OpenAI brings ChatGPT inside Excel to build spreadsheet models

Anyone who has ever spent a late night debugging a massive financial model knows the pain of a broken formula. Up until now, pasting spreadsheet data into chatbots was mostly an exercise in frustration and hallucinated numbers.

OpenAI just tackled this head-on with ChatGPT for Excel, an official add-in powered by GPT-5.4. Instead of jumping between tabs, the model now lives right inside your workbook, building formulas and stress-testing models through plain conversational prompts.

A spreadsheet copilot that actually audits

What sets this apart from typical generative AI features is that it does not just spit out static tables. ChatGPT for Excel operates natively within the spreadsheet grid, keeping your formulas, formatting, and structural assumptions intact.

It can trace formula chains across multiple sheets, explain why a KPI changed, and pinpoint logic errors when you inherit an undocumented sheet. On OpenAI's internal investment banking benchmark, GPT-5.4 Thinking hit an impressive 87.3% accuracy, a massive leap from the 43.7% scored by GPT-5.

OpenAI is also playing carefully with user trust. The add-in shows its working steps, highlights the exact cells it references, and asks for user permission before overwriting any cell. Because the math actually runs inside the Excel engine, analysts can audit every single step.

Beyond basic modeling, OpenAI rolled out integrations with heavy-hitting financial data providers including Moody's, Dow Jones Factiva, MSCI, Third Bridge, and MT Newswires, with FactSet on deck. Users can also hook up internal data using the Model Context Protocol. The beta rolled out to Enterprise, Business, Pro, and Plus users outside the EU, with Google Sheets support following shortly after.

Why it matters

We think this is one of OpenAI's smartest enterprise moves to date. Instead of trying to reinvent the spreadsheet, OpenAI met knowledge workers where they already live, respecting the native mechanics of Excel.

For analysts and associates, the immediate value is massive. Grunt tasks like building dynamic scenario tables or cross-referencing messy third-party templates can now happen in minutes instead of hours. The inclusion of audited formulas rather than hardcoded AI values solves the critical trust problem that kept finance teams away from LLMs.

For the industry, this creates a fascinating dynamic. OpenAI is directly encroaching on Microsoft Copilot's home turf. While Microsoft and OpenAI remain close partners, OpenAI is clearly building its own vertical solutions for high-value enterprise niches like Wall Street.

There is still a catch, though. OpenAI openly notes that edge-case formulas and convoluted workbook architectures may still need manual cleanup. In financial modeling, a single misplaced reference can ruin an entire forecast. We think teams should jump in to accelerate drafting and error-checking, but keep their sign-off processes strictly human for now.