Summary: OpenAI’s May 12, 2026 Academy article describes finance workflows that moved from the former Codex app to ChatGPT Work. Teams can start from close workbooks, dashboards, forecasts, prior reviews, and notes to create review-ready drafts for monthly reviews, reporting, variance analysis, and planning. The article provides no quantitative results, so its claims are directional rather than measured outcomes.
From Codex to ChatGPT Work: What Actually Shifted
Dated May 12, 2026, the Academy article “Finance workflows with ChatGPT Work” from OpenAI describes how finance teams can take the materials they already work with—close workbooks, revenue and expense dashboards, forecast updates, prior monthly business reviews, and owner notes—and use ChatGPT Work to turn them into review-ready deliverables. The notable change is one of location: these workflows used to live in the former Codex app, and they now run directly in ChatGPT Work at chatgpt.com or in the ChatGPT desktop app. Readers are also told that the accompanying webinar was recorded back when the workflows still lived in Codex; they can now follow along in ChatGPT Work instead.
For product builders, the significance of this move is directional: a standalone coding assistant has grown into a more integrated work environment. ChatGPT Work is presented as requiring no coding, which puts it within reach of finance professionals who are not developers. And the instruction is to start from the work products a team already has, not to make users adopt a new data structure or workflow.
How It Works: Existing Materials Set the Starting Point
At its core, the method is straightforward. Rather than opening a blank document, finance teams hand ChatGPT Work the very materials they already rely on. As the OpenAI article lays out, those inputs are:
- Close workbooks
- Revenue and expense dashboards
- Forecast updates
- Prior monthly business reviews
- Owner notes
From that context, ChatGPT Work produces concrete assets the team can review, refine, and share. The point is to cut the time spent assembling a first draft and redirect it toward shaping the story, validating the numbers, and getting ready for the next decision.
What stands out about this approach is what it leaves out: no demand that users change their habits, and no new system for organizing data that they must learn. The tool comes to users where they already are, and that keeps the adoption barrier low. Product teams should take this as a design principle: work with the workflows people have instead of demanding new ones.
Where Finance Teams Put It to Work
Several headline use cases for finance teams appear in the OpenAI article, but detailed step-by-step instructions do not. The use cases named are:
- Monthly business reviews
- Reporting
- Variance analysis
- Planning
All of them are routine, repetitive jobs where data from multiple sources has to be assembled into one coherent narrative. Assembly is the part ChatGPT Work takes on; analysis and communication stay with people. The article also points readers to a collection of use cases at learn.chatgpt.com/use-cases/collections/finance and to a solutions page for finance teams at openai.com/business/solutions/finance/.
Also worth flagging: the article offers no quantitative results—no hours saved, no accuracy improvements. What it presents is a direction, not hard metrics. For product builders this is a useful reminder in its own right: an AI feature does not need a benchmark to be valuable, because sometimes the value lies in making a different workflow possible.
The Caveats, and the Trade-off Behind Them
The tone is optimistic, yet the article is careful about its limits. The most visible one: the webinar was recorded in the former Codex app, and the workflows have since been migrated over to ChatGPT Work. Some of the details shown in the webinar may therefore not line up exactly with the current interface, and users are advised to follow along in ChatGPT Work directly.
A second limitation is the absence of concrete examples or case studies with measurable outcomes. No company is cited as having put ChatGPT Work to successful use in finance, and there are no before-and-after comparisons either. Missing evidence is not automatically a red flag, but it does mean readers should take the claims as directional rather than proven.
For product builders, a trade-off sits inside all this: the tool is built to be flexible and adaptable, and predictability is what that flexibility costs. How good the output turns out will vary with the quality and structure of the inputs. Getting the best results will most likely take iterating—refining the prompts and the inputs until they land.
Lessons for Product Builders
Perhaps the most striking feature of the OpenAI article is how often the term “review-ready” appears. Nothing is framed as a final answer; everything is an asset the team can review, refine, and share. Behind the wording sits a crucial design philosophy: the AI exists to support human judgment, not to replace it. People stay in the loop, and what the AI produces counts as a draft that still requires validation.
This frames a key question for product builders: is your output a draft that needs review, or a finished product that can go straight out the door? Professional teams are likelier to accept the first, since it preserves human control and gives errors a chance to be caught during review. The second is faster, but one AI mistake can start eroding trust.
A second lesson is the value of starting from inputs that already exist. Because users can upload their own workbooks and dashboards, ChatGPT Work sidesteps the friction of asking anyone to reformat their data. As an onboarding strategy that is powerful: make the tool work with what users already have, and the cost of trying it drops.
First Steps: One Concrete Way to Try It
Product builders and finance professionals alike can run the practical test the OpenAI article points toward: take the most repetitive, most time-consuming slice of your monthly reporting process, pull together the relevant workbooks and dashboards, and look at the kind of draft ChatGPT Work hands back. Perfection is not the bar. The question is whether it saves you time on assembly, so you can move faster into substantive discussion.
The use case collection and the finance solutions page are both linked in the article, and either makes a reasonable entry point. What it does not supply is specific prompts or code examples, so experimenting on your own is up to you. Treat whatever comes back as a starting point rather than a final product, and iterate against your team’s standards.
The bigger picture: ChatGPT Work is part of a shift that brings AI assistants within reach of non-technical teams. With its focus on review-ready outputs and its habit of starting from existing work inputs, it doubles as a template for designing AI tools that augment human expertise rather than replace it. The lesson for product builders is plain: design for review, not for automation.