In 2025, consumers spent more on non-gaming apps than on games for the first time – $85.6 billion versus $81.8 billion, according to Sensor Tower. The biggest driver was AI apps, which more than tripled their revenue, crossing $5 billion. Bogdan Filippov, co-founder and CTO of Clickfun, runs a consumer mobile portfolio with over 50 million downloads and five million monthly active users – and is now building in the fastest-growing segment of that market. We spoke with him about what retention actually means at scale, why the dotcom comparison doesn’t hold, and what it takes to build an AI product users trust enough to come back to.
Clickfun’s portfolio is built on educational trading simulators – not what most people picture when they think about AI or consumer tech. What does that experience actually teach you about how mobile users behave?
Users don’t open an app to interact with an interface. They open it to get something. The products in our portfolio that perform are the ones users return to because the app delivered on its promise. We track four metrics every day – ARPU, CPI, retention, and time spent. The signal that matters most is whether someone comes back the next day. If Day 1 retention is above 30%, you have something to build on. Below that, you’re acquiring users faster than you’re keeping them, and the math never works out.
That’s an unusually operational framing for product development. Where does that discipline come from?
Partly from years of iOS development before I started founding things – you get a very concrete sense of what actually ships versus what sounds good in a meeting. But a lot of it came from a failed experiment. In 2022 I co-founded CreoAI, an AI product for personalized image generation. The technology wasn’t there yet – the compute requirements, the render times made the user experience unworkable at any real scale. We shut it down. That experience gave me a much sharper sense of the difference between a viable idea and a market that’s ready to support it. They’re two very different things, and when you’re close to your own product, it’s easy to confuse them.
And that experience eventually led you to the Runet Prize jury – evaluating other founders’ projects. What did you see from that vantage point?
(The Runet Prize is Russia’s national internet industry award, organized by RAEC, running since 2004. The Expert Council – a closed group of senior industry figures – votes to select winners from shortlists compiled by a broader community of IT professionals.)
I was on the Expert Council for two years. Hundreds of applications, twenty-something categories – mobile apps, fintech, edtech, government services. The judgment you’re forced to develop isn’t about whether a product is well-built. It’s about whether the timing is right. You see a lot of projects where the idea is correct, the execution is solid – and it still doesn’t work because the market isn’t ready. The technology costs too much, users haven’t formed the habit, nobody’s willing to pay yet. Inside your own startup, that’s almost impossible to see. When you’re evaluating someone else’s work, it’s obvious within the first few minutes. After CreoAI, I recognized that pattern immediately every time I saw it.
You’ve also been running Muse Journal since 2017 – a journaling app, alongside everything else. Why does that product matter to this conversation about AI?
Because it established a principle I’m now taking into a much harder technical context. From day one we built Muse Journal on a hard privacy commitment: no data collection, no access to entries. Full privacy. In a category where most wellness apps treat behavioral data as a revenue stream, that became a defining characteristic – and it’s held for eight years. The users who stay are there because they trust it. When I move into AI skincare, that question of trust becomes structurally more difficult. A journaling app can credibly promise not to read your entries. A skincare app that’s actually personalized needs data to work. How you handle that isn’t just an ethical position – it shapes whether users come back.
That privacy principle becomes a much harder engineering problem when you move into AI – especially health. How does it carry into your next project?
That’s where it gets complicated. I’m building an AI skincare app – a personal companion designed to help users slow skin aging. The technology is finally accurate enough to build a real consumer product around: Haut.AI’s Face Analysis 3.0, one of the leading platforms in the space, analyzes 29 skin parameters from a single selfie. But what you do with that data – whether it stays on-device or goes to the cloud – is as much a product decision as the diagnostic algorithm itself. Most apps in this space are building for cosmetics companies – B2B platforms that serve as a diagnostic layer for their product lines. I’m going straight to the consumer. That means user trust has to be earned directly, not through a brand they already have a relationship with.
The AI skincare market is crowded – Perfect Corp, L’Oréal ModiFace, Haut.AI itself. Why enter now, and why direct-to-consumer specifically?
I’m entering now because the market has caught up to the technology in a way it hadn’t in 2022. After CreoAI, I spent time watching the space develop rather than forcing a product into it. The difference between then and now isn’t the quality of the ideas – it’s the infrastructure: better models, cheaper compute, users with enough AI product experience to know what they’re willing to pay for. The brand integration model leaves a gap. When you’re using a skin diagnostic embedded in a cosmetics company’s app, the recommendations get filtered through their product catalog. A standalone AI companion doesn’t carry that constraint. The recommendations can actually be personalized, not just feel that way.
You’ve said publicly that AI apps are still at their starting point. A lot of people push back on that – they see another dotcom moment. What’s your read?
The dotcom comparison doesn’t hold up when you look at the revenue data. Many dotcom companies were still searching for viable business models when they collapsed. AI apps are monetizing now – Sensor Tower data for 2025 shows AI app revenue more than tripled year-over-year, crossing $5 billion for the first time, with sessions topping one trillion. That’s not a speculation bubble. That said, there’s a real problem in the category that doesn’t get enough attention: retention is falling. The industry is adding new users faster than it’s keeping existing ones. For most players, that’s a structural issue. For a product built around daily habits and trust, that gap is the opportunity. That’s what I’m building toward.
