As an independent growth advisor and angel investor, Phil Carter helps consumer subscription companies scale their growth engines. In a conversation with David Barnard on the Sub Club podcast, he breaks down the tension between maximizing short-term revenue with hard paywalls and building massive businesses through freemium models—and why the most powerful AI features might not need the most expensive LLMs.
Moving from checkers to chess: The multi-step paywall
For the vast majority of subscription apps, a hard paywall is the right answer. The data is clear: hard paywalls often convert five times better than freemium models. If you’re a bootstrapped startup operating with limited capital, it’s the most reliable and low-risk way to grow your business.
But for companies trying to build a billion-dollar business, freemium is often the only way to attract a massive user base at the top of the funnel. Phil Carter calls the transition from a hard paywall to freemium “moving from playing checkers to playing chess” because it requires significantly more sophistication.
Recently, Carter worked with a client to navigate this transition. Instead of simply dropping the paywall, they implemented a “multi-step paywall.” The product is free, but users are offered a seven-day trial of the best version. After the trial, they are prompted to subscribe to maintain that maximum value. Combined with pricing and packaging optimizations, this shift resulted in a 75% increase in LTV per user. The business moved from excluding users with a hard paywall to growing much more quickly through organic acquisition.
The trap of the value-to-noise ratio
With AI dramatically lowering the cost of software development, the pace of innovation has accelerated. It’s easier than ever to ship new features constantly. But this speed can become a trap.
“A fallacy that a lot of companies fall into is like, well, more products and more features equals more value and therefore we should just ship products and features as fast as we possibly can,” Carter explains.
The bottleneck isn’t development speed; it’s the capacity of the human brain to absorb the product experience. As an app becomes bloated with new capabilities, the absolute value might go up, but the complexity and noise increase alongside it. This causes the “value-to-noise ratio” to plummet. To combat this, teams must rigorously analyze which features actually drive long-term retention and aggressively prune the rest.
Why cheaper LLMs can build better products
It’s tempting to default to the most powerful frontier models from OpenAI or Anthropic when building AI features. But Carter points to Gamma, an AI-powered presentation tool that reached profitability within six months of launching its AI features in early 2023.
One of the reasons for Gamma’s rapid path to profitability was their savvy use of underlying LLMs. They didn’t always use the most powerful models. Instead, they found that for their specific use cases, longer-tail models provided performance that was “good enough” while delivering significantly faster response times and drastically lower compute costs.
“The product experience is a function not just of the output that the LLM provides, but also how fast it provides that output and how expensive it is to generate that output,” Carter notes. For many consumer apps, speed and affordability are more critical to the user experience than peak AI performance.
The 400-creative-per-month advantage
AI isn’t just changing the core product experience; it’s fundamentally altering user acquisition. Carter highlights Runna, a running app that used AI tools to exponentially increase their creative testing volume.
Runna went from producing tens of creative concepts per month to over 400. This massive increase in volume isn’t just about lowering CAC by finding winning ads faster. It creates a much more rapid learning cycle. By testing hundreds of permutations—using tools like ElevenLabs for voiceovers or Suno for background music—the marketing team learns exactly what resonates with users. Those insights don’t just optimize ad spend; they feed directly back into product roadmap decisions.
In the full episode, Phil also discusses how AI can turn the first 60 seconds of onboarding into a magical experience, why nicheification is the new strategy for crowded app categories, and the importance of building extrinsic triggers to form user habits.