Founder Playbook · Sub Club by RevenueCat
15 tactics from Olivia Moore
App Revenue Is Booming and It's Not Just AI Apps
Watch the full episode“Consumer application in-app purchases finally passed games. A decade ago it was like six to one the other way, so it's been a really big and meaningful shift.”
Consumer app IAP finally passed games after a decade of 6:1 the other way
After a decade where games out-earned consumer apps roughly 6:1, non-game consumer IAP has now overtaken games. The shift is broad — it's not just AI driving it. Categories like social, utilities, and movies/TV all posted billion-dollar gains in the latest year.
“21% year-over-year growth in non-game apps which amounts to like billions and billions of dollars — and only 3.5 billion of that according to Sensor Tower was attributed to generative AI.”
21% non-game revenue growth — only $3.5B from generative AI itself
Non-game app revenue grew 21% YoY into the billions, with only $3.5B from generative AI directly. Movies/TV, social, and utilities all posted billion-dollar gains. The boom is wide, not narrowly AI-driven — and AI is lifting categories that don't even call themselves AI.
“These companies are monetizing at two times the average revenue per user, if not higher, than the pre-AI complements. Consumer businesses are ramping revenue faster than enterprise businesses, which has probably never before been true in the history of software.”
Top AI consumer apps monetize at 2x the ARPU of pre-AI peers
a16z's data on the top 50-100 AI consumer apps: ARPU is roughly 2x their pre-AI counterparts, and they're monetizing from day one rather than waiting 5-10 years. For the first time in software history, consumer companies are ramping revenue faster than enterprise.
“Product people, marketing people are scoping things out on Lovable and getting a spec or a new idea approved in like a day, where previously they would have to loop in a bunch of engineers and it would take weeks.”
Product/marketing people now scope a spec on Lovable in a day, not weeks
Non-engineers now scope and validate ideas in Lovable or Cursor in a day instead of looping engineers for weeks. The bottleneck has moved off the eng team, and PMs/marketers can prove an idea end-to-end before anyone writes production code.
“We're looking for products that are not just thin features that can capture transitory opportunities. We're looking for teams that are extremely opinionated about a specific use case and will build out and productize the models in a way that it's much more compelling, easy to use, and more effective than what a broad-based LLM platform will do.”
Be opinionated and verticalize — thin features get price-competed to zero
Thin AI wrappers get price-competed to zero as labs ship the same capabilities natively. Durable consumer AI apps are opinionated, verticalized, and productize models in ways general LLM platforms won't bother to — that's the defensibility moat.
“Pre-2023, unless you were building a marketplace, the idea that you would want to or be able to monetize your users in the first 5 or 10 years was pretty crazy. ChatGPT anchored a lot of people on $20 a month if not more, and when you incorporate usage-based pricing you can have whales paying hundreds if not thousands of dollars a month.”
ChatGPT anchored consumers on $20/mo — the $60/yr ceiling is dead
For a decade pre-AI, consumer apps couldn't monetize directly — users were the product, you served ads. ChatGPT trained a generation to pay $20/month, and that's the new floor, not the ceiling. The willingness-to-pay shift benefits every consumer app, not just AI ones.
“Sam Altman actually said something interesting on the OpenAI town hall a couple weeks ago. Someone asked him 'are you going to kill my startup?' and he was basically like: look, if you as a company are happy when the models improve, you're safe. That's a great place to be.”
Sam Altman's safety test — are you happy when the models improve?
The litmus test for whether your consumer AI app is durable: do you get stronger when GPT-5 ships, or weaker? If model improvements help you, you're on the right side of the labs. If they directly threaten you, you're a thin feature waiting to be eaten.
“The best thing that you can do to build intuition on these products is just to use them. The quality of consumer AI apps on mobile is so much better and more sophisticated than they were 6 months ago. Things are changing so fast — the best way to stay on top is just to use them.”
Use the products yourself — intuition is the moat
Olivia's tactical advice for founders trying to read the market: stop reading hot takes and actually use the AI apps shipping each month (Wabi, Gizmo, Sakai, Poke, Tasklet). Mobile AI quality jumped massively in 6 months — hands-on use is the only way to see what's resonating.
“If you look at top 10 apps by saturation for every age group — 18 to 25, 35 to 40 — ChatGPT is the only AI app that cracks the top 10 for any age groups, which is crazy. All of the top ones are ChatGPT, Claude — these very general LLM assistants. We're going to see a big and exciting expansion in the types of AI products delivered via phone.”
ChatGPT is the only AI app in any age group's top-10 saturation — headroom is huge
Despite the AI hype, ChatGPT is the only AI app cracking top-10 saturation in any demographic. The vertical/opinionated AI app wave hasn't hit mobile yet — for female users ChatGPT doesn't even make the top 10. The audience opportunity for category-specific AI apps is wide open.
“Grindr now has like a several hundred dollar a month AI that swipes and makes matches for you. Is the average user going to pay for that? No. But 1% of users will, and you're now able as an app to capture so much more value because you're essentially serving not just the product anymore but you can serve outcomes if people are willing to pay for them.”
Sell outcomes, not products — Grindr's $hundreds/mo AI auto-swipe agent
The new monetization frontier is selling outcomes, not features. A small slice of power users will pay 10-50x the base subscription for AI that does the work for them. Mix subscriptions, usage-based, and ads to capture the full demand curve instead of pricing for the median — perfect price segmentation is finally possible.
“A category that's not technically AI that saw a ton of growth is short dramas. They're making hundreds of millions of dollars now, and I've met several startups that are largely using AI to either fully generate or edit these short dramas, so they can make them at much lower cost and spend more on marketing to distribute them.”
Short-drama apps do hundreds of millions — AI collapses production cost so you outspend on marketing
Short-form drama apps are quietly a hundreds-of-millions-of-dollars category. The startups winning are using AI to generate or edit the content itself, collapsing production cost so they can pour the savings into paid acquisition. The leverage isn't in the app — it's in the cost structure that lets you dominate distribution.
“A lot of startups now are achieving really fast growth by essentially subsidizing the model cost to the end consumer — like giving away Nano Banana Pro for free. It's almost similar to the on-demand economy like Uber, DoorDash. Most of those, the unit economics catch up to you eventually. You have to be quite good at marketing and distribution and really lock users into the product if you're going to grow that way.”
Subsidizing model cost is the new Uber/DoorDash — unit economics catch up
A common 2026 distribution playbook: subsidize model cost (give away Nano Banana Pro, Sora, etc. for free) to grow fast. It works like the on-demand era — unit economics eventually catch up. Some hit escape velocity and turn users profitable, but only if you're exceptional at marketing and genuinely lock retention. Otherwise you're lighting cash on fire.
“The first version of AI takes we saw on therapy or coaching were very skeuomorphic to the pre-AI service — therapy session but it costs $10 because it's run by a voice agent for 60 minutes. What we're learning is for a lot of people the 60-minute session format doesn't work or fit into their life. AI is now allowing you to reimagine how to build a different experience.”
Don't skeuomorph the pre-AI service — reimagine the format itself
Early AI consumer apps copied the offline ritual (60-min therapy at $10, hourly AI tutoring) and underperformed. The winners reimagine the format itself — micro-sessions, on-demand, async, or letting the user co-design the experience. Don't bolt AI onto an existing format; rebuild from scratch around what AI uniquely enables.
“We invested in Gamma which does AI for presentations and slide decks. Yes you can do decks within ChatGPT, Claude, other products now — but they're never going to be a first-class citizen in the way they are for Gamma, and there's a real opportunity to retain users in that way.”
Verticalize for the retention moat — Gamma owns decks because ChatGPT never will
Retention defensibility against general-purpose LLMs comes from being a first-class citizen for one job. Gamma retains users for decks because ChatGPT will never make decks its primary surface. Verticalization is the retention moat in the LLM era — pick a job that the labs won't prioritize and own it end-to-end.
“Much has been said about the era of the one-person one-billion-dollar company, and honestly I think we're not too far from it. We've seen and invested in companies that are doing a million plus in ARR per employee, which is a metric that we haven't seen before. We're also just so early — I don't think we've yet seen what the Claude Code for marketers and product people and salespeople looks like.”
$1M+ ARR per employee is the new bar — one-person billion-dollar company isn't a meme
a16z is seeing portfolio companies hit $1M+ ARR per employee — a metric that didn't exist pre-AI. And we're still early: Claude Code shipped for engineers, but the equivalent for marketers, PMs, and sales hasn't yet. Efficiency only goes up from here, freeing humans to do the things only humans can do — generate ideas and spend time with customers.