Korva is a scroll of the internet’s best ideas.
Measure LTV against gross profit
Usage-based AI products carry a bigger average cost per user as well as a bigger average revenue per user. One user can be a $50,000 annual contract while another is $50, and Meta and Google's algorithms have no idea what to do with that variance. The anthropic bill is only one cost. Modal, inference, and every other cost of serving the product belong in the figure, and a finance leader can gut check the list.
Early on, a one-to-one LTV to CAC ratio is acceptable for a company that has raised a lot, because distribution decides who wins when a hundred new products appear on X every day and two have cloned your website. Below one-to-one money is simply burned. Later, sponsorships and AEO lift the ratio, and most teams aim for around three to one.
Why do AI products cost more to serve?
Language model providers charge by the token, which OpenAI's rule of thumb puts at about three-quarters of an English word. Cost therefore scales with how much text a user sends and receives. A heavy user who runs long tasks can cost far more than a light one. Traditional software, by contrast, costs little to serve each added user.
Where did the three-to-one LTV rule come from?
Venture investor David Skok of Matrix Partners popularized the benchmark in his widely read writing on software-as-a-service metrics. He suggested a customer lifetime value of at least three times the cost of acquiring the customer. He also wrote that the cost of acquiring a customer should be earned back within twelve months. The ratios were derived for subscription software with high gross margins.
More from The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
