Same models. Different depth of adoption.

What the data says

The frontier-to-typical output-token ratio rose from 2.6× in January to 8.3× in June 2026. Frontier firms were the top 10% of usage each month; typical firms were the middle 10%. Weekly plugin use was 21% versus 9%, and skills use 19% versus 3%.

These figures describe how deeply different groups use the tools. They do not establish an equivalent increase in revenue, profit, or completed work. The publisher explicitly warns that token volume is an imperfect value measure.

Our interpretation: make useful use easier

A practical response is not to ask employees to generate more tokens. It is to remove the friction between a real task and a useful result.

Start with a recurring job: a quote, a weekly report, or a customer follow-up. Give the team a working brief, approved source material, the necessary tool access, and a clear review boundary. Then let the person who owns the task test it against ordinary work—not just a demonstration prepared by the implementation team.

This is where upskilling becomes operational. A colleague learns to specify the task, spot a missing source, and check the proposed action. The company gains a method it can improve rather than another account people are expected to explore on their own.

Measure the work, not the activity

For a first workflow, establish four measures before rollout:

• Time from request to a reviewable output.
• Time spent checking and correcting that output.
• Errors or missing information that reach the next person.
• Whether the team continues using the workflow after the initial training.

More usage can mean deeper delegation. It can also mean a workflow is inefficient or that someone is repeatedly correcting it. The business measure has to sit beside the usage measure.

Limits worth keeping in the conversation

This is observational data from OpenAI’s enterprise customers, not a representative census of every business. The usage cohorts are defined monthly, so the comparison is not necessarily a fixed group tracked over time. It does not isolate the causal effect of training, plugins, or skills.

For a portfolio team, the sensible comparison is initially within each business: the same recurring job before and after implementation. Repeat the method across companies, but keep their data, permissions, and operating constraints separate.