believe in the company mission and you want the company to be very successful, you’re trying to behave so that the company is as successful as possible.
Yeah. Your job is okay. How do I most effectively deliver value for our customers and sellers and buyers, if you’re told by leadership that no, no, no. Your job is to make a credible shot to become the next Amazon and do it in six months. That not really a by sort of argument.
Jake: So this sounds like the GTM challenges that you’re facing
Andrew: It’s just very interesting. Yeah. It’s just that I think it’s unique to where we are. It’s unique to the companies that we’re doing. Ultimately, the way we get around it is you can’t win over everyone, but also just do a really great job, like actually have a fantastic product and that’s also why we have nine people on our team and we were last valued it a lot.
I don’t even wanna say it, we did a great job of fundraising from Y Combinator continuity. They’re great, by the way. We just don’t wanna get into like whole valuation kind of
Jake: Yeah, no worries. We can chill outside of that.
Andrew: Yeah, but the gist of it is why can we deliver so much value? And the part of it is because we just focus on building a really great product and not all of the other incentives that if you’re working at a bigger company, you build this in-house, you have these other incentives like you gotta look like you’re building a next-generation platform, like a technology company, which turns out to then suck a lot of the oxygen out of the room in terms of focusing on making the product work really really well.
Jake: That’s an angle ahead and considered for the sort of SASS model and piecing these components together and commoditizing those and then selling them. I think that’s an interesting GTM challenge that likely all of them are going to have, if they’re not, to your point, like the Stripe of X, the AWS of X model.
So that’s interesting. How, based on that, what are you gonna do next? What are the next steps for Promoted to continue to build out the product to build that value that you’ve been talking about and to build the great product that you want to build?
Andrew: Yeah. On, in one sense, it’s pretty simple. We have the customers on, on our website, teachers, big teachers, and Outschool, Hipcamp, Snackpass, and Gopuff. Other customers are in the works and somewhere not public, but every week, everyday we have a slack channel and that 1%, 1%, 1%, 1% just making these great customers even better.
And then working with the in-house engineering team and the internal product management and whatever problem that they have to fix it right away. Now that’s my top priority. I wanna be in every single one of these. Solutions engineering sort of calls. Like it’s not the B team that comes in integration and makes it. No, it’s Dan Hill and Andrew Yates.
It’s like CTO and CEO, founder. Every week, I wanna know exactly what things are now and then make it really really great. Cuz I feel like even though we have, it works, like the product works, it’s delivering value for customers at the same time, we know it can be so much better and we believe to win as a great company.
We have to not just make a good product, but a phenomenally a great product, amazing product. So that means organizing how we do our team and how we work with our customers a little bit differently than I think some of the other go-to markets like for as soon as you get like over a millionaire ARR oh yeah, scale the sales team. No, we didn’t do that. The second piece is going and getting more of those logos, honestly like working with a few of these customers, like keep doubling down, keep finding the next best customer and making it an even more of an amazing experience for them. And then these customers talk to each other, right?
You can, it’s a small world. They call ’em up and Hey, how was it? And they’re either gonna say, eh, yeah. Or they’re gonna say, oh yeah, they solved all my problems are amazing. You should work with them. And we need the latter. So those are the next two steps for us to just keep making our current customers even happier. Whatever that means.
And next is to keep getting those big logos because ultimately you don’t know how amazing you are. We are like, I could tell you, Hey, we increase your sales by 10%. There are a bazillion startups out there that will promise you something very similar. Yeah. What they don’t do so much is here are these respectable top tech companies that use Promoted.
And you don’t have to fully understand everything that we do, but you can understand that these are respectable top tech companies and you can call them up right now. And it’s not just like the VP of marketing, whatever, but the engineering team that’s responsible for dealing with this every day. And they’re like, yeah, they’re great. So that’s our top goal.
Arun: It’s goes back to that whole, like doing it well and doing it well means doing it at the highest level and doing it at the highest level, like generally means having those big logos up on your website. And I think that’s one of the ways that you prove that you can do it better than they can in some ways.
Andrew: Yeah. I think doing it better or outsourcing it, I feel like we really want people to feel like they could do it. Like we wanna reveal all of them, like we don’t have secrets, in the sense like, oh, we have some sort of secret sauce. It’s more like we want the engineers who are working on these systems to feel that they are much more efficient and effective as engineers versus they’ve outsourced something that they felt like maybe they didn’t have the expertise to do for whatever reason. I think that’s another aspect here of having these logos here and like how we wanna work with our customers is we don’t really want to fall into this bin of build versus buy in the sense of okay, we’re outsourcing this.
It’s more of this is a way for your existing technology team and your existing engineering team to advance their own careers, in fact, and of course make the company more successful, but here’s a bunch of tools and systems so that they can focus on what they’re focusing on and we provide them the tools so that they can do that even better.
Arun: So just to ask a question that’s really been foundational in the ad tech world. I think over the last, call it six months. But these IDFA changes have gone on. Yeah. You knew it was coming. You knew when you signed up, somebody’s gonna ask you about IDFA. It’s just if you come on here and you work in self-driving, you have to answer a Tesla question. It’s just the rules. I don’t make them.
So how did that impact your work at Promoted? Did you catch a tailwind from it? Did it cause problems? I’d love to hear about that.
Andrew: Yeah. Short answer is it has zero effect on us. It has actually, the longer answer is it’s the motivation of why we exist. But the shorter answer is it has zero effect on us because
Jake: that was part of
Andrew: You knew it. You knew it. So there’s a catch. There’s a catch, right?
Jake: asterisk
Andrew: We are all first party data. We’re your marketplace. We’re your e-commerce site or social media. There is no tracking problem on your own app. There’s no identity problem on your own app. You have all of that data. So the way we get around this is we’re not aggregating data in a big ball and competing with Facebook. That’s not how we work. We are your own data and we’re helping you use your own data more effectively.
And then to the point of IDFA and, and generally performance advertising not working anymore. One is, I really haven’t seen people change their behavior very much yet. There are no other breakout products in this world.
Jake: I was gonna say who’s people, like people as in consumers? Or like companies that are being built around?
Andrew: Companies. We’re the number two problem, actually, number two problem is unity efficiency. Number one problem is customer acquisition, top of funnel and doing it cost effectively, no one has a solution for number one, we have a solution for number two for unity efficiency, but top-of-funnel efficiency or paid growth, no one has a solution and it continues getting worse. Everyone knows it’s getting worse. There’s no hope in the future for it getting better. As far as anyone can see, as far as direct response type of digital advertising, that typically is going to say Facebook or Google, or there’s a variety of up and coming platforms that always seems to be in the kind of stuck road mode of up and coming.
We think we can be there. We think we can build a better system through this decent, fully distributed, decentralized way of not sharing data, but doing a really fantastic deep integration with these as individual properties.
Jake: Do you think the difference between top-of-funnel and we kind of talked about that before, too, like the leveling, like where in the funnel you sit and focus on unit economics and optimizing that problem.
Jake: The tailwind question that Arun asks, I actually am curious have you found that customers are like, oh crap, like the top of the funnel is still an issue, but now that we know that we need to optimize more of our funnel in other areas, perhaps we need to actually have unit economics be a lot more efficient as people get here because of the lack of people that don’t get here, we need to make sure the people that do get here actually convert. Has that played into any of the conversations or is it really just like separate?
Andrew: You think they would relate to each other because they’re multiplicative
Jake: Right, exactly.
Andrew: User retention. Yeah, exactly. But these are people and big organizations and some of them are moving very quickly, especially if they puffed up a lot in the last couple of years.
Jake: Yeah. Okay.
Andrew: So people don’t actually think that way, actually, these tend to be like totally separate apart. Now we think that way we’re like unity
Jake: That’s the first thing I thought about.
Andrew: You think that way, but yeah, we’ve seen like more of like desperation kind of thinking like you’ve seen this in the wall street journal every day, someone’s building a new ad platform. They weren’t building an ad platform when things looked good, they were building an ad platform when things were not looking good.
Oh no, we need a new source of revenue. What are we gonna do? But anyways, we think of it this way because of the unit efficiency. Okay. So lile keywords or like key ideas, unit efficiency on your own marketplace is somebody else’s customer, customer acquisition cost. Right.
If you can solve unit efficiency and you can do it in general then. Yeah. Well, you’ve created this matching algorithm and that’s your customer acquisition cost. This is a sort of thinking that is just so natural at, let’s say Facebook, but it may take a little bit of thinking on. You have to think about it for a little bit, but if I’m delivering media commercial media on someone else’s website or someone else’s app from that app’s perspective, it’s unity efficiency from your perspective, it’s customer acquisition cost.
Arun: Hmm.
Jake: yep.
Andrew: So you have to solve unity efficiency. If you want to have any hope whatsoever of solving the customer acquisition cost in this sort of DR Response world. And that’s how we think of it. So that’s why the IDFA is a thing, this is who’s going to eventually crack the next way that people can do performance advertising.
At scale, we know this is coming, this is continuing to come. There’s no break, sort of solution up there. There are some people who are trying to pick at it. There are a lot of these sorts of Shopify direct to consumer aggregators, like cross promote. Somehow others are all gonna cross-promote each other.
Jake: they’re basically building that network. Right?
Andrew: yeah, they’re trying to build it from like the network side. But they didn’t focus on it from this first principle side of you first need to have the unit deficiency piece and for the unit efficiency piece and you have the metrics piece, we’re solving it from the other direction.
Jake: Cool. Makes sense.
Arun: We have a tradition here on economics.
Andrew: Yes.
Arun: and the tradition is called the hot take, which since Jake has become a father, I have rebranded to Papa Jake’s hot takes, and Jake has a hot take prepared for you, Jake, would you like to give Andrew his hot take for this episode?
Jake: I would love to. We’ve been talking a lot about ad tech, networking, making this deviation from performance advertising, and there’s been some moves in the industry lately. And I think one that I found really fascinating and I wanted to get your opinion on is, which is, obviously the mobile gaming platform. They love for developers to create games. They bought iron source, which was an ad tech company from a mobile gaming perspective.
And then now AppLovin is like, no, actually we want unity, but we want it without iron source. And so there seems to be this sort of like shuffle between these like developer networks and developer platforms versus like ad tech and like ad tech, trying to buy into these developer platforms.
And I wanted to get your thoughts on first of all, why is that happening? And do you have any hot takes on this particular situation?
Andrew: Yes, first. I don’t know. I wanna be super clear. You should not take my opinions with any sort of credence. I’m gonna say dumb things, especially if people are in this industry.
Jake: That’s why this is the hot take. Don’t worry. Don’t worry.
Andrew: But then I’m gonna come up with, oh, here are all my actual opinions. right. I’m full of shit and don’t listen to anything I say.
And, and part of it is because, Hey, look, I’m a dad of two small kids and I’m running a startup and I’m also the technical founder and running enterprise sales. I’m a busy person. So one way I find time in my day to do that is not reading a lot of industry news. Although I know that sort of drifted across my hacker news feed.
Jake: See, you did know you did read the headline.
Andrew: I know. I did see it. Yeah. And also we’re not really delivering much in the mobile gaming system. I know there’s a lot of ad tech in mobile gaming. It’s a huge business. We just, for whatever reason, We landed more in the marketplace space and that’s ed tech versus gaming. Although one day maybe we will maybe.
Jake: You build your empire. Yes.
Andrew: We build our empire. I will have a gaming section to it. But going back to this idea of gaming platforms and ad tech, one of the biggest problems for ad tech, in general, is control. If you don’t have control over, what’s shown you get two very bad things that happen.
One is you get squeezed out by whoever does have control. They’re just gonna keep saying okay, we won’t show until you pay us more, you get squeezed out. And that, and ultimately the control is whoever displays the final ad, like on some sense, it’s the publisher, but it could be just the platform idea.
So this is partially why a lot of ad tech has been. As a business, as a venture funded business has been so unpopular for the last several years, is that you get squeezed on both sides. You get squeezed by the people who are buying the advertising, and you get squeezed by the platform. And the style has been like the Facebook model, the Google model, where they just own it end to end and you get squeezed out.
So that’s one issue. So no one wants to lose control. Control is so important. The front is from like a margin’s perspective, like who has negotiating power. The other part though, is the experience and this gets back into what we are talking a little bit about. Could you just take a marketplace listing and place it somewhere else and you just buy it?
Right. Whereas it is like I take a Hipcamp listing with all of its viability and formatting and just place it on a different app. And that would be a great experience. And that’s kind of least this is kind of the opposite of it, which, in addition to losing control, you end up with all of these small frictions about how the format has to be and where it can show up.
And what’s allowed to be shared and not shared in terms of data integrations. And you just make worse experience, both from buying and selling these things from like operational standpoint, but also from the user experience, it makes it less efficient and it just doesn’t work. Like eventually becomes less efficient enough that it’s not profitable to do it anymore because it’s just a bad experience.
It’s too expensive, too hard to use. That’s my opinion, that’s what I think is both from the sort of margins, Hey, I don’t want to be in a position where I’m gonna get negotiated out and lose control, but also like Apple sort of control. Like this is the business control. Then there’s like the, the user’s end user experience sort of Apple level kind of control.
To make commercial media work. It needs to be a good experience, both from buying it and consuming it as a user on whatever that platform it is. And so if you don’t have that control, you end up with like lousy experience. That’s inefficient, and if it’s inefficient enough, then you can’t even do it at all because it’s not valuable.
That’s my opinion.
Jake: Yeah, I hink that’s, that’s a great opinion. I think the other piece of this that I’ve been thinking a lot about in particular with this deal is just that you want to be where the developers are in order to have them, integrate quickly with some sort of ad network that you control to your point.
How do you get more of those ads everywhere across games? It’s if you have the entry point specifically, when they’re building the game, then like they’re gonna use your platform to show the ads. And therefore you have more slots to show ads in. And I think that’s this additional piece of this which makes a lot of sense.
And I think because of the reason of IDFA and others, there is this sort of mad dash for this like sense of control that you mentioned, but like at the developer level.
Andrew: Interesting. Yeah, I don’t know about the developer level piece because I just don’t know, actually, I’m just gonna stop talking on that part. I do. I get the network piece. That piece makes a lot of sense. I do think that there’s a lot of propaganda in terms of oh, we care about the developers, but
Jake: Oh, I don’t think they care about the developers. I think they just want them to integrate the ad network quickly. Like for what it’s worth.
Andrew: Yeah. Yeah.
Jake: Sorry for any developers that are offended by that, but yes.
Andrew: Yes. And also there’s a network effect too. It’s no one wants to participate in the second greatest network. And then the third greatest, forget it. Right. It’s like a winner takes all, and that, that’s also true for marketplaces as well, by the way, that’s also going back to Promoted.
That’s also our idea, which is if we can’t sell to the best marketplaces, we can only go after like second or thirds here,
Jake: Yeah.
Andrew: How big can we be as possibly be? So that’s why we make sure that we can deliver for the best marketplaces.
Jake: Cool.
Andrew: But I think it’s like that aspect is they want to be number one because number one is worth exponentially more or multiples more than.
Arun: Andrew, thanks for coming on and educating us on the ins and outs of ad tech and what you’re doing over there, at Promoted as well. Super interesting. I feel I say this with, I think a lot of guests, but I really think we’ve only scratched the surface of probably all the things, and all
Jake: Yeah, I feel like we need a coffee or a beer.
Arun: Yeah, yeah. So yeah, so let’s make some time to hang out, but also at some point, you have to promise us to come back on and talk about this more, especially as Promoted grows.
Andrew: I’d love to thank you.
Arun: Yeah.
Jake: Thanks, Andrew.
