#267: “AI in Action: Real-World Use Cases” Business of Apps London 2026 panel

Speaker A: Hello, welcome to the Business of Ours podcast. You know it better than us. The era of experimenting with AI is over. The question isn't whether to use it anymore, it is whether it's actually moving the numbers. In 2026, the winners aren't the teams running clever pilots. They're the ones who've wired AI into the core of their growth engine, into how they find users, activate them and figure out what those users really want. In this special episode, we're syndicating the business of Apps London 2026 panel AI in action. Real world use cases. It is an all hyped look at what's working right now, told through the examples from three people scaling it in practice. Steve Young of uh, Appmasters, Anton Volovic of Ripface, and Nathan Hudson of Perceptics. Alright, let's get into it.

Speaker B: All right, we're going to talk about your two favorite letters today. I know what James is thinking, qt, because I'm up here, but it's not qt. It is obviously a, you guys finish it, Come on. A. Thank you. Yo, you gotta get ready for happy hour, right? All right, let me introduce the panel. The one thing I don't like about panels is the moderator always says, can you introduce yourself? I'm like, screw that, I'm going to introduce them for you. The other thing I'm going to show you guys to get your phones ready. If you got questions, put them up because we're here to answer your questions, not to kind of predict what you want to hear. All right, so without further ado, let me introduce the first guest. He runs Perceptics. He works with 6, 7, 8 figure apps to try to scale them and he's been an app marketer of the year. Without further ado, put your hands together for Nathan Hudson. Good to be here. All right, next, hopefully these applause will go up in nature and excitement. All right, he turned a one feature viral app into 300 million downloads. The app is called Reface. And now, most recently, he's found success with a new app called bypal. It hit number 3 nutrition app, top 10 fitness app globally by downloads. This guy knows his ish when it comes to app growth. Let's give it up for Anton. Come on. Woo. All right, we're going to see AI in action and we're going to show you some creatives that these guys have found success with. And yes, spoiler alert, it is literally all AI The. Did you guys get the picture of the slido? Yeah. Ask questions. I'm going to monitor along the way. Thank you. This is cool. All right. What AI solution brought the most impact on business results?

Speaker C: You first coordinated M. Good question.

Speaker D: I think over the last year it's definitely been like Claude, um, just generally. Right, like using Claude, um, for all kinds of different things across the board. I think I would talk about it more from like the development perspective and like how fast now we can ship experiments in as much as like even the growth teams, um, being able to actually launch experiments, launch onboarding experiments, launch payroll experiments and not have to necessarily like rely on the dev team who are maybe building out core features. Um, this I think is one thing that's quite cool.

Speaker C: Mhm.

Speaker B: Yeah.

Speaker C: For us, I mean I would love to say that it's like a specific marketing tool for the app, uh, or your consumer business that you can just take off the shelf. But it would be also something very horizontal. So it would be an A10 actually. And uh, the reason is that we are building inside reface, this like machine, the factory of uh, the whole marketing process starting from creatives finishing with user acquisition, monetization and a lot of automation and things. The different models that we bring together, they are all connected with NA10 which we eventually upgrade to something which is more robust and durable. But uh, if I were to say one thing, it would be NA10.

Speaker B: Oh, what did you say?

Speaker C: NA10.

Speaker B: NA10. Yeah, I like that one. Okay, let me see if these clickers are working. All right Anton, we're going to hit you with this video. Yeah, so let's see, we got examples, ladies and gentlemen, of high performing ads. Here we go.

Speaker C: And I'm feeling good. I feeling good.

Speaker B: Break it down for us frame by frame.

Speaker C: Yeah, so this is basically the app of um, Sorry, this is the ad of our nutrition app, Bite Pal. That's the one which is now reaching kind of top numbers in the nutrition and health and fitness space. So this app was uh, a very high performing ad, I would say in February. Uh, so it lasted probably more than a month, which is relatively long for the current environment and also the marketing spend that we do. And uh, I think this ad is interesting because we actually took the real influencer, like the person who was basically ready to give the footage. And then we iterated around the background, the lengths, the music, obviously the pack shot. And uh, basically you probably can see like hundreds of different variants of this ad where you have this one anchor element which was actually the person who gave UGC piece.

Speaker B: So this background is not real, right?

Speaker C: Yeah, the background is not real. So the real is just the lady. And then, uh, obviously we changed the type of the food. I mean, the captions are pretty easy ones, but also the entire background is, uh, completely changeable.

Speaker A: So. Yeah.

Speaker B: Do you guys have it? You know, I always see these LinkedIn posts. Oh, I slept and I had 108 new creatives. Is that something that you guys are doing too?

Speaker D: Yeah.

Speaker C: So I think if you benchmark us versus other relatively big publishers, I think it would still be probably a bit lower on the number of creatives we produce every month, every week. But I think what we do particularly well is the creative diversity, which probably also one of the top topics of today's discussion. Um, and we have the internal tool which we've built, uh, with the help of our internal team and, uh, a bunch of other companies as well, where a designer can actually construct the creative from different pieces. So you have a lot of AI models which are plugged in, into the system and then it's basically drag and drop. You type the prompt. We also have there basically the library of assets also live, um, in this specific software. Uh, and you can basically pick up the model, whether it's an AI model or a UGC video of someone else, and then basically create, um, in front of your laptop, everything that you see

Speaker B: as someone in your intermittent fast like this is this speaks true to me.

Speaker D: Yeah, yeah.

Speaker B: Like one minute before, I'll be like, seconds and like 15 minutes. I'm like, babe, I can hold out, I can hold out.

Speaker C: Exactly. So, yeah, it speaks really well to the audience who actually fast. Yeah, I don't, but I guess it speaks well. So therefore it was a relatively good performing ad, uh, in February.

Speaker D: All right.

Speaker B: VK says, when is AI not the best solution?

Speaker D: I agree. Yes. When is AI not the best solution? No, in all seriousness, I think it's hard because it's becoming so that it is always the best solution. Right. Like in a lot of cases now, I think as we see as we go, I genuinely think that a lot can be done with AI. I think one thing that is not AI able at the moment is that kind of human aspect and that human feel to really come up with something that is like truly innovative, um, repeatedly on demand. I think ideation is still very much humane for now at least. And people talk about human taste and stuff like this. So I think it's a case of humans and AI as opposed to just AI and ubi. Um, you didn't get that one universal basic income like us not having any jobs. Um, I think for now that's the case, but we'll have to see.

Speaker B: Anton, you want to add anything?

Speaker C: Yeah, I mean, um, uh, AI is super powerful, right? And then probably like in two, three months, we would have even the different conversation. But the way I think about it is kind of it's a funnel, right? And, um, at the bottom of the funnel you have like a human need, then you have a product, then you have, let's say, different funnels and then the creatives. Right. And I think what's actually pretty hard for AI to do is to kind of find the core concept, which would be kind of the main building block of everything else that you're trying to build. So, for example, with bytepal, kind of the whole idea about the raccoon is a little bit like a Tamagotchi. You care about it. Um, it's a great idea, which, because it's the core of the product, you actually cannot iterate it as fast as AI like enables it to do. But what you can do is with the help of AI frame everything else around. So I think kind of the core idea, ah, the product use case, I think it still sits with the person, with humans with us as of now. But let's see how it changes.

Speaker B: How much of a difference did changing the background actually make, though?

Speaker C: I mean, it's actually interesting, but, uh, it can be quite a lot. Uh, and also, what did you see?

Speaker B: If you're open to sharing?

Speaker C: Yeah, I mean, maybe the difference can be 20, 30%.

Speaker B: Wow.

Speaker C: Uh, but the thing is, it also, like, if you make variations of the same creative which performs well, you're able to tap into a little bit bigger total addressable markets and therefore iterating on something that works. And changing backgrounds, changing the person, the meal itself would allow you actually to scale what works a little bit bigger to, like, different pockets.

Speaker B: Did you guys change the meal?

Speaker C: We changed meals, we changed the outfit, we changed the restaurant to the dining room. So it's just like it's a constructor, basically.

Speaker B: It'll be interesting to see the original then. That would be cool.

Speaker C: Yeah. I don't even know whether it's original or not, actually.

Speaker B: All right, let's hit the next one. We'll go to the next one. This one. I love this one. All right, I'm going to hit play.

Speaker C: Say love me information now.

Speaker B: All right, Besides us probably doing this at happy hour, that dance break us. Break it down for us. Yeah.

Speaker C: I mean, here it's, uh, it's a bit more obvious. The raccoons are not real. The song is Also a generated. So it's both the music and the, the lyrics as well. Uh, yeah. And again the elements that we play here with. But I think this creative is actually interesting because you can clearly see that it's AI but uh, it still resonated with people. People still clicked. Therefore I feel like when we talk about the creative work and the ability to storytell through creatives through our very short attention spans, I think both actually work relatively well. Like the bet on authenticity where you have actually a real person but also this a uh, bit like fake and real ads as well.

Speaker B: What did you use to make that song? Because I literally thought that was a real song.

Speaker D: O.

Speaker C: Maybe sooner but I'm not sure actually.

Speaker B: Okay. Yeah. I love. You know.

Speaker D: Yeah.

Speaker B: What kind of prompt did you tell it? I mean did you know that hey, it's like I've got dancing raccoons. Are these raccoons or seals? Are these seals?

Speaker C: They are raccoons. They should be raccoons actually.

Speaker B: That's crazy.

Speaker C: They should be raccoons.

Speaker B: Should we listen to it one more time? Cuz I, I don't even know that it was a. Ah, They love me because I buy Chanel. What is he saying? What's the words? What's the lyrics?

Speaker C: I didn't pay attention but maybe. Maybe you saw there is a back. I mean but no copyright issues. Let's uh, not talk about it probably.

Speaker B: I love this question from Vlad Vladimir is like what are the biggest security and trust risk when integrating AI into mobile apps and how can they be mitigated without degrading user experience?

Speaker C: I mean when you install any of our apps, uh, which is like Reface, uh, our kind of app that we started with bytepal. We have another one, it's called Honestly, it's like AI journaling, AI therapy. I mean you will be hit by multiple pop ups saying that hey, actually we use Gemini and Gemini may send some third party data to Google back and stuff like that. So we're actually pretty transparent within the product itself, uh, how we handle AI. And to be honest, uh, I mean we're very kind of scientific companies, probably everyone in the room. So we a b test everything and we didn't see actually a big drop off of people if they're like facing straight away that it's AI so it's kind of expected.

Speaker A: Yeah.

Speaker C: And I feel like people already started developing this. A little bit of like GDPR blindness. I would say that okay, there is something to read, something to accept and people just, they're okay using iTools I would say.

Speaker B: All right, Anton, you got a friend in the audience I think. Okay, before we move on you, Nathan, what is the one thing that you think made bytepal successful? I mean there are hundreds of apps for this.

Speaker C: M. I mean one thing, uh, looks

Speaker B: like a webflow as well. Take the test.

Speaker C: Yeah, I mean a lot of things of bytepal are pretty standard, I would say. Um, but M. You know, there's like this really nice book actually. It's called Seven Powers. It's the way how kind of companies build, uh, sustainable competitive advantage. And I think in our industry, I have to say that kind of operational efficiency is one that uh, actually super important if you manage to be better by like 2% in every single moment. We have a little bit better interface which kind of play a little bit better on TikTok ads for example. Right. And uh, we have maybe a little bit like better creative performance. So these things pile up and it actually makes you get your extra margin which allows you to scale 10, 100x. Otherwise you wouldn't be able to do.

Speaker B: Okay, I like it. All right, sorry Nathan. Bear with me bro. All right, Andrew says if you were to start bypal today from scratch right now and have limited resources, how would you grow it and what would you focus on? And did you, Anton, did you just start off with running ads right away just to see if you had product market fit?

Speaker C: Yeah, yeah, I mean we had. So we had an idea. So like what we notice right now is a lot of people just like, see uh, the industry, it's like relatively big. Then sensor towers, similar web, like other analytical tools. You see it's growing and uh, people try to launch something pretty generic and then they try to find kind of first product market fit in with the ads and then they start building the product. The problem is that like a lot of people never actually reach this state. Right. And maybe one of the ways to think about it is if you actually like the idea, kind of invest in it even before ads work. Because I mean bypal kind of existed for a year before reaching positive uh, unit economics. And uh, we just really like the concept. Uh, it's also an interesting, complex, interesting uh, concept back then. And uh, we kind of invested already so much in the product that we didn't have a chance to kind of step back and do something else because it's a little bit of a cost sunk fallacy. But it helped us to push through until it actually started scaling.

Speaker B: Were you running ads the whole time?

Speaker C: That whole year we were running ads, the whole Time. And I mean, we're in a relatively good position because we had other apps which were cash. Generative.

Speaker B: Yeah.

Speaker C: Uh, so by pal, for quite a while was a heavy roi. Negative product, I would say.

Speaker B: Okay, thanks for sharing that, man. Yeah. All right, we've got Nathan's ad as well. I'm going to give. I'm causing headache over there for the. All right, here's this video. I love this. And this is right, Nate, all AI.

Speaker D: Yeah, this video is 100% AI. Everything you're seeing in this video is.

Speaker B: And I love these comment thingies.

Speaker D: The comment thingy. Uh, the comment thingy is a legit comment.

Speaker B: Yeah.

Speaker D: Um, but you can tell by some of the text, right, that it's like, is that a username? No.

Speaker B: All right, I will hit the video right now.

Speaker D: All, uh, right.

Speaker B: What did you test? You test background? You test outfits? What did you guys. What did you guys test with this one?

Speaker D: Yeah. So for this creative. So basically, like, backstory, like, for the last 18 months, I've been working on what I call my secret project, right? And this secret project has been super secret project. Um, but essentially, like, what I've been working on is trying to build a super agent for mobile growth. Right? Like, in short. And one of the things I've been focusing on has been creative production. So the idea is how can we fully use AI to create the creative entirely? So taking it beyond using a real, uh, UGC creator and then changing different aspects and just fully prompting to generate a creative. So this was actually one shotted, right? So this was the first, like, version. Um, but there's like 15 versions I was trying to select to figure out which one to show. And I selected this one because one, it's the most authentic. Like, yeah, you can see the little comment bit that looks a little bit like, oh, that's not real. But realistically, like, if you're seeing an ad pop up on your phone, no one's looking at that. Um, and yeah, kind of. That's why we picked this one. The interesting thing is when doing this, like, it took me maybe 15 minutes to create the 20 different ads, and the majority of that time was waiting for the ad to be generated. Right. It's taken like over 18 months to actually build the thing to build the ads. So that's not been easy. Um, but what I notice across those 20 is that some of them suck, right? Like, it's exactly the same. The prompt is the same. It's the same. Why? Why is one of them. She's like throwing the clothes. And you can see that the gym set is like the, the top and the leggings are connected as if it's like a kid's dungaree set. And I'm like, well, that's what. What are you doing? Right. Like, different creators. We see different things in some, some of them, like she's holding the two shoes. Then, like, she throws up and a shoe lands on her head. And I'm like, what? I didn't say that. I'm confused. And I think what I've come to learn is that the models are not consistent yet. Um, but you can still generate creative end to end.

Speaker B: Do you have a prompt that you wanted to share or you want to wait till the end?

Speaker D: I don't for this one. I have a prompt for one of them. Um, the final video, I have the prompt I want to share with you.

Speaker B: That's a little teaser. Then I'm going to go to the next question. Tell me about, um, the tools that you're using. Are you still using Claude to create those videos?

Speaker D: No. So, like, this is. That's my secret project, right? Like, so, like, I've been working, I've been working on building and I'm not trying to sell anything, by the way. There's nothing to sell here because it's not ready. Otherwise I would tell you that it's ready and I'd try and make some money. Um, but I've just been focusing on trying to actually deliver something that can do a good job, and that's what I've been doing. So, yeah, that was custom tool, just like Anton mentioned. Like, you guys have internal tools? Um, yeah.

Speaker B: Yeah. All right. Nathan, are you worried about getting too comfortable with AI in the growth space and cost ballooning in the future once the AI models look to make more profit? This is from Alex.

Speaker D: Yeah, great question. So I think you're getting at the idea of, like, we're using models to generate stuff and then let's say they hike their API prices up and then all of a sudden things are, uh, crazy. I think specifically for generation and creation. Not particularly. I think if we look at what's happened with the, um, models for coding, for example. Yes, anthropic are leading the way. Majority of developers would argue with Opus 4. 7. Um, but OpenAI are very close with GPT. I don't know, 5, 4 now. Um, Open source models are really crushing it. So I don't know if anyone here is a developer. Maybe not, um, maybe, but like Minimax 2.7. Really solid point being I think open source versions of these models will become available. Um, and I don't think that that is something to worry about. At some point, it's like it either looks real or it doesn't. And when I can do it. Okay, great. Now, when I say it doesn't land on her head, but I don't mind waiting a little bit longer to get one. That's right.

Speaker B: Anton, what were you using to change all the backgrounds and the food and the outfits?

Speaker C: Oh, actually, that's a good question. Uh, so before this panel, I chatted with the person who leads the stream at Reface, and she showed me. So the tool is called Pavuk in Ukrainian. It's a spider. And then you have different models, legs, uh, of the spider. It's like cling AI. And then you have 11 labs. You have SUNO there. You have, uh, nano banana, obviously. So for different pieces, there are just like, different things.

Speaker B: Yeah.

Speaker C: And you just mix them together. Yeah.

Speaker B: How many music things, music tracks were you testing?

Speaker C: Oh, I think. I think quite a bit.

Speaker D: Yeah.

Speaker C: Uh, actually, sound and music is an important piece of a creative.

Speaker B: I mean, that's what a nose is in.

Speaker A: Yeah.

Speaker C: It drives performance quite a bit, actually.

Speaker B: Yeah. That's crazy. That's crazy. Even with the little mute thing off. Yeah. Yeah. All right. What's the biggest mistake you guys have made when it comes with AI creatives? That's from Rafael. Thank you, Raphael. Appreciate you. Sorry. What's the biggest mistake you've made?

Speaker D: I would say, um, underestimating the potential of AI 100%. Like, I vividly remember sitting down with the team, like, look, like, yeah, AI is not coming for creative. There's no way AI is going to be able to do creative. And then VO3 came out and I was like, look, it's like, it's trash. It's like sometimes good. And then Sora 2 came out and I'm like, so, um, I think we should be worried. And now. So I think the biggest mistake is being set in the old ways of doing things and being like, no, but this is our process. Like, this is our process. We do it this way and that's the way we do it. And that's how we've got results. Great. But your competitor doesn't care. It's a question of can we adapt faster and can we adopt faster and can we win and make better creative or any aspect, um, when you're leveraging AI.

Speaker B: Yeah, yeah, yeah.

Speaker C: For us, um, I think so. Basically, Reeface was born back in 2019, and we became relatively big in 2020 when we deployed our own uh, face swap model. So basically it like a one click, uh, three second experience of putting your face onto any video. Some people call it deep fake, but ours was not as good. So it's impossible to deceive people. It just has this kind of comical attitude. Right. And then the industry started changing quite a bit. So obviously AI became like a big thing in 2022, 2023 and we were a little bit struggling with our ego because back in 2020, 2021, we were one of the few companies in the world who were able actually to develop their own um, AI models. And the original face swap that you can see in Reface hub still, which is like a tiny feature now, but back then it was like a very big thing and not a lot of people could repeat it. And then the industry changed so much and we kept believing that, okay, we actually need to sit also in the model level and we need to have our own ML research, build models and stuff like that. And it took us some time actually to realize that, okay, I mean if you want to build models, just build models but don't do other things. And therefore we were choosing okay in this kind of value chain of consumer, uh, Internet where we wanted to play. And it took us a little bit more time to realize that actually we are uh, a company which builds consumer products. It doesn't matter what kind of AI we use. It's much more about packaging, UI, UX and also the distribution machine that we built.

Speaker B: Yeah, 100%. How do the roles in reface marketing team change as a response to AI?

Speaker C: Uh, the role of the marketing team? Yeah, I mean look, to be honest, like the role of the marketing team is, I mean I would say number one for the companies like us. It used to be like this before and it's exactly the same now. And I think it actually intensifies because I believe that the cost of the production and the ability to copy things, uh, the barriers to entry is just like so low. Like anyone can build bypal kind of in a few days with a few developers, although five years ago it was much harder. So the question is where the competition happens. And I uh, think it happens on the distribution layer. Distribution layer is marketing. So I think the role is even bigger than before.

Speaker B: Did it shrink? Did it expand?

Speaker C: It expand and it became the uh, key competitive mode. Uh, like if you were to ask what's our competitive mode? I would say, I mean a lot of it is marketing.

Speaker B: Yeah. Did your, did your like interview Questions

Speaker C: change, like interview questions.

Speaker B: If you're going to hire somebody, it's like, you better know how to prompt properly.

Speaker C: I mean. Yes, but I think what worked for us really well is we try to take, uh, people who are just naturally curious and motivated. I think this is what worked the best for us. And we have, like, so many examples of a person who was doing completely different thing, joined Reface. We saw that, okay, there is some m energy to do things, and then these people learn very fast. Also, like, to be honest, like, we have experts, we have tools, we have resources. So now I think also, like, if we talked a little bit about the future of the apps, and I believe distribution is the core, I think, for a person, for, uh, whatever person who is searching for a job or starting something is, I think is like the core motivation. Like, what drives you. And are you willing to kind of every day work hard? That's the biggest competitive advantage, I would say.

Speaker B: I can't teach hustle. I can teach you skills. I can't teach Hustle.

Speaker C: Yeah, exactly. Exactly.

Speaker B: I'm gonna assume this is yes, but have you guys seen a significant reduction in cost by using AI, or is starting with UGC a better. A real UGC a better option?

Speaker D: Yes. So I would say, yes, we've seen a significant. Well, we're seeing increasingly costs decrease. However, I would argue that for maybe even five years, like three years, creative production costs haven't been the big issue, Right? Like, it's not a problem getting creative at, uh, scale. It's a problem finding winning ads, right? Like, however much you're spending on creative production, you're spending more on meta ads or TikTok ads, right? Like, you're paying Zuck, and no one has a problem lining his pockets ranting, but you know what I mean? Like, for example, if you're spending five times as much on creative production, and Instead of spending 1,000amonth, you spend 5,000amonth on creative production, but you're paying a million in ad spend. I struggle to see what the issue is here. Um, especially if you're getting significantly better results. So when AI models came out, and I think back in October last year, I was really playing with Sora too. And sometimes, yeah, we could get creatives cheaper. But were they winning? Mm, I'm not sure. And ultimately, we're here to find winning ads, right? We're not here to reduce our creative production cost, um, because that's where the uplift is. So I think, yes, costs are coming down, and it's great, like, now. Yeah. Like, this ad hasn't been ran. By the way, I made this ad this morning at 2:00am um, and if you don't believe me, you can ask Emily, because I was meant to submit the ad to yesterday.

Speaker B: Right.

Speaker C: I was in the thread, in the

Speaker D: email thread, and I was like, okay, I need to do this. Um, and that's because I was building. But yeah, I think it's kind of irrelevant. We just want winners.

Speaker B: Last question before I show the next video. Are you guys using AI to come up with the script and the ideas? Or is that the human aspect of it? Like, hey, I want somebody eating their meal and I want to show when they can actually eat and do that. Or did AI come up with that script and idea?

Speaker C: Uh, I mean, for us, it's a collaboration, actually. It's a collaboration. And we also see some people who, let's say, come to reface and then they try to rely completely on AI idea generation. Generation. It actually doesn't really work. Um, so AI can be like a multiplier, I think, of your own talent and things. But still, I, uh, think for you to kind of create a really meaning creative, you really need to understand how a person thinks. A person who does fasting, for example. So I'll just give you an example. So I'm, I love cycling, so I do a lot of cycling. I'm very interested in this topic and I can clearly see if the company who advertises to me kind of talks in very general generic things or actually they know like how I live, how I think and stuff like that. So there is definitely still a human element.

Speaker B: You see the same thing.

Speaker D: Yeah, I would agree. I would say so the video we just saw, that was a collaborative effort, right? That was me and my super agent. And that was the outcome, right? That was like collaboration. My concept there was like a try on haul. I want, like to combine a Zumba dance workout at home with a try on haul. Because I was like, I was in my zone. I was like, yeah, I've got an idea. Um, and then this was the outcome. Um, the video that we'll see at the end was 100% generated by AI, from everything from idea to script to storyboard to everything. Um, and I'll tell you what the prompt was that I gave it. Um, but yeah, it's a moving target.

Speaker B: It's not this one, is it?

Speaker D: Not this one. This one.

Speaker B: Can I come back to this one?

Speaker D: We can play the next one, is it? Oh, not, not this one, the other one. We haven't there was twos and whatnot. Yeah, this one's the one.

Speaker B: This is the one completely. AI.

Speaker D: No, no, this is. There were three. This is the second one that we should play now. This one. I didn't come up with the script.

Speaker B: Um, AI came up with this.

Speaker D: AI came up with the script. I can't remember what I came up with. Let's watch it and then we'll see.

Speaker B: All right.

Speaker D: Yeah.

Speaker B: Shall we? Is it ridiculous doing a Zumba workout at the same time as filming a try on haul? Probably, but I'm gonna do it anyway. No, who says try on hauls can't actually burn calories? Real talk.

Speaker D: If I can't Zumba in it, I'm not buying it.

Speaker B: Anyway, I've got a workout to do.

Speaker D: Now I remember. So I made the first one first and then I was like, this is cool, but I want a voiceover. Uh, right. And I was like, ah. I came up with. What did I come up with? I don't know. Maybe I came up with the hooks. I can't even remember. But we were sparring. We were sparring and then we came up with some script. I was like, yeah, I like the sound of this. Let's go with this. Um, and then it was really hard to get this right. So the other one, I said it was fast to make this one. I've got 20 versions and 19 suck. Like, really tricky. Um, I think it's because I was asking for quite a lot. There's quite a lot going on with the script, the storyboard, the cuts. Um, but, yeah, this one was, again, a mix.

Speaker B: Nice. All right.

Speaker D: And one thing I will say actually, that I want to touch on. I got my notes. That's why I've got this notepad. I'm funny because technically, I should not tell you that I've got my phone inside the notepad with notes. Because otherwise, like, just why are you sitting there with the notepad? Just use the phone. But I did tell you. Yeah. So none of these have been ran. Right. Um, and a question is, I'm looking at them thinking, do I think that they're going to be winners? Which would be an interesting thing. And I guess what I will do is I'll run them and then, I don't know, maybe I'll do, like a post on LinkedIn to say, like, which out of the two Zumba ads performed better? I, uh, have to ask the client. I made these this morning, so they don't. They haven't even seen them. I might get in trouble if I get in trouble. I'm blaming Babs. Anyway, um, but are they going to be winners? And I was sitting there thinking, and I was like, I don't know, maybe, right? Like, maybe there'll be winners, maybe they won't be winners. But I actually don't care because I'm not attached to the creative. And this is what really stood out to me. And it's the same thing I've noticed with, like, coding, right? You become less attached to your. Your code, but you also. I become less attached to the creative. I didn't spend hours and hours grafting on a single script and, like, going back and forth with the creating. No, no. Like, what are you doing? Why did you say the word that way? Why don't you say the word this way? It took me hardly any time. So whether or not it wins or doesn't win, I don't care. And I'm also not going to try and reverse engineer the win and be like, yeah, I think this one won. Because if you notice, because as I said, there's like 20 variations of each of these, and it's way more efficient for me to focus my time on coming up with really killer concepts and sparring and just letting matter do its thing than it is for me to get emotionally invested into a creative. And the irony is, I'm pretty certain that this time last year, I was saying the opposite thing. And I was a big advocate of really diving into the performance. Like, of course we want to know the hook rate. I still do. Of course I want to know the hold rate. I still do. But why? What is it? Is it the accent? Is it, like, the hook words? Is it the tonality? Is it the clothing she's wearing? Is it the backdrop? Is it the lighting? Is it the this? Now, uh, I don't care. Because across multiple clients and multiple apps, I see all these things. I think that's it. And then when I try it somewhere else, it doesn't work. Or it was the case. But now the trend's moved on, and what I thought was a winning ad is no longer a winning ad. So now I'm just like, let's just come up with ideas and make good creative and then hopefully we can grow.

Speaker B: Yeah. Anton, has AI surprised you in unexpected ways?

Speaker C: Uh, I mean, yeah, it kind of, um, probably keeps surprising.

Speaker D: Yeah.

Speaker B: Yeah. Have they grown in unexpected ways? And I'm assuming that audience, that she had a female, like, it was targeting male and female.

Speaker C: You, uh, mean our products?

Speaker B: Yeah. Well, no, that. That ad, it was just a broad

Speaker C: targeting Yeah, I think, I think it's broad targeting.

Speaker D: Yeah.

Speaker C: And meta decides and I mean, going back to your point, actually we also like at reface, sometimes try to predict which creatives would work better. I mean, we're never able to do it. And uh, sometimes, obviously we have this metric of cost per creative and then we have very cheap creatives, let's say the static creative, which is completely just like you click the button and been generated. Then you have a big storyteller where we actually try to find a specific person. If this person is real, then we put together like a long storyteller. And I mean these ads lose and then the generated one in one second wins. So it's actually, it's impossible to predict. You just. Yeah, throw everything there, I think as well.

Speaker D: Like, because I was having this discussion with someone, I think earlier in the week about like predicting winners and we were debating and my argument is, look, anyone who claims that they can predict a winner's line because otherwise there should be someone out there with like a lifetime 80% plus win rate just like dropping killer creatives or like you show them ten crazy like, look, mate, don't even waste your money on that. That's your winner. And they would be a billionaire. And I haven't met that person. And there's a reason why we aim for typically like what, 10 to 15% win rate is considered standard. So if we can't predict and we spend all of our time building creatives and working on creatives, we've got to really ask ourselves, why are we so obsessed with prediction? And I think you're right about it's the cost per creative. Like we get attached to. What's the cost of creative? Can I reduce that or can I reduce my cost per winner? But then you can take it even further and say, who cares about cost per winner? What about costs per, uh, creative that I can spend half a million on?

Speaker B: Right, like, all right, let's watch this last video and you will share the prompt.

Speaker D: I will share the prompt. In my head, I was always going to start running. Then I finally did and now I'm out here actually running. My first 10k runner got me here. Start your plan today. Are you not entertained? Is this not why you are here? Um, no. So the prompt, the prompt create me a killer ad creative for the runner app that I can show off at a demo today. They are not a client yet, but make them an ad that will make them want to be. That was my prompt. No, no, lie. That was my prompt.

Speaker C: Um, and Then no editing afterwards.

Speaker D: No editing afterwards.

Speaker B: So that was.

Speaker D: That was one shot. And there's about three of these that are like, really good. And I picked this one. Um, there was one that was a little bit. Arguably a little bit better. But I was like, no, but that one would have required tweaking and then I wouldn't have been able to say that. And I thought, what's the bigger wow factor? It's being able to say that was the prompt. Um, but yeah, I think with this creative, it's clearly AI Right. Um, but I think there are two types of AI Creatives. There's AI Creatives where we're trying to make it seem as though it's not AI, which I think there's an argument to be. We can discuss the ethics there. But I think there's an argument that those creatives perform well. But then there's creatives like the one you showed with the meerkat, like this one here, where it's like, well, clearly this is not completely human. It either is. And they've spent a crazy amount of production on this. But it's not like ugc. Even if someone was to recreate this with, I don't know how Hollywood do the movies and Iron man and Marvel, but yeah, I'm sure expensive production studios could recreate this. But do I care if I'm looking at an ad that was created by an expensive production company or AI?

Speaker C: No.

Speaker D: The goal behind this creative is just. It's kind of entertaining. Yeah. The script is not that strong. Right. Like, as a growth person, I'm looking at the script thinking, okay, yes, it's a decent story. Um, but, yeah, it took me.

Speaker B: Sorry, Nathan, I'm going to interject. Uh, Anton, I would like to get your thoughts on this. James says, do you think AI Creatives cheapen your brand identity?

Speaker C: Cheapen our brand identity? Uh, no. No, I don't think so, actually. Um, so at the beginning of our journey, uh, we try to, uh, kind of, um, follow the, um, kind of brand guidelines for all the creatives that we produce. And obviously it's a constraint. So it constrained the creative ability of, uh, our team. And the creatives were not as broad as possible. And then once we started just like, uh, creating everything without very little constraints, I would say the performance marketing started working. So the way how we think about brand is actually creative. No one actually remembers what you clicked on. What, uh, you remember is your interaction with the product. Uh, and therefore we kind of try to think about the brand and build brand down the funnel. Not, not uh, at the creative level.

Speaker A: Yeah.

Speaker C: And it's been working actually quite, quite fine. Yeah, bypal has a lot of organic traffic, actually a lot of word of mouth. It's a really good product. Um, and the ads are, I mean they are not.

Speaker B: I like that answer.

Speaker C: Quality.

Speaker B: Yeah, it's the product. When you have to search for an AI tool to be used during, during your day to day operations, what are the key aspects that you take into account when making that choice?

Speaker D: I don't know about anyone else, but I really struggle adopting, I did struggle adopting new AI tools in as much as like, for me it has to, there has to be like a reason to give this a go. Like if it's just like the same thing as something else a little bit differently, I kind of can't be bothered. So it's a tough one. I think we've probably all fallen, um, anxious around all of like the LinkedIn comment date. Like if there's enough comments, like, you know what I mean? Like uh, comment, comment skills. Oh yeah, comment, um, runner and then get the ad or whatever, you know what I mean? Like this kind of stuff that makes me think, oh no, I need to try that. Like a thousand people have commented and I'm like, oh my gosh, am I missing out on something? But I've come to realize that every time it's a disappointment. So now it's just like, is there a need? Do I have a problem with my current workflow or do I genuinely think that this might enhance it?

Speaker B: All right, last question. Now with Vive coding, we all seen the stats. There's so many apps out there. Where do you guys see? Where are we headed here? What are we talking about next year

Speaker C: in terms of the apps ecosystem? And uh, Joe is broad.

Speaker B: Take it wherever you want to go.

Speaker C: I mean, I think, uh, two things, right? So now obviously it's much easier to develop an app, but it's uh, exactly the same kind of companies that uh, are at the top. Right. I mean obviously there is some rotation, but the fact that it's much easier to code doesn't mean that you can crack distribution and therefore the distribution is very heavy on one side.

Speaker B: I would say too.

Speaker C: Yeah, right. But um, if we think about kind of uh, what's coming next. So I think, I mean obviously a lot of apps now try to kind of implement AI somehow. And the kind of, the clearest use case is the assistant who can like talk to you and help with something. I think probably the next layer would be agentic flows when the app actually can do something for you. Um, I mean, we haven't seen a lot of examples of it working and actually, to be honest, we haven't seen a lot of examples assistants working as well or meaningfully improving the app, uh, by adding them. So I think it's still a lot to go and I, uh, feel like technologically we're way more ahead of the product, uh, application of what we already have. So even if we stop everything that's happening on the technological side of things, I think there's still five years of innovation that we can implement.

Speaker B: Uh, yeah, because they can get done with a smaller team than like, I don't think the need for a big team is that necessary anymore. That's why I was asking if you.

Speaker C: I think it's marketing. You can be definitely smaller, you can be definitely smarter. Yeah. And, uh, I think if reface were born right now without obviously the processes, the legacy that we have, we would probably be able to be much smaller and do the same things. But, uh, it's something for the new companies to do much easier than for the others who existed already.

Speaker D: Right.

Speaker B: Closes out.

Speaker D: Yeah, I mean, I completely agree. I was struggling to think what do I think the future is going to look like next year? And I don't know what next year is going to look like, but one of the big bets I'm going to betting on is this premise that the future of successful companies, successful companies that are starting out like today and like startups at the moment and mobile apps that aren't already massive, isn't having a big team. It's having significantly smaller teams of great, great operators. Right. Like if you can have three to five A players who are using AI to the fullest, you're going to win versus a team of 50 who stuck doing things the old way. Um, and I think you gain efficiency and I think that's kind of where we're headed.

Speaker B: Ladies and gentlemen, do not go anywhere because we're about to do prizes. But give a round of applause for Nathan and Anton.

Speaker A: And that was the episode 267Ai in action. Real World Use Cases featuring the discussion panel from the Business of Apps London 2026. To listen to more Episodes, subscribe to our podcast on iTunes, Spotify, Stitcher, YouTube, podcasts. Search for business of apps and you will find us easily. Remember, we release Episodes on Mondays, so subscribe and you will be able to get new episodes on your smartphone, tablet or computer as soon as we release them. And please don't forget to leave us your view or comment on itunes. It is highly appreciated and all episodes will also be available on, um, business.com thank you for listening. See you next week.

Speaker D: Sam mhm.