Silicon Valley's hottest AI financial metric is also its least trusted

The co-founder of Cluely — an Andreessen Horowitz-backed startup with the motto, "Cheat on everything" — ignited a controversy earlier this month after confessing to lying to a reporter about his company's performance on a popular Silicon Valley metric: annual recurring revenue, or ARR. 

Cluely Chief Executive Officer Roy Lee told a TechCrunch reporter the company's ARR doubled in one week to $7 million. Later on X, Lee said he "got a random cold call from some woman asking about numbers and told her some BS," calling it "the only blatantly dishonest thing I've said publicly online." He added that the post would serve as "my formal retraction," adding that the actual number was $5.2 million. 

While Lee may be unusually bold, taking liberties with ARR is common for artificial intelligence startups, Silicon Valley investors say. A measurement of current sales extrapolated over 12 months, the metric is squishy enough that it leaves plenty of room for interpretation. Even as it's become ubiquitous in the AI era — it's also become one of the least trusted yardsticks for gauging a startup's growth. 

"The startup world has always been a bit more of a Wild West," said Chuck Eesley, a professor of management science and engineering at Stanford University. "There are no audit requirements, there are no SEC definitions, so basically there's no cop on the beat other than the VCs and acquirers doing their due diligence. So essentially, the number can mean whatever the founder needs it to mean when they walk in to do a deal or do a fundraise." 

The basic components of ARR calculations are simple: Take one month's revenue from recurring contracts and multiply by 12 for an annual projection. It's also not to be confused with "annual run rate revenue" — a similar, perhaps even more popular metric with an identical initialism, but that doesn't concern itself with whether the sales are recurring.

Companies reporting ARR for individual products or overall sales have included Anthropic, Glean and Cursor, the coding company investors have called the fastest-growing startup of all time in terms of revenue. Meanwhile, the recurring revenue of specific OpenAI products has become a closely watched number for media outlets. 

There's nothing inherently wrong with tracking ARR, and if a company is adding new subscription clients each month, it can present a more accurate picture of revenue than looking backward at actual sales. Until recently, ARR was thought of as a trusty benchmark for software businesses, particularly those selling predictable services to other businesses, said Darren Yee, a senior venture associate at NYU's Innovation Venture Fund.

"This worked really well when subscription pricing was very straightforward," Yee said. "And that's been true for a long time, basically up until AI."

But with recurring revenue, there's enough leeway in how exactly to measure it — like what contracts count and what time period to use — that it's relatively easy for startups to massage the figures. The numbers can be particularly variable if revenue fluctuates from week to week, or some recurring subscriptions lapse. 

For example, Stanford's Eesley said that plenty of AI business customers are eager to try out new tools, but drop them after a trial period. That kind of revenue can be counted as "recurring," even though the contract doesn't renew. "A lot of companies want to experiment with AI these days because they recognize that it's a big trend, it's a big shift," Eesley said. "There's a lot of budget there for experimentation, but not all of those experiments pan out into actual recurring revenue."

Another reason ARR isn't as reliable today is that startups are often charging customers based on how much they use the product, shifting away from more regular subscription models. "Customers may have a nominal subscription number but are paying mostly for usage. This gives very lumpy revenue attribution in the early days," Yee said. "You can't just take one month of subscriptions and multiply by 12 and get what that represents in an annual contract, because it probably won't play out that way." 

Is recurring revenue still a useful measurement? Cluely's Lee doesn't think so. In an expletive-laden email to Bloomberg, Lee expressed his distaste for the media, as well as ARR as a metric for startup growth.

"What the f--- even is ARR for a company that is less than a year old?" he wrote. "The calculation doesn't even make f---ing sense for us, it's a fake accounting number developed by fake accounting people." Lee added that his measurement of the metric was changing 20% week to week. 

Yet there may not be many better options, and more elaborate auditing processes could be more trouble than they're worth. "I think we should be careful about imposing a lot of auditing and accounting costs on small startups and stifling a lot of the innovation and experimentation that should be going on," Eesley said.

Instead, Chris Sloan, co-chair of Baker Donelson's emerging companies group, recommends broad transparency. "Always err on the side of disclosing too much rather than too little," Sloan said. Even if it's not technically securities fraud, he added, "If you have broken trust with a potential investor, you're never getting that back."