On April 17, Scott Stevenson, CEO of legal AI startup Spellbook, wrote on X: "time to expose a massive scam in AI startups." The post got over 200 reposts and reactions from major investors and founders. TechCrunch published an investigation after talking to a dozen founders, investors, and startup finance people.
How the scheme works
ARR (annual recurring revenue) is a traditional SaaS metric — revenue from customers who are already paying. Many AI startups count signed contracts as ARR even when the product hasn't been deployed and no money has changed hands. That's a different metric — CARR (contracted ARR) — and it's a much stretchier number. Stevenson claims he reviewed the financials of one enterprise AI company that was publicly reporting ARR five times higher than its actual revenue.
Specific cases
One VC told TechCrunch they'd seen companies with a 70% gap. One startup was counting a customer in its ARR who was still running a free year-long trial and could walk away without paying a cent. Another was publicly claiming $50M ARR against $42M in its financial documents. In both cases, investors saw the real numbers — but didn't push back publicly, because the big headline figure helps portfolio companies land clients and raise new rounds.
> "An unspoken pact between founders and VCs — don't discuss the gap with the press, use the bigger number for reach." — Scott Stevenson
Why this is structural
The pressure on numbers has become structural. General Catalyst CEO Hemant Taneja put it plainly back in September:
> "Growth from '1 to 3 to 9 to 27 million' impresses nobody anymore — you need to go '1 — 20 — 100'."
With that kind of bar, the temptation to juice your ARR is basically baked into the system.
A second flavor
There's also a second flavor of the same manipulation. Some startups take revenue from one good week or month, multiply by 52 (or 12), and call it "annualized ARR." The metric is called annualized run-rate revenue — just without the word "contracted."