You’ve probably been told that if you build a great AI tool, the users will come, and the money will follow. That’s a lie.
In the generative AI era, free users aren’t a growth metric. They are a financial liability.
In 2022, a startup named Tome launched. It did one thing: you type a sentence, and it spits out a complete presentation deck. In 134 days, they hit 1 million users—faster than Slack, Dropbox, or Zoom. By 2024, they had 25 million users. Investors like Reid Hoffman and Eric Schmidt threw $81 million at them, valuing the company at $300 million. They were the ultimate AI darling.
But in late 2024, founder Keith Peiris looked at the books. 25 million users. $3 million in Annual Recurring Revenue (ARR). That’s roughly a dime per user per year. He did the unthinkable: he killed the product. Not sold, not put on life support. Executed.
A problem “everyone has” and a problem “someone will pay a lot of money to solve” are two entirely different things.
Why did it fail? Because Tome solved the “0 to 60” problem, not the “60 to 95” problem. Generating a 10-page deck from a single prompt is a great party trick. But if you’re a sales rep pitching a $1 million deal to Nike, the AI doesn’t know that the VP of Logistics cares about inventory, or that your competitor just undercut your last quote. General LLMs don’t have that context. As Peiris put it: even if you hire a person with an IQ of 160 to make your slides, if they know nothing about your client, the presentation is useless.
AI is brilliant at helping you cross the blank page. It is terrible at helping you close the deal.
Realizing this, Peiris consulted Slack founder Stewart Butterfield, who had twice built failed games only to pivot their internal tools into massive successes. The advice? Shrink the team and start over. Tome went from 70 employees down to 6. They shut down the 25-million-user app completely and built Lightfield, an AI CRM that connects to company emails and meeting notes to provide actual, proprietary context.
The Silicon Valley playbook of “grow first, monetize later” is obsolete. Every time a free user generates a PPT, it costs compute. You’re burning cash to entertain students and freelancers who might use the tool once a month. If your AI startup is bragging about user acquisition, you might just be advertising your own bankruptcy. The real question isn’t how many people said “wow” at your demo. It’s how much pain they feel when you take it away.
FAQ
Q: Couldn't Tome have just forced free users to pay?
A: No. The demand was broad but shallow. Forcing payment on a low-value, occasional-use feature doesn't create revenue; it just drives users away, leaving you with high inference costs and zero goodwill.
Q: What's the practical implication for AI founders?
A: Stop optimizing for viral 'wow' demos. If your AI only solves the 0 to 60 problem (getting past the blank page), it's a novelty. You need to solve the 60 to 95 problem using deep, proprietary context to become a mission-critical tool.
Q: Is high user acquisition always a bad sign in AI?
A: In the GenAI era, yes. Free users burn compute. If they aren't paying, they are actively destroying your runway. High user acquisition without immediate monetization is a net-negative cost center.