The Bill Came Due

The Bill Came Due

The AI conversation turned to the invoice as much as the technology.

Metrics that Measure Up put numbers on it. Token spend across Ramp's customer base is up 13x since January 2025. Only 27% of enterprise executives say their AI investments have met ROI expectations. And the value, where it exists, hides. Ray Reich and Peter Buchanan call it the measurement paradox: individuals are demonstrably faster, but it does not reach the financial statements. Drop the cost of a legal document from $400 to $5 and all the dashboard shows is the token bill. One executive running a thousand Claude Code engineers put the productivity math bluntly: "one plus one plus one plus one equals one and a half, not four." By an estimate they cite, just 18% of tokens spent on coding become product that reaches a user. Even Meta's CTO told staff that "AI motion is not progress."

The trap is that usage-based pricing now punishes the gap. GitHub, Anthropic, OpenAI, Google, and Salesforce all moved enterprises off fixed subscriptions this year, so the meter runs while the ROI stays theoretical. Agents spawn subagents and loop, so spend is multiplicative, not linear. One company kept paying for an agent working on behalf of an employee who had already left.

The constructive counter came from Rahul Rekhi, president of Rogo, on Run the Numbers. His frame: stop token-maxing, start ROI-maxing. Finance is the "goldilocks zone" for applied AI because willingness to pay for intelligence is high, but the winners will be the organizations that measure output against cost, business unit by business unit. He thinks the gap between firms that spend tokens well and firms that don't will run to hundreds of millions of dollars. And the people who close it are not the pure technologists. They are domain experts. Rogo's fastest-growing role is the "forward-deployed banker," someone who knows the work and can aim the model at it.

The throughline for the week: the hype phase is over and the accounting phase has started. Nobody serious is saying AI doesn't work. They are saying prove it, and most companies still can't.

Listen:

• Metrics that Measure Up — AI Math Is Not Adding Up: Where Is the ROI? (Jul 14). The numbers, and the "dark output" problem.
• Run the Numbers — Why "Token Maxing" Is a Trap, Rogo's Rahul Rekhi (Jul 13). What measuring ROI actually looks like.