Summary
OpenAI has added analytics in the ChatGPT Admin Console to show how teams use ChatGPT Work and Codex, including task categories, spending and Codex contributions to software development. The company also outlines a framework for comparing AI-assisted workflows with business outcomes such as time saved, quality and revenue.
OpenAI has added analytics to the ChatGPT Admin Console to help organisations connect ChatGPT Work and Codex usage with business outcomes. The tools combine usage and cost data with task insights and selected outcome metrics, giving administrators a way to examine where teams use AI, where support may be needed and which workflows should be measured more closely.
What the new analytics show
The Usage view brings together active users, credits and token usage across ChatGPT Work and Codex. Administrators can filter the data by group or user to identify areas with low adoption, concentrated spending or a need for additional training.
An Insights task classifier groups a sample of messages into work categories and tasks. These include software engineering activities such as feature development and code maintenance, as well as sales and revenue work such as account research and planning. An overview shows the distribution of work, while a detailed table breaks it down by credits, messages and active users.
Task details also show how credits are distributed across model, reasoning and speed settings. OpenAI presents this as a way to assess whether a workflow is using an appropriate configuration. For example, a routine brief could be tested with a faster or lower-cost setup, while the team compares the resulting quality and the time required for review and correction.
The same area shows the plugins and skills used for a task. OpenAI says this can help administrators identify access or training needs, maintain useful workflows and assign ownership for frequently used tools.
For software teams, the Outcomes view tracks Codex contributions to merged commits and lines of code alongside code-review activity. Engineering leaders can use group, user or repository filters and compare these trends with review time, defects and rework when assessing how an AI-assisted development process is performing.
The Admin plugin can compare adoption, spending and task trends and turn the findings into reports. The Admin API allows organisations to bring the data into their own dashboards and combine it with business-system information, such as displaying AI credit use alongside support-ticket resolution time.
From AI activity to business value
OpenAI’s framework treats usage data as a starting point rather than the final measure of value. Administrators and business owners are expected to select a workflow linked to a business priority, establish a baseline and compare results over a defined period.
The baseline can include how often a task is performed, how long it takes and what constitutes acceptable quality. The comparison should include time spent reviewing and correcting AI-generated work, so a shorter initial completion time is considered alongside the quality of the finished result.
The company illustrates the approach with sales account research. In its hypothetical example, 20 sellers prepare two account briefs per week and save three hours per brief over 46 weeks. That produces 5,520 hours of annual time savings. If half of that time becomes productive work valued at a fully loaded employee cost of $75 per hour, the estimated annual capacity value is $207,000. After assumed first-year costs of $60,000 for AI, setup, training and support, the example produces a 245% illustrative ROI.
OpenAI explicitly labels these figures as hypothetical. The calculation excludes possible effects such as higher win rates, larger deal sizes and other sales outcomes. The broader measurement process is to connect workflow improvements to outcomes such as delivery time, quality, lower costs, qualified opportunities, revenue or contribution margin.
Customer examples and rollout
OpenAI says 1Password uses Codex to build, review and test software and estimates a 553% ROI along with $0.8 million in annual engineering capacity value. The ATV Big Air Tour reported reducing weekly listing reviews from eight hours to one hour and inventory work from two to three days to two to three hours after using ChatGPT Work. Playco reported creating and testing playable game prototypes with GPT-6 Astra and making 50% fewer manual fixes than with its previous model.
To begin, OpenAI recommends opening Insights in the Admin Console, choosing a common task tied to a business priority, agreeing on a baseline and outcome with the relevant business owner, and setting a date to review progress. The company also says it does not train its models on an organisation’s business data by default.