OpenAI details how to turn ChatGPT Work and Codex usage into ROI

OpenAI has published a guide explaining how administrators and business leaders can use Analytics in the ChatGPT Admin Console to move past raw usage and spend numbers and connect ChatGPT Work and Codex adoption to measurable business value. The guide's starting point is that usage and cost figures alone do not show what people actually accomplish with the tools, so admins also need to see the tasks being done and the outcomes that follow.

The Admin Console brings together several views for this. The Usage view shows active users, credits and token usage across ChatGPT Work and Codex, filterable by group or user, so admins can spot where adoption is low and review starting workflows or training needs with a team owner. The task classifier in Insights groups a sample of messages into use cases and tasks: software engineering, for instance, covers feature development and code maintenance, while sales and revenue covers account research and planning. An Overview tab shows the mix of work at a glance, and a Use cases tab gives a detailed breakdown table. Within a task, Models, Reasoning and Speed breakdowns show each setting's share of credits, which can point admins toward testing a routine task on a faster or lower-cost setup. The Plugin leaderboard and Skills view show which tools support a task, so low use of a relevant plugin can flag an access or training gap. For Codex specifically, the Outcomes view tracks its contribution to merged commits and lines of code alongside code-review activity, letting engineering leaders compare a growing Codex share of merged code against review time, defects and rework. The Admin plugin inside ChatGPT Work can turn these findings into reports, including a leadership deck with charts and recommendations, and the Admin API lets teams pull the same analytics into their own dashboards next to other business-system data, such as ticket resolution time.

The guide frames the actual work of proving value as a conversation between admins and business owners, built around five questions: what outcome the team wants to improve, what the process looks like today (frequency, duration, what a good result looks like), what changes once AI is introduced, what the resulting time or quality gains make possible for the team, and whether the benefit is worth what is being spent on the tool, setup and support.

OpenAI walks through an illustrative ROI calculation for a sales team: 20 sellers, each preparing two account briefs a week and saving three hours per brief with AI, saves 5,520 hours over a 46-week year. Assuming half of that time converts into productive work valued at a fully loaded employee cost of $75 an hour, the estimated annual capacity value is $207,000. Against an assumed first-year cost of $60,000 for the AI subscription, setup, training and ongoing support, that works out to a 245% illustrative ROI. OpenAI is explicit that every figure in this example is hypothetical, that the ROI reflects only estimated capacity value, and that it excludes any additional benefit from a higher win rate or larger deal sizes.

The guide also cites three customers' own reported results. 1Password says it uses Codex to build, review and test software, and estimates a 553% ROI and $0.8 million in annual engineering capacity value from it. ATV Big Air Tour uses ChatGPT Work to check event listings, plan inventory and improve its website's visibility, and reports that weekly listing reviews fell from eight hours to one hour, and inventory work from two or three days to two or three hours. Playco uses GPT-6 Astra through OpenAI's API to build and test playable game prototypes, and says it built three themed prototypes from a single foundation with 50% fewer manual fixes than it saw with the previous model. None of these figures is described as independently audited, and no timeframe is given for how long any of the three have been using the tools.

The guide closes with a concrete next step: open Insights in the Admin Console, pick a common task tied to a business priority, review it with a business owner, agree on a baseline and the outcome to measure, and set a date to check progress.

Key facts

OpenAI's Admin Console Analytics combines a Usage view, a task classifier in Insights, and a Codex-specific Outcomes view across ChatGPT Work and Codex.
Illustrative example: 20 sellers saving 3 hours per brief on 2 briefs a week over 46 weeks yields 5,520 hours saved a year, worth an estimated $207,000 in capacity value at 50% utilization and $75 an hour, against an assumed $60,000 first-year cost, for a 245% illustrative ROI.
1Password reports a 553% ROI and $0.8 million in annual engineering capacity value from using Codex to build, review and test software.
ATV Big Air Tour says ChatGPT Work cut weekly listing-review time from eight hours to one hour, and inventory work from two or three days to two or three hours.
Playco says it built three themed game prototypes from one foundation using GPT-6 Astra via OpenAI's API, with 50% fewer manual fixes than the previous model.

Why it matters

Enterprise AI spend is easy to track and hard to justify: usage and cost dashboards show that people are using ChatGPT Work or Codex, not what that use is worth. OpenAI's guide targets exactly that gap, packaging usage data, a task-level breakdown and, for Codex, a code-outcomes view into a single console so admins can move from "adoption is up" to a specific claim about time, quality or cost.

Who it affects

The guide is written for admins and business leaders running ChatGPT Work and Codex inside a company, plus the business owners of individual teams (the example used throughout is a sales team) who supply the context, baseline and target outcome that usage data alone cannot. Engineering leaders get a dedicated angle through the Outcomes view, which ties Codex activity to merged code and review load.

How to use it

Start in the Usage view to see where adoption and spend are concentrated, then open the task classifier in Insights (Overview for a quick mix, Use cases for a detailed table) to see what kind of work AI is doing for a given group. The Models, Reasoning and Speed breakdown inside a task shows whether the setup fits the work, and the Plugin leaderboard and Skills view show which tools are actually supporting it. For code, the Outcomes view adds Codex's share of merged commits and lines of code alongside review activity. The Admin plugin in ChatGPT Work can turn any of this into a report or leadership deck, and the Admin API exposes the same data for a team's own dashboards. OpenAI frames the actual value case as five questions to work through with a business owner: what to improve, today's baseline, what changes with AI, what that enables, and whether the benefit is worth the spend.

How solid is it

The sales-team ROI figure (245%) is explicitly labeled illustrative: every input, the 3-hours-saved-per-brief, the $75 hourly cost, the $60,000 first-year spend, is an assumption OpenAI chose for the example, not a measured result. The three customer cases are the companies' own self-reported or self-estimated figures (1Password's 553% ROI and $0.8 million capacity value, ATV Big Air Tour's time reductions, Playco's 50% fewer fixes); none is described as independently audited, and the article gives no timeframe for how long any of them has used the tools, no byline and no publication date.

Risks and caveats

This is OpenAI publishing a guide to its own admin tooling on its own site, using its own customers as evidence, so the examples are naturally selected to look good. OpenAI itself caveats the illustrative ROI as excluding upside from a higher win rate or larger deal sizes, which cuts both ways: the real number could be higher or the 50% productive-time assumption could be optimistic. No pricing, discount or contract terms for ChatGPT Work, Codex or the Admin API are given, so the $60,000 cost assumption in the illustrative example should not be read as a real price point.