OpenAI shows how finance teams use Codex for reporting and scenario planning

On May 12, 2026, OpenAI showed how Codex can be built into the daily work of finance teams, not just engineering workflows. In a new OpenAI Academy piece, the company lays out five practical scenarios, from monthly business reviews and board packs to model checks and scenario planning.

Where Codex saves time

The core idea is simple: finance staff should not have to start from a blank page or manually assemble the first version of documents before a meeting with the CFO or the board of directors. Codex is meant to work on top of artifacts teams already use: closing workbooks, revenue and expense dashboards, forecast updates, comments from metric owners and past presentations. The output is not an abstract chatbot answer but a draft of a specific file that the team can check, correct and pass along the process.

• Preparing a narrative for a monthly business review based on the period close, the forecast and owner comments
• Cleaning up and QA-ing a financial model before it goes to leadership or the board
• Updating a recurring CFO or board reporting pack with fresh data
• Building a variance bridge for revenue, margin, expenses, EBITDA and cash flow
• Recalculating base, downside and upside scenarios for the forecast and operating plan

Across all five cases, OpenAI stresses the same point: Codex takes real inputs from the current work cycle and speeds up the first assembly. That removes routine work but does not replace financial expertise. The team spends less time on mechanical consolidation and more on checking figures, wording conclusions and preparing questions for the business. This matters most where deadlines are tight and sources are scattered across spreadsheets and chats.

How the process works

In OpenAI's examples, Codex works not from a general description of the task but from a specific set of files and channels. For an MBR, these might be the close workbook, revenue dashboard, forecast update, the previous slide deck and owner notes. For scenario planning, they are the operating model, headcount plan, cash forecast, latest actuals and leadership notes. Importantly, for every material number the system is required to cite its source: a workbook tab, dashboard, tracker or owner note. That makes the result suitable for review, not just for reading.

OpenAI also emphasizes integration with a company's working tools. It lists Google Drive, SharePoint, Box, spreadsheets, documents, presentations, Slack, Teams, Gmail and Outlook Email as typical places where Codex can pull context from and return results to. In other words, this is not a demo on a synthetic dataset but a scenario in which a finance team asks Codex to update the May CFO pack, check the FY27 model or explain the April forecast-to-actual movement in terms of existing processes.

Where control is needed

Despite the automation, OpenAI spells out the limits. When Codex cleans a model, it can safely fix structure, references, formulas, period labels or sign conventions, but it should not quietly change business assumptions. If a number in a variance bridge or reporting pack lacks confirmation, it should be flagged as unconfirmed. For high-stakes reviews, the system should return not only the updated file but also a QA memo that prioritizes risks: which fixes were made, which assumptions remain disputed and which cells or tabs the finance owner should check.

In essence, OpenAI is proposing not a replacement for the FP&A team but an accelerator for its cycle. Codex handles the first pass through the data, finds gaps between sources and prepares the document structure well. But responsibility for interpretation, final wording, approving assumptions and sending materials to leadership stays with people. For finance this is critical: the cost of an error in a model or board pack is too high to leave the decision entirely to a machine.

What this means

Codex is increasingly positioned not as a tool only for developers but as a working environment for functions with heavy operations and many files. For finance teams this is a telling case: if AI can assemble MBRs, check models and quickly recalculate scenarios from real data, the next wave of AI adoption in companies will come not only through IT but also through the office of the CFO.