AI startup Cluely is embroiled in scandal over misreported revenue figures.

Cluely co-founder admits inflating a key financial metric, reigniting debate over how trustworthy popular growth metrics are in AI startups

The co-founder of Cluely, a startup backed by Andreessen Horowitz and known for its slogan “Cheat on everything”, has sparked controversy after admitting he misled a journalist about the company’s performance on a Silicon Valley favourite metric: annual recurring revenue (ARR).

Cluely’s chief executive, Roy Lee, told a TechCrunch reporter that the company’s ARR had doubled in a week to reach $7m. He later wrote on X that he “got a random call from some woman asking about numbers and I told her bullshit”, describing it as “the only explicitly dishonest statement made publicly online”. He added that the post served as an “official retraction”, clarifying that the real figure was $5.2m.

Investors in Silicon Valley say ARR manipulation is common among artificial intelligence startups. Because ARR takes current sales and projects them across 12 months, it is flexible enough to allow plenty of interpretation. In the AI era, it has remained popular-but it has also become one of the least dependable ways to gauge startup growth.

“Startup world has always been somewhat of the Wild West. There are no audit requirements, no specific requirements from the SEC [the US Securities and Exchange Commission], so there’s basically no ‘police’ except venture capitalists and buyers doing diligence. So the number can mean whatever a founder needs it to mean to close a deal or raise money,” said Chuck Eesley, a professor at Stanford University.

How annual recurring revenue (ARR) is supposed to work

In principle, ARR is straightforward: you take monthly revenue from recurring contracts and multiply by 12 to create an annualised projection.

It should not be confused with “annualised run-rate revenue”-a similar but even more widely used metric that shares the same acronym, but does not consider whether sales are actually recurring.

Some companies that report ARR for specific products or for overall sales include Anthropic, Glean and Cursor, which investors describe as the fastest revenue-growing startup in history. Meanwhile, recurring revenue for individual OpenAI products has also become a key figure tracked by the media.

Why ARR has become easier to game-especially for AI startups

ARR tracking, by itself, is not inherently unreliable. Where a company is consistently adding new subscribers month after month, ARR can offer a clearer view of revenue than backward-looking analysis of completed sales. Until recently, ARR was seen as a dependable reference point for companies-particularly those selling predictable services to other businesses-said Darren Yi, a senior venture associate at NYU’s Innovation Venture Fund.

“It worked really well when subscription pricing was very simple and up through the pre-AI era,” Yi said.

The challenge is that “recurring revenue” leaves room for discretion: which contracts are included, over what timeframe, and under what assumptions. That latitude can make it relatively easy for startups to massage the figure. The metric can be especially volatile when revenue swings week to week, or when some recurring subscriptions drop off.

Eesley noted that many AI business customers are keen to test new tools, only to abandon them after a trial period. That income may still be counted as “recurring” even if the contract is never renewed.

Another reason ARR is less dependable today is the shift away from simple subscription models. Startups increasingly charge based on how customers use a product rather than on a fixed, regular fee. “Customers may have a nominal subscription, but they’re mostly paying for usage. That leads to very unstable revenue attribution at the early stage. You can’t just take one month of subscription, multiply it by 12 and get what represents an annual contract because it probably won’t work that way,” Yi said.

Roy Lee and Cluely argue ARR is the wrong yardstick

Is recurring revenue still a useful metric? Lee does not think so. In a profanity-laced email to Bloomberg, he expressed contempt for the media and for ARR as a measure of startup growth. “What even is ARR for a company that is less than a year old?” he wrote. “This calculation doesn’t even make sense for us, it’s a fake accounting number made up by fake accountants.” Lee added that his metric was changing by 20% every week.

Still, there are not many alternatives in common use, and more rigorous audit processes may be too burdensome. “I think we should be cautious about imposing a lot of audit and accounting costs on tiny startups, so as not to squash the innovation and experimentation that should be happening,” Eesley said.

Instead, Chris Sloan, co-chair of the Emerging Companies Group at Baker Donelson, recommends broad transparency. “It’s always better to disclose too much than too little,” Sloan said. Even if it is not technically securities fraud, he added: “If you’ve lost the trust of a prospective investor, you will never regain it.”