A Guth Labs publication ↗
Back to News

Ramp’s AI spending update shows why usage and revenue need separate scorecards

Ramp’s November11 economic update says observed business AI spending rose 1.65% in October. Its newly introduced department breakdown puts engineering at 72% of classified spending and customer support/success at 13.3%, up from 0.8% in June. These are results from Ramp’s observed customer data, not a census of the entire AI industry. Dated primary update

The payment-based index draws on more than 85,000 American businesses. Ramp defines adoption through observed payments, so free tools and employees’ personal accounts can be missed. Its token metrics use a separate Token Spend Management sample and should not be treated as representing all spending on Ramp. The dashboard also mixes monthly and weekly series; those windows are not interchangeable. Dashboard and methodology

The Decoder independently reports on the update and emphasizes the sample limitations. Its coverage is commentary on Ramp’s numbers, not a separate replication. We therefore do not turn the concentration figures into claims about the complete revenues of model providers. Independent coverage

Analysis

A fleet can perform more work while its model bill grows slowly, particularly when teams change model mix, compress context or reuse results. Conversely, an increasing bill does not show that useful work increased. These data cannot tell us which explanation applies to a particular company.

The practical response is to keep separate scorecards for paid spend, usage and successful tasks. Department labels also need stable definitions before comparing periods. Support becoming a larger share of a sample is a reason to examine workflow economics, not proof that support agents now have better returns than coding agents.

Sources and citations

Revision history

Published .