OpenAI Publishes Guide to Using Admin Console Analytics Tools

The first screen covers usage. According to the announcement, the usage view shows active users, credits, and token consumption across ChatGPT Work and Codex together, and filtering by group or user surfaces where adoption is lagging, giving admins a basis for talking to team leads about onboarding workflows and training needs. OpenAI noted that the numbers shown on screen are all demo data, not real figures.

The second is the task classifier. In the insights screen, the classifier groups a sample of messages into use cases and tasks to show what AI is actually helping with. Software engineering covers feature development and code maintenance; sales and revenue cover account research and planning. An overview tab shows the task breakdown at a glance, and a use-case tab tables out credits, messages, and active users by task. In the sales team OpenAI used as an example, account research and planning accounted for the largest share of credit use.

The third area is where training is needed. In the task detail screen, the model, reasoning, and speed fields show the credit share for each setting, letting admins check whether a setting fits the work and target model-selection training accordingly. The announcement suggested that for routine briefings, it's worth testing a faster or cheaper setting and comparing quality against review time. A plugin leaderboard and skills screen show which tools are supporting the work; low usage of a relevant plugin can point to an access or training gap, and the company said frequently used skills need an owner and regular updates.

Datadog AI product manager Bharadwaj Thanikella said in the announcement, "OpenAI's analytics help our teams understand how they're using AI, and that becomes the foundation for the guidelines and policies we build going forward." He added that Datadog already uses OpenAI's task classification in Agent Console, its AI agent monitoring product, and that getting that data directly from OpenAI is "a more reliable way to deliver those insights" as customers' AI usage keeps growing.

On the coding side, a performance screen handles the numbers. According to the announcement, the performance view shows Codex's share of merged commits and lines of code alongside code review activity, with filters by group, user, and repository to decide where to expand access and which teams to help. The company explained that if Codex's share of merged code is rising, engineering leads can weigh that trend against review time, defects, and rework to judge whether the team is shipping software more effectively.

There are also two tools for turning the numbers into reports. ChatGPT Work's admin plugin compares adoption, spending, and tasks to build reports for budget and rollout decisions, and can even produce an executive presentation complete with charts, key findings, and next-step recommendations. The Admin API lets teams automate reports in their own dashboards and merge the data with business-system data — for example, placing credit usage next to ticket-resolution time on a support dashboard.

OpenAI was clear that the analytics numbers alone don't prove value. Usage and task data are only a starting point for admins to work through value together with business leaders, who need to add the context on what actually changed in the workflow, whether outcomes improved, and how much that improvement is worth. The announcement laid out five questions: what you want to improve, what the current process looks like, what changes with AI, what that makes possible for the team, and whether the payoff is worth the investment. It also cautioned to include the time spent reviewing and fixing AI output when making that comparison.

A hypothetical calculation was also included. If 20 sales reps each write two account briefings a week and save three hours per briefing, that adds up to 5,520 hours saved over 46 weeks; assuming half of that time goes toward productive work, and plugging in a labor rate and first-year adoption cost, the example works out to a 245% return on investment. OpenAI stated plainly that these figures are all assumptions, and that they exclude sales outcomes like higher win rates or bigger deal sizes.

Three customer case studies were included as well. According to the announcement, 1Password estimated a 553% return on investment from building, reviewing, and testing software with Codex, and ATV Big Air Tour used ChatGPT Work to cut its event-listing checks from 8 hours a week to 1 hour, and inventory work from 2-3 days down to 2-3 hours. Gaming company Playco reported that using GPT-6 Astra through the OpenAI API let it build three differently themed prototypes on a single foundation, cutting manual fixes by 50% compared with the previous model. METAL has previously reported on how ATV Big Air Tour turned three days of inventory cleanup into three hours.

This guide sits alongside the string of enterprise products OpenAI has rolled out recently. METAL has reported on OpenAI adding a data-querying agent to ChatGPT Work, and also covered its launch of a ChatGPT built specifically for financial firms. In effect, OpenAI has written the next chapter after the sale itself — the part where a company has to justify the spend.

The full announcement METAL reviewed came with an 8-minute-29-second audio version, credited to OpenAI. The recommended starting steps are short: open Insights in the admin console, pick one common task that supports a business priority, agree with a business leader on a baseline and the metrics to measure, and set a date to review progress. The announcement closes with a line noting that, by default, the company does not train its models on an organization's business data. The next question in AI adoption isn't how much you're spending, but where that spending pays off — and OpenAI chose to let admins work out that answer for themselves.