ChatGPT Admin Analytics Link AI Usage to Rollout Decisions

As AI use expands across an organization, the administrative question is not simply who has adopted the tools. It is where that use is producing value, where spending is accumulating, and where additional access or training is justified.

The Admin Console’s Usage view is positioned as the starting point for that assessment. It shows daily active users alongside credits and messages, with filters for groups and individual users. In the displayed organization-wide view for August 12 through September 10, the dashboard reports 7.4 million credits and 6.7 million messages. An administrator can narrow that activity to departments including Engineering, Sales, Finance, Customer Support, Marketing, Operations, Product & Design, IT, and HR & recruiting.

That makes usage growth legible by team, but the intended use is operational: determine where to direct enablement, allocate budget, or broaden access. A rise in credits is not presented as an outcome by itself. It is a signal to investigate what work is being done and whether the configuration fits that work.

The Insights view classifies ChatGPT Work and Codex activity into predefined use-case categories and then into specific tasks. The categories displayed include software engineering, sales and revenue, customer service and support, research and planning, design and media, strategy and business operations, and marketing.

Administrators can filter these insights by group to see how different teams use AI. A Sales-filtered example breaks the sales-and-revenue category into account research and planning, prospecting and qualified pipeline creation, customer onboarding and adoption, sales enablement and coaching, and remaining tasks.

The task-level detail is where configuration decisions enter. For account research and planning, the console exposes breakdowns for model, reasoning level, and speed. The stated purpose is to assess whether the setup matches the task and identify cases where a faster, lower-cost option may be sufficient.

Plugin activity provides a second diagnostic. In the account-research example, the plugin leaderboard shows 308,100 Google Drive invocations, 306,500 Salesforce invocations, and 76,800 Slack invocations. The speaker’s expectation is that Salesforce should supply relevant account context for this workflow. If it is not being used for the task, that gap becomes a target for training rather than merely a utilization statistic.

The same view surfaces reusable skills, including Account Brief, Customer Handoff, Renewal, and Account Plan. Skills are described as a way to make workflows repeatable and consistent with company practices. Their observed use can therefore inform which skills an administrator should maintain and which should be distributed more broadly.

For coding work, the console presents a more direct set of measures: Codex contributions to commits, merged code, and code review. OpenAI frames these measures as evidence an administrator can use with engineering leaders when evaluating a rollout and deciding where to extend access. They are signals of Codex’s presence in the engineering workflow, rather than a complete measure of engineering value.

The Outcomes dashboard shown in the demonstration reports that Codex contributed to 90% of 120,000 commits. It also shows 108,000 commits merged with Codex, 15.3 million lines of code merged with Codex—85% of 18 million total lines—and 42,000 pull requests reviewed, with 64,600 issues found.

Following these measures over time and across developers gives administrators material for a more specific discussion: where Codex is contributing to commits, merged code, and reviews, and where expanded access may warrant consideration.

The ChatGPT Admin plugin brings these analytics into a single conversation. An administrator can explore adoption or spend, combine views, and ask follow-up questions while investigating an AI rollout.

In the example request, the administrator asks for a monthly AI rollout review: trends in Work and Codex usage and credits by team over the previous 30 days; each team’s leading use cases and task breakdown; and, for Engineering, Codex contributions to commits and code review. The request also asks the system to identify the largest changes and the next areas of focus.

The resulting analysis displays team credit totals including 3.1 million Codex credits and 934,200 Work credits for Engineering; 526,100 Work credits for Sales; 408,000 Work credits and 22,100 Codex credits for Product & Design; 360,000 Work credits for Marketing; 201,000 for Finance; and 198,000 Work credits plus 1,400 Codex credits for IT. It identifies software engineering as Engineering’s top task area, at 3.2 million exchanges, and sales and revenue as Sales’ leading category, at 1.2 million exchanges.

The follow-up instruction is practical: turn the analysis into a short leadership deck with charts, key takeaways, and recommended next steps, then share it in the AI rollout Slack channel. The displayed result says it created six leadership slides and two appendix slides, shared them in #ai-rollout, and made an editable PowerPoint available.

The generated deck characterizes the period as one in which Work and Codex are both growing, with 45% growth in credit usage to 6.81 million credits. It says Engineering accounts for half of the additional credits, that each team has a clear primary use case, and that Codex is embedded in engineering work, citing the 90% commit-contribution figure.