Thousands of AI startups are fighting for the VC funding needed to win a slice of the enterprise market. But according to Scott Stevenson, cofounder and CEO of the legal AI startup Spellbook, many are inflating their real revenue to get it. In a viral tweet on April 17, Stevenson called out these fledgling companies for perpetuating a “huge scam” in their metric reporting.
Specifically, Stevenson’s tweet concerned the misuse of a revenue metric common in the AI startup world. Annual recurring revenue, or ARR, is meant to show the annualized value of recurring subscription contracts. It’s typically calculated by projecting the current month’s subscription revenue over a full year. So if a startup invoices $1 million in January, its ARR for the current year would be $12 million, on the assumption that the same monthly revenue will continue.
He said some AI startups have begun basing ARR figures on future revenue that is far from certain, noting that they do this by blurring ARR with something called CARR, or “contracted annually recurring revenue,” which can include future revenue.
“Often in decks CARR and ARR are reported as separate metrics, but when companies go to press they are actually reporting CARR and calling it ARR in order to have the biggest number possible,” Stevenson told Fast Company in an email exchange.
CARR can be used legitimately to describe the value of long-term contracts, such as in healthcare AI or energy optimization, where revenue accrues gradually over a lengthy deployment. “Initially this may have been innocent as companies were trying to get a little extra credit for deals they signed that were not live,” Stevenson said.
But CARR shouldn’t be confused with ARR, which includes only subscription revenue that can be invoiced to the customer. “The gap between these metrics has grown massively,” Stevenson said. “I know 100% of confirmed cases where the gap is as much as 3-5x.”
In practice, the obfuscation can take a few different forms. A startup might, for example, count a full year of revenue even if its contracts allow a customer to opt out after one month. Or a startup might count a free three-month “pilot” as three months of real revenue.
“I was talking to an investor yesterday who sees that all the time from early-stage companies,” Stevenson said on a recent TBPN podcast. “Coming out of accelerators, saying they have a million ARR, and they look under the hood and it’s just all pilots that haven’t converted yet.”
Or a startup might write in a contract that the customer will start paying for a certain feature after it’s built. The startup then counts revenue from the months during which the feature is being built. But there’s just no guarantee the feature—or the revenue—will ever come to fruition.
The post also drew a wave of agreement from founders and VC partners in the replies. “This is rampant and it’s honestly distorting the benchmarks for everyone,” wrote Equal Ventures partner Rick Zullo. FPV Ventures partner Nikunj Kothari added, “I have stopped looking at headline number for this reason.”
As some commenters on Stevenson’s X post pointed out, a VC considering an investment will likely examine a startup’s contracts and separate real revenue from projected revenue. Journalists, by contrast, typically lack access to those contracts and may take startups at their word that ARR reflects actual revenue.
According to Stevenson, journalists should probe startups on whether their whole ARR number really reflects “live” revenue (invoiced revenue) or if some of it is “contracted ARR,” noting that some VCs may go along with the deception.
“I feel like there is a bit of a ‘silent pact’ between founders and VCs not to discuss the difference with press, and to often use the bigger number for more coverage,” he said.
Some insidious second-order effects could follow.
If one AI startup in a given space begins inflating its revenue using an elastic definition of ARR—or even just appears to—others in the space, perhaps fearing the appearance of falling behind, may feel pressured to follow suit.
“These illusions can create mania, cause companies to chase each other’s ghosts, and to do risky things that they shouldn’t—also very bad for employees who may not understand real ARR numbers, and for customers trying to understand the landscape,” Stevenson said.
There is already widespread skepticism about the earning potential of AI companies. That skepticism extends to Big Tech firms and AI labs spending heavily on large models and data centers, as well as to smaller startups building enterprise applications on top of those models. Overestimating the impact of any of these players only adds more air to the bubble.
