Expanded admin tools relate AI use, costs and tasks to operational results, while OpenAI warns that activity metrics cannot prove value by themselves.
OpenAI announced expanded administrative analytics for ChatGPT Work and Codex on September 16, aiming to help organizations relate AI use and expenditure to measurable business results. The ChatGPT Admin Console brings usage, cost, task and outcome information into one place. OpenAI cautions, however, that activity figures are only a starting point and cannot demonstrate economic value without operational context.
Administrators can use the Usage section to examine active-user counts, credits and token consumption, narrowing the data by team or employee. Insights applies a classifier to a sample of messages to group them by task and use case. Further breakdowns distinguish credit use by model, reasoning option and speed setting. Plugin and Skills sections indicate which supporting tools are being used and may reveal needs involving access, ownership or training.
The engineering-oriented Outcomes section relates Codex contributions to merged commits, lines of code and review activity, with filters for repositories, users and groups. OpenAI says these contribution patterns should be considered together with review duration, defects and repeated work. Its qualification matters because output volume on its own does not establish better engineering performance.
An Admin plugin within ChatGPT Work can compare adoption, spending and task patterns, then produce materials including reports for leadership. Organizations can also use the Admin API to feed the information into their own reporting systems and place it alongside operational data, such as the time required to resolve tickets.
OpenAI’s proposed measurement process starts by defining an outcome that matters and recording its initial level. The organization then observes change over a set interval, includes the effort needed to inspect and correct AI-assisted work, and weighs the resulting benefit against AI, deployment, training and support expenses. A hypothetical sales calculation produces a 245% first-year return, based on $207,000 of estimated capacity value and $60,000 of costs.
That calculation is illustrative rather than evidence of a typical return: OpenAI says every figure is hypothetical. It assigns value only to assumed productive capacity and leaves out potential effects such as improved win rates, larger transactions or other sales results. The screenshots likewise contain demo data, while the classifier assesses only a message sample. A credible evaluation would therefore still require workflow knowledge and outcome measurements from the relevant business owners.
OpenAI states that organizational business data is not used to train its models by default. The announcement does not identify plan-level restrictions, a staged release schedule or access differences among ChatGPT Work offerings. It therefore does not confirm that every organization can use all of the described capabilities at the same time.
| Section | What it shows | How to use it |
|---|---|---|
| Usage | Active users, credits and tokens | Filter by groups and employees |
| Insights | A sample of messages grouped by use case and task | Identify patterns in AI use |
| Outcomes | Codex commits, lines of code and review activity | Compare with development quality and speed |
| Plugin and Skills | Use of supporting tools | Identify needs involving access, ownership and training |
Sources
Originally published by AI Feed: OpenAI Adds Business-Outcome Analytics for ChatGPT Work and Codex.
Top comments (0)