OpenAI: Frontier AI Adopters Now Generate 8.3x the Output Per User of Typical Enterprise Customers

OpenAI’s “Enterprise Signals” usage data puts a number on something consultants have mostly argued from anecdote: frontier-usage companies — the top 10% of customers by AI adoption depth — now generate 8.3 times the output tokens per active user of a typical enterprise customer, up sharply from a 2.6x gap measured in January 2026. The adoption gap between AI leaders and laggards isn’t closing. It’s compounding.

The pattern OpenAI extracts from three case studies — Basis, Clay, and Exa Labs — is specific enough to be useful rather than aspirational: pick one consequential workflow, define the outcome and how it’s measured, write an explicit “job description” for the agent handling it, build human escalation paths around the agent rather than assuming full autonomy, and make experiment results visible and reusable across teams instead of siloed in whichever group ran the pilot.

ABC Legal is a concrete instance of that pattern working at a mid-size scale that will feel more familiar to most Banyan clients than a frontier AI lab’s own customers. A 15-person steering committee at the 1,100-employee legal-delivery company built working agents within a week of adopting Claude Managed Agents, and scaled to more than 50 agents in production within a month — covering service of process, eFiling, compliance, and finance. By July, roughly 310 employees used the platform daily, agents covering certain tasks had already cut costs by about 50% before any real optimization, and a compliance-review agent nicknamed “Charvis” now agrees with the human compliance team’s judgment 98% of the time.

The throughline across both stories is sequencing, not tooling: OpenAI’s own framework and ABC Legal’s rollout both start by naming one workflow and its owner before anyone touches automation. For a consulting practice arguing that transformation succeeds or fails on operating discipline rather than model choice, the 8.3x number is the argument’s clock — it says the cost of waiting to define that discipline is now compounding, not holding steady.