Josh Chandley: Felix The Chief AI officer of Unity. you launched Vector a year and a half ago.
Felix The: Mm-hmm.
Josh Chandley: What did you believe about the product at the announcement that you no longer believe now?
Felix The: Well, let's just say before I touch upon the product, probably the more important bit is when you're launching something new in which you believe it's transformative, but you don't have any empirical evidence outside of your own lab of what that thing says, right? Not yet launched to the customer. The only thing that will carry you forward is the conviction you have based on your expertise, research and evidence and otherwise that. This is the moment that you do not lose steam despite whatever you see out there. so if there's one thing that I guess I have now that I didn't have back then is the empirical evidence about the fact that the customers actually love it. So yeah.
Josh Chandley: Well, yeah, well, an upside surprise is a very positive thing to be wrong about. and so I've I've been watching Vector's progress as a customer, you know, both on this demand and the supply side. we've been active users. What was the hardest thing to get working that you didn't expect? Like, cause what you're describing is totally true. Like when you hit the road, that's the moment when every little
Felix The: Mm-hmm. Mm-hmm.
Josh Chandley: mistake becomes much more obvious. It's also the moment you can improve the fastest. So what was the hardest thing to get working?
Felix The: it's a zero to one moment because it's it's very hard for most companies to basically even have the ambition to disrupt what they already have. Especially when you say like what you already have is like not bad. It's just that for the ambition that you believe and what the industry really should have, you could do better. it's kind of like you have a plane but you have to make a new one too at the same time when the plane is still flying. That transitionary moment when you need to disrupt yourself and you need to beat yourself was hard. because like think about it, like when Factor was still baby in the lab, it doesn't get traffic even in our own environment, right? Like our legacy model basically got the majority of the traffic. That switchover, when you basically starve the baby, but you don't want to starve the baby, but you have to run the plane, that was a very artful orchestra that we need to basically meet a trade-off. And in some cases, when you're in that moment Being courageous is the most difficult thing. Which is like, I know, it doesn't look like at this current state that it is ready for prime time, but we gotta give it some air debris because it's going to be born. That was hard, but we did it.
Josh Chandley: It's so interesting to me because you are talking to me like every GM that launches a game, where,
Felix The: Mm-hmm.
Josh Chandley: yeah, my team launches a new game every like many years. This is always a soft skill. This is always challenging. and there's always this period when you launch it, it it needs to get volume, but it's gonna be a little rough at first. And we have to kind of accept that going into it.
Felix The: Yeah. Yes, and and and that's like say launching a game. there are two cases. Like if you're launching a game in which your displacement costs or your opportunity cost is like
Josh Chandley: Mm.
Felix The: zero zero, that's like a risk you take. But in my case, I gotta take air from the legacy. Yeah,
Josh Chandley: You were killing the old game. Yeah.
Felix The: correct, correct. Yeah. So that was hard. And then when you basically take air of the legacy and you know it's not exactly at a performance state yet because you need to give it air, you're gonna You're gonna be okay with the fact that you are you're taking some risk. and we did it and I was very pleased with the outcome.
Josh Chandley: there's lot of pressure in that moment. There's a lot going on externally too. I I would say that the early skepticism was particularly ungentle. You know,
Felix The: Mm-hmm.
Josh Chandley: Unity had been promising the engine data advantage for years and never shipped it. The read was, you know, vector was more more of the same, almost like a distraction. Was any of
Felix The: Yeah. Yeah.
Josh Chandley: that fair?
Felix The: I would say there's no way for me to be in a position to characterize fair or unfair. The market spoke. The market has, you know, wisdom in the crowd about what they believe in based on performances of history of companies of the past. it is what it is. All we could say is like, all I could say is you don't let fairness or unfairness of a comment to distract you about what you would like to do. And we were focused on the mission and we delivered.
Josh Chandley: Why didn't the runtime or developer data ship in that previous era of Unity pre vector?
Felix The: I I think Unity has gone through many different stages. Many different stages. the company itself is, you know, it's it's it's decade long old. There's usually a time in a company's stage of maturity in which a certain type of product is something that the company is able to deliver and for the community to accept and for us to be able to actually form it in a way that we basically Provided with our customers in mind. In the earlier stage of Unity, Unity has always been focused on like, you know, if you want to call it as we still do today, a game development engine. Yes? A game development engine, and when you break it down into its core of primitive, the most core primitive component, you gotta be able to use it to make something. So the idea of like data, when you're focused on that principal core, is like, is that Just a luxury item that we need to think about now. But as games become more sophisticated, as the industry becomes more competitive, the idea of data is something that we believe the developers should own because they could actually control their own destiny, what they would like to do with that information. You want to use it to manage your stuff, your games, your live ops, your whatever, in many different ways. Data gives you the insights to do so. You wanna basic a partner with certain vendors in which that data can actually amplify the activities you're trying to do, you can do so. So that's the nature of how we ended up with developer data framework. We believe this is the moment where Unity can cement its role. It is not just a classical game engine where you can make stuff with it, but you can actually operate your titles with it with confidence too.
Josh Chandley: That framework, as it applies to vector, is it rolled out and bidding live and auctions today at scale? Or is there data flowing and it's just not being quite used yet?
Felix The: The developer data framework runtime is
Josh Chandley: Mm.
Felix The: being used. We're very pleased with the results that we have seen today. we have more room to grow on adoption, and that adoption is just a number of things that we naturally will see as people upgrade and they basically release new games. And also the way we basically present these options for developer to opt in. we display it prominently in our dashboard, So we can we can partner together on what you would like us to do with it or you would like to do it yourself. so it's a it's something that is already used, but we are very interested and excited about what other surface areas we can actually display this opt-in that will accelerate its adoption.
Josh Chandley: I see. So opt-in is still a key area. Well, let's look back to Q2. Last quarter, vector's growth rate accelerated to 23% sequentially. You passed
Felix The: Mm-hmm.
Josh Chandley: a billion dollar run rate. If you had to name the single biggest driver, what was it?
Felix The: It's hard for me to name a single biggest driver, not because I don't want to name it. There is one if I just basically stack rank like the list of things we did in a descending order, there is one. But in the aggregate whole, just to answer your question directly, it's a combination of like model architectural improvements, a combination of like, you know, how do we basically do feature engineering? It's hard to pinpoint to one because we never basically have that singular aha moment. It's a continuous improvements of the model over time. If you twist my arm and really have to pick one that is
Josh Chandley: I will.
Felix The: probably in a recent memory, I think you won't be surprised. Day twenty-eight has been quite successful. and day 28 is quite successful because you know, you you operate video games yourself. day seven sometimes often is just not sufficiently long for you to actually understand the maturity of your cohort, especially when your game mechanics and that's the interaction between your users and your game. You don't really get that clear signal on day seven about who's going to basically behave in a certain way, you know what I mean? So day twenty eight is is such a successful product for us, so that in recent memory, one thing I would volunteer.
Josh Chandley: Yeah, something we've been looking forward to for a long time. I mean, a a lot of IAP monetized games, for instance, say that their median, their median whales first purchase doesn't occur until week two. So so having the depth beyond day seven is incredibly valuable. And that's I'm assuming that's why over a quarter of your customers adopted day twenty eight in Q2. What kind of advertiser moved first?
Felix The: You won't be surprised. there's usually like people market typically reacts in a rational manner. the people that move first are developers that represent the genre in which like longer retention players are key to them. in which like the monetization events are not clear that you can get that clarity on the early days. RPG, strategy games, like but there's also puzzle and casual. So It's not limited to one genre. It really depends on your game. If your game values that monetization unlock moment on the latter days, people tend to move quicker. The second part also, day twenty-eight, especially when you're basically launching a new product, it does carry more degree of risk, right? When you basically launch a new product, the earlier adopter is are the people that basically use the product first. So what we find, the developers that move are the people that actually are comfortable with the idea of day twenty-eight. They tend to be very sophisticated. They also appreciate the fact that day twenty eight is the prime maturity cohort they care most about and they also have larger budget. that's
Josh Chandley: I see.
Felix The: that's the that's the type of people that adopt it.
Josh Chandley: Okay. And then I'm assuming the the adoption is lagging for essentially the opposite. customers that have essentially faster turnover, they're less total yield focused, more IRR focused, just turning their money over quickly.
Felix The: Yeah. Correct. Yeah.
Josh Chandley: I see. Why was moving from day seven to day twenty eight, ROAS optimization, such a such a step change?
Felix The: well one is day seven is day seven. Day twenty-eight is day twenty-eight, day whatever, and day seven. Because like you can't assume a successful day twenty-eight if you're completely blind of the prediction mode in day seven. i in a way like the reason why day twenty eight is really key for us is not only because like it is Pretty much a natural explanation about how the market wants to see cohort maturity because day seven doesn't guarantee the outcome of day twenty eight. But the model quality itself is a far harder model quality to achieve greatness at. Because you need to be able to predict day seven really well if you want to have any attempt in predicting day twenty-eight. so that's why I think it's such a wonderful step function because it forces us to actually up our game in the sophistication of our prediction. And given the fact that we are getting really comfortable that we're getting momentum on factor day seven, that gives us the ambition and and the appetite to tackle day twenty eight the right.
Josh Chandley: So I the way I visuali visualize your problem, which may be insultingly simplistic to you. is this, you know, if you build a a rocket to take satellites to space, that
Felix The: Mm-hmm.
Josh Chandley: doesn't mean you can get to the moon by making a four times larger rocket. You know, this harder, longer mission creates new failure points. I'm assuming
Felix The: Mm-hmm.
Josh Chandley: that was the same kind of challenge you face with D-28. So what got harder to do at day twenty eight versus day seven?
Felix The: like for example, if you predict what I'm gonna do in the next thirty minutes, the next three hours, the next three days, the next three weeks. There's three in common there. The next thirty minutes, unless you feel like, you know, you don't enjoy this conversation so much, you're gonna kick me out early, chances are we're still gonna be talking.
Josh Chandley: Ha ha.
Felix The: so it's an easy prediction. It's an easy
Josh Chandley: Yeah.
Felix The: pr prediction. But three hours? Yeah,
Josh Chandley: I'm glad you're that confident, but go on.
Felix The: okay, okay. three hours, harder. Three days. Three weeks, it's far harder. not just because the echo predicting a longer event horizon is harder, the distribution skew of a heavy tail is hard. external factors that you basically didn't predict is hard. to so to bring it to video games again, day twenty-eight. What makes you think I can predict it correctly? Because you might change your game by that time, right? But it's day seven, the likelihood of you changing your game in seven versus twenty eight days, statistically speaking, I can tell you there's far less variability about an external factor. kind of getting in the way of the complexity. So if you compound all those factors, it just requires a different modeling, sophistication and treatment to get comfortable with the idea of day twenty eight.
Josh Chandley: That's interesting.
Felix The: there's one more. Sparse event and a slow data coming back. So you you get feedback on day seven after seven days, right? So in this case you're
Josh Chandley: Yes.
Felix The: gonna you're gonna wait. you're gonna wait. And then when when you wait, when you basically then ingest the feedback back, whether you predict it correctly or not, and calibrate, that whole process takes a more arduous process and takes longer. So your appetite for risk and being right, has to be high. before you basically embark on day twenty eight.
Josh Chandley: why hasn't the whole industry already moved to day 28? Is there something about the the external environment that made it hard until now, or did we just not prioritize it? Like why why is it just two NPA ad networks?
Felix The: well I don't know how to I don't know how to basically s answer that. I can speculate. I mean we we launched day twenty-eight quite recently too. We started with day seven. So I think typically when people do R and D development, it's quite common for you to basically do a trade-off about like risks of a major failure, like your rocket ship launch analogy, right? Versus something that we know like this is a guaranteed or a more guaranteed success, which is the first path that typically most R and D will go for. so if if there is a reason why day twenty eight say it's not that common in the industry, it might just because it's a far harder problem to solve than day seven, but we did it. So that's pretty cool.
Josh Chandley: I I I wonder though, like one one thing I wonder is that clearly day twenty-eight is a is a large structural advantage when almost nobody else offers it. But
Felix The: Mm-hmm.
Josh Chandley: what happens to that advantage when the majority of in-app ad networks also eventually release those day twenty-eight campaigns? How do you see that?
Felix The: We technically never I mean, before day twenty eight was launched, like day seven was the competitive territory, right? When we launched Vector, we launched it on day seven. We are pretty proud of that. So we don't see entry as a as a as a threat. it doesn't change our R and D roadmap. It does not change the fact that we will continuously make day twenty eight a better product.
Josh Chandley: Are you gonna go longer? Like day ninety or PLTV or PROS?
Felix The: no no no comment on that yet.
Josh Chandley: All right. The reason I ask is because, you know, if you're not, then this advantage does have an expiry date. You know, seeing your incredible progress in Q two is probably a strong signal to other ad networks. if if you are, then it that it's an arm race and that's really interesting.
Felix The: If you wanna predict a longer horizon, the the amount of knowledge that you have about behaviors on that longer horizon becomes key.
Josh Chandley: well let's talk about this quarter. Bromberg said that unlike in Q two, you won't ship twenty enhancements every quarter, but Unity is still guided to a a similar growth rate anyway. You know, so if product velocity slows down, what carries that number?
Felix The: I I wouldn't characterize product velocity slowing down. I think maybe one way you
Josh Chandley: Okay.
Felix The: can interpret that that number is like you can't equate, like let's say if I tell you I ship 20 releases in Q1, I ship 10 releases in Q2. I'm making this number up. There's no way to say this 10 is less valuable than the 20 because this number is just a
Josh Chandley: Correct.
Felix The: quantity of release, not the quality of those release. So I think using quantity as a matrix of velocity and impact is a vanity matrix. It doesn't really tell the the underlying value beneath. so our plot product velocity is as probably as fast as as it has ever been. It's one of those things, you know? the law of motion, when things are moving, it's pretty hard to stop it.
Josh Chandley: And you've been moving for I think six straight quarters now of growth. before developer data or runtime was in the models. So something else was the advantage. What was it?
Felix The: It's it's the it's the it's the golden triangle if you wanna call it. Like what what feature you use is quite important, your model architecture is quite important. How do you basically decide how these things basically fit together in your platform environment is important. When I say platform environment also like the underlying support tools, so you can basically do this philosophy at scale, right? Like for example, if I were to basically have like a completely different model for everything that I do, completely different. Completely different data pipeline, completely different model, completely different blah, completely different training process, complet you get the point. That's gonna slow me down quite, quite tremendously. What we've done is we we know how to do good engineering architectural design. Things are fast because we design it in a way to allow us to do R and D fast. So it's all all those three. They they they don't there there's no way for for for any companies I think in the personalization space to say, I have a great model. Like that's just a false statement. It's it's it's always one of the several things that you need to do to execute.
Josh Chandley: Do you still believe there's a a lot of runway there even without the the developer data?
Felix The: Yes, absolutely. the thing that is nice when you're basically making R and D development that is gaining momentum is it does open you up to what more you can do in the same area that you already started. So we're we're not done. The model architecture side, the model architecture of Vector now versus what we did when we first launched it, is not the same. so we we're still innovating. I there's a lot of room.
Josh Chandley: So was engine data ever in your mind the critical differentiator, or do you view it as more the the cherry on top?
Felix The: The engine data there there's there's a very subtle way to basically answer that question, which is like differentiator. You can answer differentiator in in two ways, right? One, the only thing that matters, or one, something you have that say, several others, if not many others, don't have. for us, we view engine data mainly from the lens of the following. And I think like a lot of people also confuse the the concept of engine data like, what type of data do you collect? that's a very narrow definition about about about about data. probably this is not a this is not like a a big reveal, but you know, the industry in in AI, in in personalization have really strongly adopted the idea of like sequence and causality is very important, right? You know, people basically using GPUs, they use Transformers. if you wanna basically do s causality and sequence, the idea of data that you need to use in which you need to basically join like let's say a number of disparate data sources from different people is hard because I can tell you like The idea of timestamp as a concept by itself is something you should actually worry. It's a suspect, like you know, like the the timekeeper and the clock, it's like different clocks, right? But so the idea of like data and runtime, don't don't define it narrowly as like, so what type of data do you have that other people don't have? It's the fact that when you boot up your game, and since
Josh Chandley: Mm-hmm.
Felix The: games are made with Unity, you know, a a good number, more than a good number, those games boot up our runtime. Our runtime becomes the best record keeper about what happened when in what sequence at all times, from the moment you start to the moment you end the game. This type of information, less so the individual information that is being collected, is an ideal platform when you want to know that your data quality is pristinely quality clean for the purpose of understanding what people do.
Josh Chandley: what you're what you're sort of arguing for here is that there's yeah, there's how much data, there's what data points you come in, you have coming into you, but you're describing what what I would call first party versus third party data sources. The ability to actually own the whole pipe and determine what's coming in and standardize and make it one clean thing. Is that a good layman way of thinking about that?
Felix The: maybe first party data typically you would associate it with like let's say there's a there's a technology company that owns their property. They own the video game or they own the app, right? You can probably think about a couple companies that do this. That is the classical definition of first party data because it's yours. It's it's it's not you representing someone else. It's yours because the application is yours. Unity is not that. We don't own the applications. Like when when I partner with you, it's your game, it's yours. so the difference between first party data and what I describe is in the third-party app open app environment, which is what I describe like video games that got released to like Play Store and App Store, there is no such thing as a software provider to have a first party data. Because the first party data that that Can be generated is if you yourself collect that data. Our role as a platform and us powering the runtime of many of these games put us in a position where we can basically make sure that the data that is being transmitted, collected, on behalf of you is clean, standardized, and of quality.
Josh Chandley: And what does that engine and developer data add to vector? Do you think about it as it being more of the same or or something fundamentally different?
Felix The: It's my answer prior. Basically,
Josh Chandley: Okay.
Felix The: there's a a a bit of both, right? Maybe there are some data events about like something that is game specific that standard typical events from ad stuff doesn't doesn't really collect. Like there's many taxonomy and nomenclature that is very quite video game specific. Like level. What is a level? Like outside of video game, what what what is that? your player currency. What what what is that, right? There's a lot of nomenclature that runtime data can collect that is very specific to video games, so that's one. But the second part is what I described earlier, which is like the cleanliness and the pristine quality and the preservation of time order sequence and causality is very very important. So don't think of runtime as just like what data, but the platform itself is the number one source when it comes to a mate with Unity game being open to the moment it closed, we have the time order sequence from us directly.
Josh Chandley: What is the key choke? Is it developer opt-in? Is it the the version of the Unity engine each game runs on, or or is it work on your side that hasn't happened yet?
Felix The: Gosh. You know the one thing that I feel like we are actually quite excited about, to answer your question in in a different way.
Josh Chandley: A non answer or an answer?
Felix The: one of the one of the things that actually was hard in the past before the advent of AI, you know, coding agents and all that stuff, is Upgrading something is hard, you know. But with AI, making changes, testing, making sure that changes are okay becomes easier. so we have two departments in AI that I manage. One is what you call vector, the other one is AI for the game engine itself. You're probably not surprised that we actually have the developer's best interest that if you are one of those studios where You have made your game, your game has shipped. you want to consider how could you upgrade a Unity engine version without actually breaking your game, and that's your impediment. We're working very actively on how to solve that problem for you. That I'm excited about.
Josh Chandley: that is definitely a big challenge as someone who has my oldest Unity game was launched in 2016.
Felix The: Mm-hmm.
Josh Chandley: it it there's an effort. There's effort and risk there.
Felix The: Mm-hmm. Mm-hmm.
Josh Chandley: So what does a developer actually end up getting for turning developer data on? Or enabling the developer data pass back.
Felix The: well I'll I'll I'll answer the obvious. you own that data. So the first thing that you get is data. I'm not trying to be cheeky, but it is the first thing you get is data. So because the difference is like all we provide is the technology to basically collect their data in a standardized quality way. but you
Josh Chandley: Mm-hmm.
Felix The: own that data. So that means developers can use that data in in many many ways. You you you typically would use insights and information with. Like you could use it to manage your game, discover information that you would not otherwise know before about like how people f engage with your game, issues, technology or otherwise with your game. but beyond the ownership of data and insights that you get, activating that on will also mean Should you agree on partnering with Unity in which we can basically utilize that data on behalf of you to improve the services that we render on your behalf, you will experience a better performance in the products that you utilize us for.
Josh Chandley: If we use vector I or unity ads, I assume
Felix The: Mm-hmm.
Josh Chandley: that is something that should pret increase materially also.
Felix The: That is one of the things that we we believe developers will enjoy. And also just to put a final point on it, in addition to just the the the data itself, like there's a lot of products that we basically offer, right? data products like diagnostics is one example. So I think between those three, you activating data, it's your data. Data oriented products that you can use, diagnostics is one example. And third is any services that you use with us in which the utility of data will materially change the quality of the product, factors included.
Josh Chandley: But say I opt into data sharing today. How long before I see a Unity Ads UA uplift? Is it like a month? Is it months? Is it T B D? Help me understand kind of my measure my measurement window.
Felix The: it's a hard machine learning problem to answer because generally speaking like it really depends is the answer. i it really depends about like, you know, in in any personalization AI it's like you wanna basically see the the long tail, right? You wanna see information that you don't otherwise know. So you can basically influence the quality of your prediction. So it really depends on That long tail behavior. If you turn it on and we we see you know the data emitted from your game environment, we're like, yeah, we already know this stuff, then the the good news is like the the improvement is probably quite immediate, it's quite fast. But at the same time, because there's no that long tail power, we probably won't basically materially change the performance performance right away. It will be more modest. But if you produce information, long tail information that we actually have not seen before. I think the improvement is gonna be material. But at the same time, typically machine learning models and AI models are trained so that when you see long tail, don't get overly confident until you actually see enough of that distribution data for you to be confident to basically go all in. so that's why it depends.
Josh Chandley: be comfortable exploiting it takes time. If we if I enable engine data, is that gonna help just my properties or is it gonna help the whole ecosystem?
Felix The: It's it's both but not in equal measure because
Josh Chandley: Okay.
Felix The: it's it because it will primarily help your properties, more so than the the ecosystem overall. Because I I think again in in AI models, assuming every single game is wild card game is is a it's a false it's a it's an incorrect design, right? So that means if the data is being generated by wildcard games That data is mostly relevant for wild card games. From what how we understand how your g game mechanics work, how it will benefit you if certain things are done in a certain way, the players that might enjoy your game, all those. So the answer is both, but the impact is is not proportionate. It should disproportionately favor your environment because it is based on you.
Josh Chandley: That is such an interesting response because one of the biggest concerns I hear is that if I share my my data, my players' behavior will help other advertisers compete with me for the very same players. and it'll help Unity target my players with competition. How do you I think I I know, but how do you think about that?
Felix The: I I think again, like my my understanding is Data and information has a disproportionate value based on who provide that information. Like for example, if there's a machine learning software that tries to predict you and I, if you basically say, Nope, you can't know anything about me, but I said yes, you can know stuff about me, I can tell you they can predict me better than you. so will will me volunteering my information make that machine learning model better predict you? Probably. well one. w we're both humans, to start. So we probably have human behaviors that they can predict quite well. Maybe they will then realize like, that guy is also in the video game industry. Maybe they they have some similar passion for video games. I'm making this up, but you get my point. Like it will always be more specific to me. but you will benefit from the fact that we have common traits and attributes, but it's very glossy.
Josh Chandley: And so that's also why you don't t necessarily have a free rider problem. There's no one that can let everybody else opt in and reap the rewards without sharing their data. They end up just benefiting the least. That's your argument, right?
Felix The: I y yes. My argument is like to your point, I think in this type of data situation, right? the idea of an asymmetry on on data economy exchange is is something that I think a lot of people think about. the good news is in in AI and machine learning, you know, just the way naturally the product's designed, people that basically provide information are the people that will benefit the most from it.
Josh Chandley: It's similar to MP post box.
Felix The: Yeah. And by the way, when I say that, remember I make that statements in generalities. So that means like let's say there's one developer that basically share information and we find based on that sharing of information is like, you know, the the players playing their games are not legitimately staying players or whatever else, then that benefit statement does not materialize. it will be beneficial for the industry though, when when that happens. So
Josh Chandley: In that niche corner case. I see. I see.
Felix The: Yes, yes.
Josh Chandley: I'm also interested in D to C. That's becoming incredibly important. as purchases move more and more to web shops, you know, those transactions stop flowing through all of these pipes in the exact same way. what I wanna understand is Unity launched up a new product commerce. Do
Felix The: Mm-hmm.
Josh Chandley: you view that as an entirely new data source for Vector? Or is it a way of making sure that you don't lose purchase signal that you already had? Like how do you how do you visualize that opportunity?
Felix The: our philosophy when it comes to launching a product is we launch the product for the sake of that product itself. So that means on the commerce solution, we genuinely remember, like I understand the nature of your question because, like, you know, vector, all that, right? But we're not an ad network in the largest, grand scheme of things. We are Unity. We have the game engine. We have relationships with developers that allow us to basically extend our services beyond just vector. a lot of people make their games with unity and they want a commerce solution. They want a web shop. So that is the product. The product is as you make your games with Unity, we believe we have an opportunity to provide you you and developers with a commerce solution. So we want that commerce solutions to be robust, strong, all that stuff. Now, back to the vector side. commerce solutions, yes and no because like Some people have web shops today, you know? and that
Josh Chandley: A lot.
Felix The: web shop yeah a lot. And some of that information about purchase that is being made on those web shops is already reported to the MMPs. So in that sense, like the data's already there, so it's nothing new. but are there web shops and are there transactions that we have seen that does not materialize elsewhere? Yes. Absolutely yes. So I guess it's a little bit of both.
Josh Chandley: So it is incremental to the data set in terms of seeing new purchases in theory.
Felix The: incremental some are incremental, some are probably higher quality if we were to process it differently. And others could just be, you know, what you said. Like we don't know how the industry's gonna go when we can predict. But risk insurance against a future data going to a different channel is always a good risk insurance policy to have. So it's all of them.
Josh Chandley: So bringing this kind of all together. I think Bromberg said the the vast majority of game advert advertisers already know Unity and already spend with you in some capacity. So
Felix The: Mm-hmm.
Josh Chandley: if the next billion dollars in in vectors scale isn't just new logos, what is it?
Felix The: It I I I think it would be also new logos because again like new logos because again like as Unity like peop people use us for game development for game engine. So we do like the idea of new logos because we believe like one of the ways for people to create an impact for the video game industry is to be able to express their creative vision and bring it to life. and as part of us basically being able to participate in that journey With them from the very beginning. we do like the ideas in new logos. I will say that. but on on the other side of the spectrum, beyond new logos, conversion rate in the industry is in a single digit percent, in a low single digit percent, Josh, as you know. So so growth is not predicated on just new logos. Growth is predicated on the fact that you predict correctly. if the conversion rate today, call it 1%, for every hundred, ninety-nine are missed. That's growth. That is a pretty big sizable opportunity for growth. and then materially for the developers that we have today, not just conversion, but prediction of like LTV, you mentioned, right? A longer prediction cohort. That's also an expression of growth. So an expression of growth for me is both new logos, but also in the existing customers that we have, we have so much potential. To generate much more value given the fact that the baseline that we have today, as we celebrate everything that we've done, we got a lot more room.
Josh Chandley: To to visualize that, like with your example, it it does almost lower the bar, but it is incredibly hard the task ahead. You know, if one in a hundred install and and two in a hundred purchase, that means really only one in five thousand actually even do the first target event of an IAP ROAS campaign. To to some extent You know, I actually as a side note, I I hear developers complain about you know CPI rising a lot, but
Felix The: Mm-hmm.
Josh Chandley: if You know, of the five thousand impressions you buy, one causes a purchase. If you purchase two and a half thousand impressions fewer that were doing nothing for you, your revenue creation is still the same. And so the effective
Felix The: Yeah.
Josh Chandley: CPI should double, right? And so that's what you're kind of describing as the opportunity.
Felix The: I I am a big believer of not wasting transactions. I agree with you completely. we track internally about the that area that you mentioned about like, yes, we're proud about the modeling improvements and everything that we've done, but we are very sensitive to make sure that the money spent is well worth it. So that means any prediction that is inaccurate is a transaction that adds no value, and we are here to combat that.
Josh Chandley: Yeah. and even just going down funnel, buying a hundred installs and having five purchasers that spend a hundred dollars versus having twenty installs and having five purchasers spend a hundred dollars. some of the metrics that publishers want to have go down, like CPI, actually lose their correlation with value creation in that way.
Felix The: Yeah. which is which is why I believe like we're you know we're we're one of the earlier, you know, proponents of ROAS as a concept.
Josh Chandley: Mm-hmm.
Felix The: I do believe like the industry, I never have again like affinity about just this or that, right? Metrics need to be rational and needs to make sense when it's tied to your success and the developer's success. When the product expresses that and that is the goal that it seeks, you And the industry will be better for it. So I agree with you. We don't tie CPI, we tie ROS.
Josh Chandley: Which is it's just the way to do it. the one thing on the other side of that though is outside of efficiency gains, there is a competitive aspect where you don't own the inventory. and when my when my business scales an in app ad network and it's successful, that budget very usually comes partially at the expense of other in app ad networks. Like absolutely it's true that there's a value creation component.
Felix The: Mm-hmm.
Josh Chandley: but if I'm a solitaire app, you know. E everyone knows I want to buy solitaire app traffic. There
Felix The: Mm-hmm.
Josh Chandley: is some level of direct competition. I assume that's true for your customers, too right? And how much how much do you think of your growth is going to be value creation versus share capture?
Felix The: Meaning like, how much of the new customers we're getting that is coming from share capture versus new?
Josh Chandley: Share capture versus you found a new user for this app that nobody else knew about.
Felix The: I d I'm not sure if I'm positioned to answer that question because I don't have access to my competitors information.
Josh Chandley: That's true.
Felix The: all I can say is like we focus on delivering the best ROS and the best quality products for our customer and our customer will then decide what type of actions they want to take. And what we've seen so far, like for example with the t day twenty launch, as they're satisfied with the ROS that we deliver for them at scale, they basically just say, I want more of this. Whether that action impacts our competitor or whether that's just truly incrementally new budget, I am not privy to that information. All I know is I receive, you know, more spend.
Josh Chandley: You just yeah, you just had you have more inventory flowing. Or more auctions flowing.
Felix The: Yeah. But I w right. But I would like to think like generally speaking for this industry, because I really truly believe it's not a zero sum game, the industry will benefit by the many network providers to basically better at their product and then you, in terms of acquiring players, you can basically have more choices to basically spend more dollars in return for a greater return. So as long as that greater return is true, you technically would always increase your appetite to basically acquire more players because it will guarantee you return. That is one way to look at it. But again, like that's just again us talking about generally, like mathematically in the industry. But from my side, I am not privy to you know how how you manage your businesses with the other networks. I can only speak to mine.
Josh Chandley: And so that that kind of covers, you know, gaming demand flowing into in app at Where else could the next billion come from? Like what's more exciting to you? Is it new demand that isn't gaming, or is it the same gaming demand flowing into inventory that isn't an app?
Felix The: Both could be exciting. I would say on the on the the modeling side, when you already have a model that is sophisticated enough to predict multiple things, right? Like day twenty eight, day seven, retention, purchase, this, that. the idea of prediction in a separate category is not an insurmountable task. In terms of your other question about channel, like again, like I think I think you're you you you probably know like generally speaking, we don't never we never view about channel as a as a identity for a product. Like, yes, Factor is in mobile in app today, but we never actually design Factor so that this is definitely a hundred percent mobile in app product. The idea that we have about Vector is is player acquisition product. We want to find players where they're at.
Josh Chandley: So W what I just heard, if I'm parsing this correctly, is like the the the general like principle here is that good prediction shouldn't be limited by demand type or inventory type. was it the plan for Vector all along to be as extensible as possible in the way we're talking about?
Felix The: I think there's two ways to answer that question. You don't make a plan, assuming your plan is perfect, and you don't change it as new information is available to you. But if you don't have a conviction about where the puck is going and what's possible from the very beginning, then you might make the wrong thing. So factor the way we design it is meant to be extensible. It is meant to be efficient. The platform and the underneath the chassis that support it is meant for it to be able to grow and expand. but whether we from day one had that expansive idea from the very beginning, I would say that would be more like a hypothetical ambition rather than a plan. But the plan all along has always been how do we transform our AI personalization product so we can achieve The success that the developers deserve that is not just today, but for years to come.
Josh Chandley: So It's interesting to me because that what the way that makes me think is it makes me think like so in 2024 the the Unity narrative was as I say pessimistic. some people would say it was a dying company, which is ex you know, a strong word. But actually you were you were planning a strategy to scale not just gaming and app, but to create something as a that could potentially expand the TAM quite dramatically.
Felix The: You you let me I I I hear your question. Let me put a final point to it. when you plan, especially in twenty twenty four, if we rewind back the time, when you actually are thinking about something more crucial than just mere expansion,
Josh Chandley: Mm-hmm.
Felix The: it would be unwise to basically think about the grand ambition to stand in the way of doing what's right at the moment of time, to be good at the the core value that you're offering. But it would also be equally foolish if you basically designed it so bespoke That it basically limits your ability for what will happen next should you be successful. There's a fine balance on how do you do it, but in this case, it's not a trade-off. They kind of converge. A good engineering is good engineering. a good engineering means when you basically design a system, you should not design it bespoke. You should design it to be really good at what it does now and it's capable to be good for the new stuff.
Josh Chandley: I see. Does a model does a model trained on game behavior predict anything useful about someone who isn't playing a game?
Felix The: it depends on the the the prediction head, right? So the prediction
Josh Chandley: Mm-hmm.
Felix The: head that is like say if you're trying to predict level complete. Yeah, absolutely not, because there's there's nothing in common. But if the prediction head is like your willingness to purchase something, you buy stuff in general. So that would be relevant. So it depends on the prediction head. But in any personalization, you don't predict things individually, right? Like you you you combine the prediction head to basically say this is the quality of my outcome. This is what I recommend. So the answer is yes for some, no for others.
Josh Chandley: Okay, let's start to land this. There's a there's a decision maker listening, you know, right now and he's he or she is setting budgets for the rest of Q four. Make the case. Like why should we why should we test now rather than waiting a few more quarters?
Felix The: tests day twenty eight or vector in vector in general
Josh Chandley: vector in general, or day twenty eight, yeah. Vector in general.
Felix The: vector in general I I hope we don't have to convince people to to test we are we are kind of beyond the testing phase. people have been using it at scale pretty much if not all of the traffic on our network is basically powered by vector. so if you're new to vector Give it a shout, give it a try. it's been proven. If it's a newer subsection of vector product, which is day twenty eight, mm my advice would be we have seen tremendous success of people that adopted day twenty eight. So I would highly recommend everybody to try it. But I would also advise this: don't try by turning off a product that already works for you and then trying to hop this on. You should always make a rational choice. And a very permissible choice for you so you don't impose yourself a risk of s something. keep the products that work, try new things, see the metrics, see the ROAS. When you're happy, you allocate budget proportionally based on the metrics that you see are leaning more in your favor.
Josh Chandley: So my final question about runtime and developer data, so we can kind of understand this is six months from now, what is operators if they're if they share runtime and they are spending into vector and investing into that product, in six months, what's one thing that an operator can observe in their data that would tell them if this thesis is playing out the way you expect?
Felix The: You would not probably see it in the data explicitly, per se. You would see it based on your performance is the easiest way to to to answer it. But you would not technically be able to say like all right, I see plus X percent. A portion of that X is runtime, a Y portion of it is this, Z is that. It's just a compound effect. Because for us, we never actually parse that information for developers to to justify like this is the benefit from X, Y, and Z. Our loyalty is we want to b give you the best outcome, period. So whether that outcome comes from this or that, we're working our hardest to make sure that they are all materially going to benefit you in a meaningful way.
Josh Chandley: Unity has come a long way with with Vector. what are you most confident about in regards to its its long-term chance of success? What do you think it's gonna change for publishers?
Felix The: Our industry at large has room to to materially improve itself. there are so many ways. if you want to talk about the quality of ads being shown, the the nature of the at unit, how how user-friendly is it for your player so they will be retained in your game. whether it's the idea of Understanding players' preferences so we can actually predict things better, like the conversion rate in the single digit percent that I mentioned. I believe the future of this industry is bright because there's so many i inefficiencies that can be squashed, there are so many opportunities for us to improve, whether it's the actual behavior of the player and the conversion manifested by the ad unit they see and the implication to LTV
Josh Chandley: Mm-hmm.
Felix The: in the game, or the prediction quality, all of these are nowhere close to being enough. So I believe by having providers such as us to focus a lot, if not the only soul focus we have, at the heart of the matter, which is providing quality prediction, quality user acquisition for our developers, and also on the publisher side, to make sure that the experience that the players see is of quality, the industry will be better overall for it.
Josh Chandley: Well, Felix, it it's been a pleasure. Thank you for joining me on Mobile Signal.
Felix The: Thank you, Josh.
Josh Chandley: For the audience, if you made it this far, thanks for listening. few more thoughts after talking with Felix. First, D twenty-eight is real. My expectation is that if you're outperforming by some percent on day twenty eight, you're outperforming by more at day three sixty five. that's a claim though that you can test on your own portfolio. So run it alongside day seven and measure maturation. Bigger question though is developer data, the runtime signal. My read is vector is still early in using it. Their growth so far has been from solving other other problems. So runtime and executing against that is the next major test for Unity, where if they can succeed, that will unlock a new level of growth. I think for publishers The best outcome is just more networks competing on prediction quality. and I think that's coming. That's good for everybody. I would say in twelve months, so Oc October twenty twenty-seven, at least at least three more networks go GA with day twenty-eight optimization. I'll come back and score that call. So subscribe if you want to see how it lands.
Felix The: Mm-hmm.
Josh Chandley: What did you believe about the product at the announcement that you no longer believe now?
Felix The: Well, let's just say before I touch upon the product, probably the more important bit is when you're launching something new in which you believe it's transformative, but you don't have any empirical evidence outside of your own lab of what that thing says, right? Not yet launched to the customer. The only thing that will carry you forward is the conviction you have based on your expertise, research and evidence and otherwise that. This is the moment that you do not lose steam despite whatever you see out there. so if there's one thing that I guess I have now that I didn't have back then is the empirical evidence about the fact that the customers actually love it. So yeah.
Josh Chandley: Well, yeah, well, an upside surprise is a very positive thing to be wrong about. and so I've I've been watching Vector's progress as a customer, you know, both on this demand and the supply side. we've been active users. What was the hardest thing to get working that you didn't expect? Like, cause what you're describing is totally true. Like when you hit the road, that's the moment when every little
Felix The: Mm-hmm. Mm-hmm.
Josh Chandley: mistake becomes much more obvious. It's also the moment you can improve the fastest. So what was the hardest thing to get working?
Felix The: it's a zero to one moment because it's it's very hard for most companies to basically even have the ambition to disrupt what they already have. Especially when you say like what you already have is like not bad. It's just that for the ambition that you believe and what the industry really should have, you could do better. it's kind of like you have a plane but you have to make a new one too at the same time when the plane is still flying. That transitionary moment when you need to disrupt yourself and you need to beat yourself was hard. because like think about it, like when Factor was still baby in the lab, it doesn't get traffic even in our own environment, right? Like our legacy model basically got the majority of the traffic. That switchover, when you basically starve the baby, but you don't want to starve the baby, but you have to run the plane, that was a very artful orchestra that we need to basically meet a trade-off. And in some cases, when you're in that moment Being courageous is the most difficult thing. Which is like, I know, it doesn't look like at this current state that it is ready for prime time, but we gotta give it some air debris because it's going to be born. That was hard, but we did it.
Josh Chandley: It's so interesting to me because you are talking to me like every GM that launches a game, where,
Felix The: Mm-hmm.
Josh Chandley: yeah, my team launches a new game every like many years. This is always a soft skill. This is always challenging. and there's always this period when you launch it, it it needs to get volume, but it's gonna be a little rough at first. And we have to kind of accept that going into it.
Felix The: Yeah. Yes, and and and that's like say launching a game. there are two cases. Like if you're launching a game in which your displacement costs or your opportunity cost is like
Josh Chandley: Mm.
Felix The: zero zero, that's like a risk you take. But in my case, I gotta take air from the legacy. Yeah,
Josh Chandley: You were killing the old game. Yeah.
Felix The: correct, correct. Yeah. So that was hard. And then when you basically take air of the legacy and you know it's not exactly at a performance state yet because you need to give it air, you're gonna You're gonna be okay with the fact that you are you're taking some risk. and we did it and I was very pleased with the outcome.
Josh Chandley: there's lot of pressure in that moment. There's a lot going on externally too. I I would say that the early skepticism was particularly ungentle. You know,
Felix The: Mm-hmm.
Josh Chandley: Unity had been promising the engine data advantage for years and never shipped it. The read was, you know, vector was more more of the same, almost like a distraction. Was any of
Felix The: Yeah. Yeah.
Josh Chandley: that fair?
Felix The: I would say there's no way for me to be in a position to characterize fair or unfair. The market spoke. The market has, you know, wisdom in the crowd about what they believe in based on performances of history of companies of the past. it is what it is. All we could say is like, all I could say is you don't let fairness or unfairness of a comment to distract you about what you would like to do. And we were focused on the mission and we delivered.
Josh Chandley: Why didn't the runtime or developer data ship in that previous era of Unity pre vector?
Felix The: I I think Unity has gone through many different stages. Many different stages. the company itself is, you know, it's it's it's decade long old. There's usually a time in a company's stage of maturity in which a certain type of product is something that the company is able to deliver and for the community to accept and for us to be able to actually form it in a way that we basically Provided with our customers in mind. In the earlier stage of Unity, Unity has always been focused on like, you know, if you want to call it as we still do today, a game development engine. Yes? A game development engine, and when you break it down into its core of primitive, the most core primitive component, you gotta be able to use it to make something. So the idea of like data, when you're focused on that principal core, is like, is that Just a luxury item that we need to think about now. But as games become more sophisticated, as the industry becomes more competitive, the idea of data is something that we believe the developers should own because they could actually control their own destiny, what they would like to do with that information. You want to use it to manage your stuff, your games, your live ops, your whatever, in many different ways. Data gives you the insights to do so. You wanna basic a partner with certain vendors in which that data can actually amplify the activities you're trying to do, you can do so. So that's the nature of how we ended up with developer data framework. We believe this is the moment where Unity can cement its role. It is not just a classical game engine where you can make stuff with it, but you can actually operate your titles with it with confidence too.
Josh Chandley: That framework, as it applies to vector, is it rolled out and bidding live and auctions today at scale? Or is there data flowing and it's just not being quite used yet?
Felix The: The developer data framework runtime is
Josh Chandley: Mm.
Felix The: being used. We're very pleased with the results that we have seen today. we have more room to grow on adoption, and that adoption is just a number of things that we naturally will see as people upgrade and they basically release new games. And also the way we basically present these options for developer to opt in. we display it prominently in our dashboard, So we can we can partner together on what you would like us to do with it or you would like to do it yourself. so it's a it's something that is already used, but we are very interested and excited about what other surface areas we can actually display this opt-in that will accelerate its adoption.
Josh Chandley: I see. So opt-in is still a key area. Well, let's look back to Q2. Last quarter, vector's growth rate accelerated to 23% sequentially. You passed
Felix The: Mm-hmm.
Josh Chandley: a billion dollar run rate. If you had to name the single biggest driver, what was it?
Felix The: It's hard for me to name a single biggest driver, not because I don't want to name it. There is one if I just basically stack rank like the list of things we did in a descending order, there is one. But in the aggregate whole, just to answer your question directly, it's a combination of like model architectural improvements, a combination of like, you know, how do we basically do feature engineering? It's hard to pinpoint to one because we never basically have that singular aha moment. It's a continuous improvements of the model over time. If you twist my arm and really have to pick one that is
Josh Chandley: I will.
Felix The: probably in a recent memory, I think you won't be surprised. Day twenty-eight has been quite successful. and day 28 is quite successful because you know, you you operate video games yourself. day seven sometimes often is just not sufficiently long for you to actually understand the maturity of your cohort, especially when your game mechanics and that's the interaction between your users and your game. You don't really get that clear signal on day seven about who's going to basically behave in a certain way, you know what I mean? So day twenty eight is is such a successful product for us, so that in recent memory, one thing I would volunteer.
Josh Chandley: Yeah, something we've been looking forward to for a long time. I mean, a a lot of IAP monetized games, for instance, say that their median, their median whales first purchase doesn't occur until week two. So so having the depth beyond day seven is incredibly valuable. And that's I'm assuming that's why over a quarter of your customers adopted day twenty eight in Q2. What kind of advertiser moved first?
Felix The: You won't be surprised. there's usually like people market typically reacts in a rational manner. the people that move first are developers that represent the genre in which like longer retention players are key to them. in which like the monetization events are not clear that you can get that clarity on the early days. RPG, strategy games, like but there's also puzzle and casual. So It's not limited to one genre. It really depends on your game. If your game values that monetization unlock moment on the latter days, people tend to move quicker. The second part also, day twenty-eight, especially when you're basically launching a new product, it does carry more degree of risk, right? When you basically launch a new product, the earlier adopter is are the people that basically use the product first. So what we find, the developers that move are the people that actually are comfortable with the idea of day twenty-eight. They tend to be very sophisticated. They also appreciate the fact that day twenty eight is the prime maturity cohort they care most about and they also have larger budget. that's
Josh Chandley: I see.
Felix The: that's the that's the type of people that adopt it.
Josh Chandley: Okay. And then I'm assuming the the adoption is lagging for essentially the opposite. customers that have essentially faster turnover, they're less total yield focused, more IRR focused, just turning their money over quickly.
Felix The: Yeah. Correct. Yeah.
Josh Chandley: I see. Why was moving from day seven to day twenty eight, ROAS optimization, such a such a step change?
Felix The: well one is day seven is day seven. Day twenty-eight is day twenty-eight, day whatever, and day seven. Because like you can't assume a successful day twenty-eight if you're completely blind of the prediction mode in day seven. i in a way like the reason why day twenty eight is really key for us is not only because like it is Pretty much a natural explanation about how the market wants to see cohort maturity because day seven doesn't guarantee the outcome of day twenty eight. But the model quality itself is a far harder model quality to achieve greatness at. Because you need to be able to predict day seven really well if you want to have any attempt in predicting day twenty-eight. so that's why I think it's such a wonderful step function because it forces us to actually up our game in the sophistication of our prediction. And given the fact that we are getting really comfortable that we're getting momentum on factor day seven, that gives us the ambition and and the appetite to tackle day twenty eight the right.
Josh Chandley: So I the way I visuali visualize your problem, which may be insultingly simplistic to you. is this, you know, if you build a a rocket to take satellites to space, that
Felix The: Mm-hmm.
Josh Chandley: doesn't mean you can get to the moon by making a four times larger rocket. You know, this harder, longer mission creates new failure points. I'm assuming
Felix The: Mm-hmm.
Josh Chandley: that was the same kind of challenge you face with D-28. So what got harder to do at day twenty eight versus day seven?
Felix The: like for example, if you predict what I'm gonna do in the next thirty minutes, the next three hours, the next three days, the next three weeks. There's three in common there. The next thirty minutes, unless you feel like, you know, you don't enjoy this conversation so much, you're gonna kick me out early, chances are we're still gonna be talking.
Josh Chandley: Ha ha.
Felix The: so it's an easy prediction. It's an easy
Josh Chandley: Yeah.
Felix The: pr prediction. But three hours? Yeah,
Josh Chandley: I'm glad you're that confident, but go on.
Felix The: okay, okay. three hours, harder. Three days. Three weeks, it's far harder. not just because the echo predicting a longer event horizon is harder, the distribution skew of a heavy tail is hard. external factors that you basically didn't predict is hard. to so to bring it to video games again, day twenty-eight. What makes you think I can predict it correctly? Because you might change your game by that time, right? But it's day seven, the likelihood of you changing your game in seven versus twenty eight days, statistically speaking, I can tell you there's far less variability about an external factor. kind of getting in the way of the complexity. So if you compound all those factors, it just requires a different modeling, sophistication and treatment to get comfortable with the idea of day twenty eight.
Josh Chandley: That's interesting.
Felix The: there's one more. Sparse event and a slow data coming back. So you you get feedback on day seven after seven days, right? So in this case you're
Josh Chandley: Yes.
Felix The: gonna you're gonna wait. you're gonna wait. And then when when you wait, when you basically then ingest the feedback back, whether you predict it correctly or not, and calibrate, that whole process takes a more arduous process and takes longer. So your appetite for risk and being right, has to be high. before you basically embark on day twenty eight.
Josh Chandley: why hasn't the whole industry already moved to day 28? Is there something about the the external environment that made it hard until now, or did we just not prioritize it? Like why why is it just two NPA ad networks?
Felix The: well I don't know how to I don't know how to basically s answer that. I can speculate. I mean we we launched day twenty-eight quite recently too. We started with day seven. So I think typically when people do R and D development, it's quite common for you to basically do a trade-off about like risks of a major failure, like your rocket ship launch analogy, right? Versus something that we know like this is a guaranteed or a more guaranteed success, which is the first path that typically most R and D will go for. so if if there is a reason why day twenty eight say it's not that common in the industry, it might just because it's a far harder problem to solve than day seven, but we did it. So that's pretty cool.
Josh Chandley: I I I wonder though, like one one thing I wonder is that clearly day twenty-eight is a is a large structural advantage when almost nobody else offers it. But
Felix The: Mm-hmm.
Josh Chandley: what happens to that advantage when the majority of in-app ad networks also eventually release those day twenty-eight campaigns? How do you see that?
Felix The: We technically never I mean, before day twenty eight was launched, like day seven was the competitive territory, right? When we launched Vector, we launched it on day seven. We are pretty proud of that. So we don't see entry as a as a as a threat. it doesn't change our R and D roadmap. It does not change the fact that we will continuously make day twenty eight a better product.
Josh Chandley: Are you gonna go longer? Like day ninety or PLTV or PROS?
Felix The: no no no comment on that yet.
Josh Chandley: All right. The reason I ask is because, you know, if you're not, then this advantage does have an expiry date. You know, seeing your incredible progress in Q two is probably a strong signal to other ad networks. if if you are, then it that it's an arm race and that's really interesting.
Felix The: If you wanna predict a longer horizon, the the amount of knowledge that you have about behaviors on that longer horizon becomes key.
Josh Chandley: well let's talk about this quarter. Bromberg said that unlike in Q two, you won't ship twenty enhancements every quarter, but Unity is still guided to a a similar growth rate anyway. You know, so if product velocity slows down, what carries that number?
Felix The: I I wouldn't characterize product velocity slowing down. I think maybe one way you
Josh Chandley: Okay.
Felix The: can interpret that that number is like you can't equate, like let's say if I tell you I ship 20 releases in Q1, I ship 10 releases in Q2. I'm making this number up. There's no way to say this 10 is less valuable than the 20 because this number is just a
Josh Chandley: Correct.
Felix The: quantity of release, not the quality of those release. So I think using quantity as a matrix of velocity and impact is a vanity matrix. It doesn't really tell the the underlying value beneath. so our plot product velocity is as probably as fast as as it has ever been. It's one of those things, you know? the law of motion, when things are moving, it's pretty hard to stop it.
Josh Chandley: And you've been moving for I think six straight quarters now of growth. before developer data or runtime was in the models. So something else was the advantage. What was it?
Felix The: It's it's the it's the it's the golden triangle if you wanna call it. Like what what feature you use is quite important, your model architecture is quite important. How do you basically decide how these things basically fit together in your platform environment is important. When I say platform environment also like the underlying support tools, so you can basically do this philosophy at scale, right? Like for example, if I were to basically have like a completely different model for everything that I do, completely different. Completely different data pipeline, completely different model, completely different blah, completely different training process, complet you get the point. That's gonna slow me down quite, quite tremendously. What we've done is we we know how to do good engineering architectural design. Things are fast because we design it in a way to allow us to do R and D fast. So it's all all those three. They they they don't there there's no way for for for any companies I think in the personalization space to say, I have a great model. Like that's just a false statement. It's it's it's always one of the several things that you need to do to execute.
Josh Chandley: Do you still believe there's a a lot of runway there even without the the developer data?
Felix The: Yes, absolutely. the thing that is nice when you're basically making R and D development that is gaining momentum is it does open you up to what more you can do in the same area that you already started. So we're we're not done. The model architecture side, the model architecture of Vector now versus what we did when we first launched it, is not the same. so we we're still innovating. I there's a lot of room.
Josh Chandley: So was engine data ever in your mind the critical differentiator, or do you view it as more the the cherry on top?
Felix The: The engine data there there's there's a very subtle way to basically answer that question, which is like differentiator. You can answer differentiator in in two ways, right? One, the only thing that matters, or one, something you have that say, several others, if not many others, don't have. for us, we view engine data mainly from the lens of the following. And I think like a lot of people also confuse the the concept of engine data like, what type of data do you collect? that's a very narrow definition about about about about data. probably this is not a this is not like a a big reveal, but you know, the industry in in AI, in in personalization have really strongly adopted the idea of like sequence and causality is very important, right? You know, people basically using GPUs, they use Transformers. if you wanna basically do s causality and sequence, the idea of data that you need to use in which you need to basically join like let's say a number of disparate data sources from different people is hard because I can tell you like The idea of timestamp as a concept by itself is something you should actually worry. It's a suspect, like you know, like the the timekeeper and the clock, it's like different clocks, right? But so the idea of like data and runtime, don't don't define it narrowly as like, so what type of data do you have that other people don't have? It's the fact that when you boot up your game, and since
Josh Chandley: Mm-hmm.
Felix The: games are made with Unity, you know, a a good number, more than a good number, those games boot up our runtime. Our runtime becomes the best record keeper about what happened when in what sequence at all times, from the moment you start to the moment you end the game. This type of information, less so the individual information that is being collected, is an ideal platform when you want to know that your data quality is pristinely quality clean for the purpose of understanding what people do.
Josh Chandley: what you're what you're sort of arguing for here is that there's yeah, there's how much data, there's what data points you come in, you have coming into you, but you're describing what what I would call first party versus third party data sources. The ability to actually own the whole pipe and determine what's coming in and standardize and make it one clean thing. Is that a good layman way of thinking about that?
Felix The: maybe first party data typically you would associate it with like let's say there's a there's a technology company that owns their property. They own the video game or they own the app, right? You can probably think about a couple companies that do this. That is the classical definition of first party data because it's yours. It's it's it's not you representing someone else. It's yours because the application is yours. Unity is not that. We don't own the applications. Like when when I partner with you, it's your game, it's yours. so the difference between first party data and what I describe is in the third-party app open app environment, which is what I describe like video games that got released to like Play Store and App Store, there is no such thing as a software provider to have a first party data. Because the first party data that that Can be generated is if you yourself collect that data. Our role as a platform and us powering the runtime of many of these games put us in a position where we can basically make sure that the data that is being transmitted, collected, on behalf of you is clean, standardized, and of quality.
Josh Chandley: And what does that engine and developer data add to vector? Do you think about it as it being more of the same or or something fundamentally different?
Felix The: It's my answer prior. Basically,
Josh Chandley: Okay.
Felix The: there's a a a bit of both, right? Maybe there are some data events about like something that is game specific that standard typical events from ad stuff doesn't doesn't really collect. Like there's many taxonomy and nomenclature that is very quite video game specific. Like level. What is a level? Like outside of video game, what what what is that? your player currency. What what what is that, right? There's a lot of nomenclature that runtime data can collect that is very specific to video games, so that's one. But the second part is what I described earlier, which is like the cleanliness and the pristine quality and the preservation of time order sequence and causality is very very important. So don't think of runtime as just like what data, but the platform itself is the number one source when it comes to a mate with Unity game being open to the moment it closed, we have the time order sequence from us directly.
Josh Chandley: What is the key choke? Is it developer opt-in? Is it the the version of the Unity engine each game runs on, or or is it work on your side that hasn't happened yet?
Felix The: Gosh. You know the one thing that I feel like we are actually quite excited about, to answer your question in in a different way.
Josh Chandley: A non answer or an answer?
Felix The: one of the one of the things that actually was hard in the past before the advent of AI, you know, coding agents and all that stuff, is Upgrading something is hard, you know. But with AI, making changes, testing, making sure that changes are okay becomes easier. so we have two departments in AI that I manage. One is what you call vector, the other one is AI for the game engine itself. You're probably not surprised that we actually have the developer's best interest that if you are one of those studios where You have made your game, your game has shipped. you want to consider how could you upgrade a Unity engine version without actually breaking your game, and that's your impediment. We're working very actively on how to solve that problem for you. That I'm excited about.
Josh Chandley: that is definitely a big challenge as someone who has my oldest Unity game was launched in 2016.
Felix The: Mm-hmm.
Josh Chandley: it it there's an effort. There's effort and risk there.
Felix The: Mm-hmm. Mm-hmm.
Josh Chandley: So what does a developer actually end up getting for turning developer data on? Or enabling the developer data pass back.
Felix The: well I'll I'll I'll answer the obvious. you own that data. So the first thing that you get is data. I'm not trying to be cheeky, but it is the first thing you get is data. So because the difference is like all we provide is the technology to basically collect their data in a standardized quality way. but you
Josh Chandley: Mm-hmm.
Felix The: own that data. So that means developers can use that data in in many many ways. You you you typically would use insights and information with. Like you could use it to manage your game, discover information that you would not otherwise know before about like how people f engage with your game, issues, technology or otherwise with your game. but beyond the ownership of data and insights that you get, activating that on will also mean Should you agree on partnering with Unity in which we can basically utilize that data on behalf of you to improve the services that we render on your behalf, you will experience a better performance in the products that you utilize us for.
Josh Chandley: If we use vector I or unity ads, I assume
Felix The: Mm-hmm.
Josh Chandley: that is something that should pret increase materially also.
Felix The: That is one of the things that we we believe developers will enjoy. And also just to put a final point on it, in addition to just the the the data itself, like there's a lot of products that we basically offer, right? data products like diagnostics is one example. So I think between those three, you activating data, it's your data. Data oriented products that you can use, diagnostics is one example. And third is any services that you use with us in which the utility of data will materially change the quality of the product, factors included.
Josh Chandley: But say I opt into data sharing today. How long before I see a Unity Ads UA uplift? Is it like a month? Is it months? Is it T B D? Help me understand kind of my measure my measurement window.
Felix The: it's a hard machine learning problem to answer because generally speaking like it really depends is the answer. i it really depends about like, you know, in in any personalization AI it's like you wanna basically see the the long tail, right? You wanna see information that you don't otherwise know. So you can basically influence the quality of your prediction. So it really depends on That long tail behavior. If you turn it on and we we see you know the data emitted from your game environment, we're like, yeah, we already know this stuff, then the the good news is like the the improvement is probably quite immediate, it's quite fast. But at the same time, because there's no that long tail power, we probably won't basically materially change the performance performance right away. It will be more modest. But if you produce information, long tail information that we actually have not seen before. I think the improvement is gonna be material. But at the same time, typically machine learning models and AI models are trained so that when you see long tail, don't get overly confident until you actually see enough of that distribution data for you to be confident to basically go all in. so that's why it depends.
Josh Chandley: be comfortable exploiting it takes time. If we if I enable engine data, is that gonna help just my properties or is it gonna help the whole ecosystem?
Felix The: It's it's both but not in equal measure because
Josh Chandley: Okay.
Felix The: it's it because it will primarily help your properties, more so than the the ecosystem overall. Because I I think again in in AI models, assuming every single game is wild card game is is a it's a false it's a it's an incorrect design, right? So that means if the data is being generated by wildcard games That data is mostly relevant for wild card games. From what how we understand how your g game mechanics work, how it will benefit you if certain things are done in a certain way, the players that might enjoy your game, all those. So the answer is both, but the impact is is not proportionate. It should disproportionately favor your environment because it is based on you.
Josh Chandley: That is such an interesting response because one of the biggest concerns I hear is that if I share my my data, my players' behavior will help other advertisers compete with me for the very same players. and it'll help Unity target my players with competition. How do you I think I I know, but how do you think about that?
Felix The: I I think again, like my my understanding is Data and information has a disproportionate value based on who provide that information. Like for example, if there's a machine learning software that tries to predict you and I, if you basically say, Nope, you can't know anything about me, but I said yes, you can know stuff about me, I can tell you they can predict me better than you. so will will me volunteering my information make that machine learning model better predict you? Probably. well one. w we're both humans, to start. So we probably have human behaviors that they can predict quite well. Maybe they will then realize like, that guy is also in the video game industry. Maybe they they have some similar passion for video games. I'm making this up, but you get my point. Like it will always be more specific to me. but you will benefit from the fact that we have common traits and attributes, but it's very glossy.
Josh Chandley: And so that's also why you don't t necessarily have a free rider problem. There's no one that can let everybody else opt in and reap the rewards without sharing their data. They end up just benefiting the least. That's your argument, right?
Felix The: I y yes. My argument is like to your point, I think in this type of data situation, right? the idea of an asymmetry on on data economy exchange is is something that I think a lot of people think about. the good news is in in AI and machine learning, you know, just the way naturally the product's designed, people that basically provide information are the people that will benefit the most from it.
Josh Chandley: It's similar to MP post box.
Felix The: Yeah. And by the way, when I say that, remember I make that statements in generalities. So that means like let's say there's one developer that basically share information and we find based on that sharing of information is like, you know, the the players playing their games are not legitimately staying players or whatever else, then that benefit statement does not materialize. it will be beneficial for the industry though, when when that happens. So
Josh Chandley: In that niche corner case. I see. I see.
Felix The: Yes, yes.
Josh Chandley: I'm also interested in D to C. That's becoming incredibly important. as purchases move more and more to web shops, you know, those transactions stop flowing through all of these pipes in the exact same way. what I wanna understand is Unity launched up a new product commerce. Do
Felix The: Mm-hmm.
Josh Chandley: you view that as an entirely new data source for Vector? Or is it a way of making sure that you don't lose purchase signal that you already had? Like how do you how do you visualize that opportunity?
Felix The: our philosophy when it comes to launching a product is we launch the product for the sake of that product itself. So that means on the commerce solution, we genuinely remember, like I understand the nature of your question because, like, you know, vector, all that, right? But we're not an ad network in the largest, grand scheme of things. We are Unity. We have the game engine. We have relationships with developers that allow us to basically extend our services beyond just vector. a lot of people make their games with unity and they want a commerce solution. They want a web shop. So that is the product. The product is as you make your games with Unity, we believe we have an opportunity to provide you you and developers with a commerce solution. So we want that commerce solutions to be robust, strong, all that stuff. Now, back to the vector side. commerce solutions, yes and no because like Some people have web shops today, you know? and that
Josh Chandley: A lot.
Felix The: web shop yeah a lot. And some of that information about purchase that is being made on those web shops is already reported to the MMPs. So in that sense, like the data's already there, so it's nothing new. but are there web shops and are there transactions that we have seen that does not materialize elsewhere? Yes. Absolutely yes. So I guess it's a little bit of both.
Josh Chandley: So it is incremental to the data set in terms of seeing new purchases in theory.
Felix The: incremental some are incremental, some are probably higher quality if we were to process it differently. And others could just be, you know, what you said. Like we don't know how the industry's gonna go when we can predict. But risk insurance against a future data going to a different channel is always a good risk insurance policy to have. So it's all of them.
Josh Chandley: So bringing this kind of all together. I think Bromberg said the the vast majority of game advert advertisers already know Unity and already spend with you in some capacity. So
Felix The: Mm-hmm.
Josh Chandley: if the next billion dollars in in vectors scale isn't just new logos, what is it?
Felix The: It I I I think it would be also new logos because again like new logos because again like as Unity like peop people use us for game development for game engine. So we do like the idea of new logos because we believe like one of the ways for people to create an impact for the video game industry is to be able to express their creative vision and bring it to life. and as part of us basically being able to participate in that journey With them from the very beginning. we do like the ideas in new logos. I will say that. but on on the other side of the spectrum, beyond new logos, conversion rate in the industry is in a single digit percent, in a low single digit percent, Josh, as you know. So so growth is not predicated on just new logos. Growth is predicated on the fact that you predict correctly. if the conversion rate today, call it 1%, for every hundred, ninety-nine are missed. That's growth. That is a pretty big sizable opportunity for growth. and then materially for the developers that we have today, not just conversion, but prediction of like LTV, you mentioned, right? A longer prediction cohort. That's also an expression of growth. So an expression of growth for me is both new logos, but also in the existing customers that we have, we have so much potential. To generate much more value given the fact that the baseline that we have today, as we celebrate everything that we've done, we got a lot more room.
Josh Chandley: To to visualize that, like with your example, it it does almost lower the bar, but it is incredibly hard the task ahead. You know, if one in a hundred install and and two in a hundred purchase, that means really only one in five thousand actually even do the first target event of an IAP ROAS campaign. To to some extent You know, I actually as a side note, I I hear developers complain about you know CPI rising a lot, but
Felix The: Mm-hmm.
Josh Chandley: if You know, of the five thousand impressions you buy, one causes a purchase. If you purchase two and a half thousand impressions fewer that were doing nothing for you, your revenue creation is still the same. And so the effective
Felix The: Yeah.
Josh Chandley: CPI should double, right? And so that's what you're kind of describing as the opportunity.
Felix The: I I am a big believer of not wasting transactions. I agree with you completely. we track internally about the that area that you mentioned about like, yes, we're proud about the modeling improvements and everything that we've done, but we are very sensitive to make sure that the money spent is well worth it. So that means any prediction that is inaccurate is a transaction that adds no value, and we are here to combat that.
Josh Chandley: Yeah. and even just going down funnel, buying a hundred installs and having five purchasers that spend a hundred dollars versus having twenty installs and having five purchasers spend a hundred dollars. some of the metrics that publishers want to have go down, like CPI, actually lose their correlation with value creation in that way.
Felix The: Yeah. which is which is why I believe like we're you know we're we're one of the earlier, you know, proponents of ROAS as a concept.
Josh Chandley: Mm-hmm.
Felix The: I do believe like the industry, I never have again like affinity about just this or that, right? Metrics need to be rational and needs to make sense when it's tied to your success and the developer's success. When the product expresses that and that is the goal that it seeks, you And the industry will be better for it. So I agree with you. We don't tie CPI, we tie ROS.
Josh Chandley: Which is it's just the way to do it. the one thing on the other side of that though is outside of efficiency gains, there is a competitive aspect where you don't own the inventory. and when my when my business scales an in app ad network and it's successful, that budget very usually comes partially at the expense of other in app ad networks. Like absolutely it's true that there's a value creation component.
Felix The: Mm-hmm.
Josh Chandley: but if I'm a solitaire app, you know. E everyone knows I want to buy solitaire app traffic. There
Felix The: Mm-hmm.
Josh Chandley: is some level of direct competition. I assume that's true for your customers, too right? And how much how much do you think of your growth is going to be value creation versus share capture?
Felix The: Meaning like, how much of the new customers we're getting that is coming from share capture versus new?
Josh Chandley: Share capture versus you found a new user for this app that nobody else knew about.
Felix The: I d I'm not sure if I'm positioned to answer that question because I don't have access to my competitors information.
Josh Chandley: That's true.
Felix The: all I can say is like we focus on delivering the best ROS and the best quality products for our customer and our customer will then decide what type of actions they want to take. And what we've seen so far, like for example with the t day twenty launch, as they're satisfied with the ROS that we deliver for them at scale, they basically just say, I want more of this. Whether that action impacts our competitor or whether that's just truly incrementally new budget, I am not privy to that information. All I know is I receive, you know, more spend.
Josh Chandley: You just yeah, you just had you have more inventory flowing. Or more auctions flowing.
Felix The: Yeah. But I w right. But I would like to think like generally speaking for this industry, because I really truly believe it's not a zero sum game, the industry will benefit by the many network providers to basically better at their product and then you, in terms of acquiring players, you can basically have more choices to basically spend more dollars in return for a greater return. So as long as that greater return is true, you technically would always increase your appetite to basically acquire more players because it will guarantee you return. That is one way to look at it. But again, like that's just again us talking about generally, like mathematically in the industry. But from my side, I am not privy to you know how how you manage your businesses with the other networks. I can only speak to mine.
Josh Chandley: And so that that kind of covers, you know, gaming demand flowing into in app at Where else could the next billion come from? Like what's more exciting to you? Is it new demand that isn't gaming, or is it the same gaming demand flowing into inventory that isn't an app?
Felix The: Both could be exciting. I would say on the on the the modeling side, when you already have a model that is sophisticated enough to predict multiple things, right? Like day twenty eight, day seven, retention, purchase, this, that. the idea of prediction in a separate category is not an insurmountable task. In terms of your other question about channel, like again, like I think I think you're you you you probably know like generally speaking, we don't never we never view about channel as a as a identity for a product. Like, yes, Factor is in mobile in app today, but we never actually design Factor so that this is definitely a hundred percent mobile in app product. The idea that we have about Vector is is player acquisition product. We want to find players where they're at.
Josh Chandley: So W what I just heard, if I'm parsing this correctly, is like the the the general like principle here is that good prediction shouldn't be limited by demand type or inventory type. was it the plan for Vector all along to be as extensible as possible in the way we're talking about?
Felix The: I think there's two ways to answer that question. You don't make a plan, assuming your plan is perfect, and you don't change it as new information is available to you. But if you don't have a conviction about where the puck is going and what's possible from the very beginning, then you might make the wrong thing. So factor the way we design it is meant to be extensible. It is meant to be efficient. The platform and the underneath the chassis that support it is meant for it to be able to grow and expand. but whether we from day one had that expansive idea from the very beginning, I would say that would be more like a hypothetical ambition rather than a plan. But the plan all along has always been how do we transform our AI personalization product so we can achieve The success that the developers deserve that is not just today, but for years to come.
Josh Chandley: So It's interesting to me because that what the way that makes me think is it makes me think like so in 2024 the the Unity narrative was as I say pessimistic. some people would say it was a dying company, which is ex you know, a strong word. But actually you were you were planning a strategy to scale not just gaming and app, but to create something as a that could potentially expand the TAM quite dramatically.
Felix The: You you let me I I I hear your question. Let me put a final point to it. when you plan, especially in twenty twenty four, if we rewind back the time, when you actually are thinking about something more crucial than just mere expansion,
Josh Chandley: Mm-hmm.
Felix The: it would be unwise to basically think about the grand ambition to stand in the way of doing what's right at the moment of time, to be good at the the core value that you're offering. But it would also be equally foolish if you basically designed it so bespoke That it basically limits your ability for what will happen next should you be successful. There's a fine balance on how do you do it, but in this case, it's not a trade-off. They kind of converge. A good engineering is good engineering. a good engineering means when you basically design a system, you should not design it bespoke. You should design it to be really good at what it does now and it's capable to be good for the new stuff.
Josh Chandley: I see. Does a model does a model trained on game behavior predict anything useful about someone who isn't playing a game?
Felix The: it depends on the the the prediction head, right? So the prediction
Josh Chandley: Mm-hmm.
Felix The: head that is like say if you're trying to predict level complete. Yeah, absolutely not, because there's there's nothing in common. But if the prediction head is like your willingness to purchase something, you buy stuff in general. So that would be relevant. So it depends on the prediction head. But in any personalization, you don't predict things individually, right? Like you you you combine the prediction head to basically say this is the quality of my outcome. This is what I recommend. So the answer is yes for some, no for others.
Josh Chandley: Okay, let's start to land this. There's a there's a decision maker listening, you know, right now and he's he or she is setting budgets for the rest of Q four. Make the case. Like why should we why should we test now rather than waiting a few more quarters?
Felix The: tests day twenty eight or vector in vector in general
Josh Chandley: vector in general, or day twenty eight, yeah. Vector in general.
Felix The: vector in general I I hope we don't have to convince people to to test we are we are kind of beyond the testing phase. people have been using it at scale pretty much if not all of the traffic on our network is basically powered by vector. so if you're new to vector Give it a shout, give it a try. it's been proven. If it's a newer subsection of vector product, which is day twenty eight, mm my advice would be we have seen tremendous success of people that adopted day twenty eight. So I would highly recommend everybody to try it. But I would also advise this: don't try by turning off a product that already works for you and then trying to hop this on. You should always make a rational choice. And a very permissible choice for you so you don't impose yourself a risk of s something. keep the products that work, try new things, see the metrics, see the ROAS. When you're happy, you allocate budget proportionally based on the metrics that you see are leaning more in your favor.
Josh Chandley: So my final question about runtime and developer data, so we can kind of understand this is six months from now, what is operators if they're if they share runtime and they are spending into vector and investing into that product, in six months, what's one thing that an operator can observe in their data that would tell them if this thesis is playing out the way you expect?
Felix The: You would not probably see it in the data explicitly, per se. You would see it based on your performance is the easiest way to to to answer it. But you would not technically be able to say like all right, I see plus X percent. A portion of that X is runtime, a Y portion of it is this, Z is that. It's just a compound effect. Because for us, we never actually parse that information for developers to to justify like this is the benefit from X, Y, and Z. Our loyalty is we want to b give you the best outcome, period. So whether that outcome comes from this or that, we're working our hardest to make sure that they are all materially going to benefit you in a meaningful way.
Josh Chandley: Unity has come a long way with with Vector. what are you most confident about in regards to its its long-term chance of success? What do you think it's gonna change for publishers?
Felix The: Our industry at large has room to to materially improve itself. there are so many ways. if you want to talk about the quality of ads being shown, the the nature of the at unit, how how user-friendly is it for your player so they will be retained in your game. whether it's the idea of Understanding players' preferences so we can actually predict things better, like the conversion rate in the single digit percent that I mentioned. I believe the future of this industry is bright because there's so many i inefficiencies that can be squashed, there are so many opportunities for us to improve, whether it's the actual behavior of the player and the conversion manifested by the ad unit they see and the implication to LTV
Josh Chandley: Mm-hmm.
Felix The: in the game, or the prediction quality, all of these are nowhere close to being enough. So I believe by having providers such as us to focus a lot, if not the only soul focus we have, at the heart of the matter, which is providing quality prediction, quality user acquisition for our developers, and also on the publisher side, to make sure that the experience that the players see is of quality, the industry will be better overall for it.
Josh Chandley: Well, Felix, it it's been a pleasure. Thank you for joining me on Mobile Signal.
Felix The: Thank you, Josh.
Josh Chandley: For the audience, if you made it this far, thanks for listening. few more thoughts after talking with Felix. First, D twenty-eight is real. My expectation is that if you're outperforming by some percent on day twenty eight, you're outperforming by more at day three sixty five. that's a claim though that you can test on your own portfolio. So run it alongside day seven and measure maturation. Bigger question though is developer data, the runtime signal. My read is vector is still early in using it. Their growth so far has been from solving other other problems. So runtime and executing against that is the next major test for Unity, where if they can succeed, that will unlock a new level of growth. I think for publishers The best outcome is just more networks competing on prediction quality. and I think that's coming. That's good for everybody. I would say in twelve months, so Oc October twenty twenty-seven, at least at least three more networks go GA with day twenty-eight optimization. I'll come back and score that call. So subscribe if you want to see how it lands.
