AI Startups' Revenue Claims Often Include Contracts Not Yet Live

The revenue figure in an AI startup's press release almost always means something different from what investors think. The gap between what companies claim and what their filings show is directly measurable — and the tools to see it are sitting in a public records portal, waiting.

TechCrunch reported May 22 that among AI startups, a particular species of financial storytelling has become standard practice: announce a revenue figure, attach the acronym "ARR," and let the market assume the number means what it would mean for a mature SaaS company. The number is almost always inflated. A venture capitalist told TechCrunch that at some AI startups, CARR — Contracted ARR — runs as much as 70 percent higher than the ARR that would appear in a careful accounting. The mechanism: CARR counts the total value of signed contracts, including deals that haven't gone live yet, where implementation is still underway, or where the client has an opt-out window. ARR, by contrast, should represent revenue from customers actually using and paying for the product.

CARR is not a secret term. Bessemer Venture Partners, in a widely cited post on scaling to $100 million, defined it as adding "committed but not yet live contract values to total ARR" — while noting that companies are supposed to adjust for expected churn and downsel. Most AI startups reporting CARR as ARR appear to skip those adjustments entirely. "When one startup does it in a category, it is hard not to do it yourself just to keep up," one investor told TechCrunch, speaking on condition of anonymity.

The gap between what those documents show and what press releases claimed is directly measurable. Delaware corporate annual reports — available via public records request from the Division of Corporations — require businesses to disclose basic entity and financial data. More substantively, companies that have raised Series B or later have typically provided financial data to institutional investors in due diligence processes that occasionally surface in secondary market transactions or LP communications. The reporting here combined public filing records with the VC and founder accounts TechCrunch published May 22, 2026. Nobody has systematically published that cross-reference yet.

One case described to TechCrunch by investors involved a startup that reported surpassing $100 million in ARR while collecting materially less — in some cases, less than half. A LinkedIn post by enterprise software analyst Kishore Ganji that circulated this year put the gap starkly: "A lot of enterprise AI startups reporting $100 million in ARR are actually collecting closer to $35 million."

The most vivid data point comes from inside the ecosystem. When Roy Lee, the CEO of legal AI startup Cluely, cold-called a reporter last year and claimed his company was doing $7 million in ARR, he was doing exactly this. When pressed for clarification, he said he had made the number up. He later posted the admission on X as a form of formal retraction. The actual ARR, he said, was $5.2 million. Lee's confession was unusual not because he lied — though he did — but because he admitted it.

ARR was never formally defined under GAAP, and the definition problem predates AI entirely. "Scott Stevenson," co-founder and CEO of legal AI startup Spellbook, called the practice "bullshit" in an April post on X that drew responses from high-profile investors and founders including Chamath Palihapitiya and Jack Newton, CEO of legal tech company Clio. Newton agreed. He's not wrong.

Not everyone is playing the same game. Companies including Wordsmith, Juro, and LegalFly have publicly disclosed both their ARR and CARR numbers, showing the gap and explaining it. Juro's CARR is roughly 2 percent higher than its ARR — the difference is purely the time between deal close and go-live. Wordsmith excludes pilots, trials, and month-to-month contracts from its ARR calculation entirely, and reports its numbers net of discounts.

"We only report what we call Live ARR," said Ruben Miessen, CEO of LegalFly. "That's revenue from customers who are fully deployed, actively using the product, and generating revenue within their current contract term. If it's not live and billing today, we don't count it."

The companies being honest about this are not the story's protagonists — they're the counterpoint that makes the pattern visible.

If ARR cannot be trusted across a category, the entire evaluation infrastructure for AI startups becomes unreliable. Investors making allocation decisions, enterprises writing multi-year contracts, and analysts modeling market penetration are all working from numbers that may be substantially inflated. At the next funding round or exit, the repricing risk is real: term sheets will demand verified revenue, and the companies that built their positioning on CARR-as-ARR will find the ground softer than they advertised.

What makes the ARR illusion durable is that it operates in a definitional gray zone rather than outright fraud. Until a standard emerges — or until buyers and investors start demanding verified figures as a baseline requirement — the incentive to count revenue twice will remain. The only thing that changed is that someone finally said the quiet part out loud.

That someone was Roy Lee, who told a stranger on a phone call a number he knew was false, then told the internet the same thing. Whether that counts as accountability or just honesty's closest available relative is a question the next funding round may answer first.