Why AI Apps With 50,000 Engaged Users Still Shut Down

There’s a pattern playing out across the consumer AI landscape that most founders don’t see coming until it’s too late. You build something people love. You hit 50,000 engaged users. Your retention numbers look great. And then you look at the unit economics and realize the whole thing is underwater.

Nic Baird, CEO of Koah Labs — the company building a full-service ad network for AI-native apps — watched this happen repeatedly while at South Park Commons. Friends and fellow founders would build consumer AI products that users genuinely loved, then shut them down because the math never worked. It’s the problem that led him to start Koah in the first place.

The inference cost trap

The core issue isn’t that consumer AI apps can’t find users. It’s that every engaged user costs money in a way traditional apps never did. When your core experience is generative — meaning the app calls an LLM for every interaction — you’re paying inference costs on every session. The more engaged your users are, the more money you lose.

“We saw all these companies that were getting a lot of engagement and growing really quickly to shut down because they look at the projections of unit economics going forward with subscriptions only converting at three to 5% and say, this is never gonna make money,” Nic explains.

This is fundamentally different from a traditional SaaS business where serving an additional user has near-zero marginal cost. In consumer AI, growth can actively accelerate your burn rate.

Why subscriptions aren’t enough

The subscription conversion rate for consumer AI hovers at 3-5%. That means for every 100 engaged users, 95-97 are using your product for free — and each free session costs you inference fees. Even the users who do pay often don’t cover their own inference costs at typical price points.

Some founders tried dropping in legacy ad solutions like AdMob or AppLovin. The results weren’t much better. Character AI tried full-screen interstitial ads when users switched characters and got immediate backlash from their Reddit community. Perplexity launched ads at a $50 CPM, attracted initial interest, then shut the project down because they couldn’t get it right.

“Even teams like Perplexity are struggling to figure it out. It’s not something that you can just say, ‘Oh great, let’s do sponsored follow-up questions,’” Nic says. “This is a really hard problem to solve.”

The engagement paradox

Here’s what makes this particularly brutal: the apps with the worst economics are often the ones users love most. Character AI users spend an average of over three hours per day on the platform — twice YouTube and TikTok. That level of engagement is remarkable, but it also means every disruptive ad risks breaking a session that could last another two and a half hours.

This creates a paradox. The more engaged your users are, the more it costs to serve them, the more you need monetization, but the more you risk losing them if you monetize poorly. Traditional display ads solve the revenue problem by creating the churn problem.

What actually works

Nic points to Open Evidence as the proof that the model can work. The company serves a million doctors with an AI tool and generates $150 million in annual revenue — entirely from advertising, with no subscription option. The key: an extremely valuable, highly engaged niche audience that advertisers want to reach.

“Why don’t they deserve to live?” Nic asks about the consumer AI apps shutting down. “That’s why I think all great companies are built from serious pain points.”

The path forward isn’t subscription-or-ads. It’s building for engagement first, reaching a critical mass of users who genuinely need your product, and then layering in monetization that doesn’t destroy the experience they came for.

FAQ

Why do AI apps fail even with high user engagement?

Consumer AI apps face a unique unit economics problem: inference costs make every user session expensive, while subscription conversion rates sit at 3-5%. An app with 50,000 engaged users can grow quickly and still lose money on every interaction, making the business unsustainable without alternative monetization.

What are the main monetization options for consumer AI apps?

The three primary options are subscriptions, traditional display advertising, and native AI-specific ad formats. Subscriptions convert at 3-5% for most consumer AI apps. Display ads (AdMob, AppLovin) generate revenue but damage user experience and increase churn. Native ad formats like Koah’s aim to maintain engagement while generating $5 CPM.

How much does it cost to run a consumer AI app per user?

Costs vary by model and usage intensity, but every generative interaction carries inference costs. Unlike traditional apps with near-zero marginal cost per user, AI apps pay per query. High-engagement apps like Character AI (3+ hours per user per day) face especially steep costs, making unpaid users a direct financial burden.

Why did Perplexity shut down its advertising program?

Perplexity launched ads at approximately $50 CPM and attracted initial advertiser interest, but shut the project down after finding the execution too difficult. Getting targeting, brand alignment, user experience, and conversion right simultaneously requires deep ad infrastructure expertise — not just contextual relevance.

Why did Character AI’s ad approach fail?

Character AI implemented traditional full-screen interstitial ads that appeared when users switched between characters. The Reddit community reacted with immediate backlash because the format was disruptive to a deeply immersive experience. Users spending 3+ hours per session don’t tolerate interruptions that break their flow.

What is the Open Evidence monetization model?

Open Evidence serves one million doctors with an AI medical tool and generates $150 million in annual revenue purely from advertising — no subscription option. The model works because doctors are an extremely valuable audience for pharmaceutical companies and premium brands, making each user worth significant ad revenue.

When should an AI app founder start thinking about monetization?

Plan your monetization strategy from day one, but don’t implement it immediately. Focus first on building something 10,000 users genuinely engage with. Once you’ve proven retention and value, then layer in monetization. Starting with user experience and adding revenue later is more sustainable than optimizing revenue early and losing users.

How do inference costs compare to traditional SaaS hosting costs?

Traditional SaaS has near-zero marginal cost per additional user — serving the 50,001st user costs almost nothing. AI apps pay per inference call, so every query, every session, and every engaged minute adds direct cost. This makes the standard SaaS growth playbook of “grow first, monetize later” financially dangerous without alternative revenue streams.

Can a vibe-coded AI app actually generate meaningful ad revenue?

Yes, if the app serves a specific audience with commercial intent. A niche AI tool with engaged users can generate approximately $5 CPM with native ad formats. At a 30% fill rate, ad revenue can cover inference costs and enable sustainable growth — but only if the audience is specific enough for relevant ad targeting.