ARR Doesn't Mean What It Used To
VCs are calling out the metric inflation game: ARR has splintered into so many variants (daily revenue × 365, quarterly × 4, GMV disguised as recurring revenue) that the gold standard for SaaS valuation has become meaningless, and the same gaming pattern is spreading to every other startup metric.
Read Original Summary used for search
TLDR
• ARR definitions have fractured—some founders annualize daily revenue, others use quarterly × 4, some conflate it with GMV, making cross-company comparisons impossible
• CARR (contracted ARR) worries investors more because signed contracts often don't convert—customers change minds, implementations drag, and the gap between contracted and realized revenue is unpredictable
• AI companies game gross margin by pushing field deployment engineering costs into OpEx instead of COGS, showing "70%+ margins" that collapse when you properly allocate scaling costs
• The old VC playbook of benchmarking growth/NRR/GRR/burn is broken—AI's growth curves and consumption pricing models make historical comps useless, forcing qualitative judgment over metrics-based investing
• Physical AI companies show the same pattern with different metrics: "95% autonomy" measured in controlled demos and non-binding LOIs counted as pipeline
In Detail
The metric that's supposed to anchor SaaS valuations has quietly stopped meaning one thing. VCs report seeing ARR calculated as quarterly revenue × 4, daily revenue × 365, GMV rebranded as recurring revenue, and everything in between. The problem isn't just definitional sloppiness—it's that founders are optimizing for the highest number they can defend rather than the most accurate representation of their business. One VC's advice: put your ARR definition on slide one, because explaining your math builds more credibility than any growth rate.
CARR (contracted or committed ARR) triggers even more skepticism. Eight-figure contracts look impressive until you realize they're contingent on milestone delivery, customer acceptance, and the vendor's ability to scale complex work. For AI companies selling training data to frontier labs, the gap between signed bookings and realizable revenue can be massive—recognition depends on evolving definitions of what counts as "high-value data" as model capabilities change. Gross margin faces similar gaming: AI-native companies with field deployment engineer motions push those costs below the line into OpEx, showing clean 70%+ margins that only hold up if customers don't actually use the product much. NRR gets inflated by converting small POCs to contracted ARR and counting that as expansion.
The pattern extends beyond SaaS. Physical AI companies quote "95% task completion, fully autonomous" measured in controlled pilots with clean lighting and an engineer standing by, then present it as steady-state performance. Non-binding LOIs and MOUs show up as pipeline despite carrying no penalty for walking away. The underlying issue is the same across all metrics: numbers generated under ideal conditions getting quoted as if those conditions are permanent. The old VC playbook of benchmarking metrics against comps is broken—AI's unprecedented growth rates and consumption pricing models make historical comparisons useless, forcing investors toward qualitative judgment about durability rather than metrics-based heuristics.