Copilot in Excel just leveled up for finance pros

Finance professionals are notoriously hard to impress with AI. In corporate finance and investment banking, a model cannot just look right; every single formula must trace back to a verified filing, and every line item must withstand an audit. General-purpose chatbots that hallucinate numbers simply have no place in a quarterly close.

Microsoft rolled out a heavy-duty update for Copilot in Excel designed specifically for financial workflows. Instead of pitching generic chat magic, the company is treating AI like a junior analyst who needs rigid templates, verified institutional feeds, and an inescapable audit trail.

Hardcoding models with SKILL.md

The centerpiece of the update is a new feature called Skills.

Instead of prompting Copilot from scratch every month to run variance analyses or build a discounted cash flow (DCF) model, finance teams can now codify repeatable workflows into an open-standard SKILL.md file saved directly in OneDrive. Copilot reads these markdown guidelines and executes the financial modeling step-by-step, adhering to your firm's specific formatting, formula architecture, and structure.

Microsoft built a starter library of finance skills and partnered with the Financial Modeling Institute (FMI) to benchmark Copilot against real-world modeling exams. Third-party providers like LSEG, Ramp, and Velixo are scheduled to bring their own packaged workflows to the Microsoft 365 Admin Center in Q3 2026.

Copilot also gains Personalization and workbook-level rules, meaning style guides and naming conventions can live right inside a dedicated tab in the workbook itself.

Wall Street data pipes and mandatory pre-flight plans

A financial model is only as credible as its underlying data. Following integrations with LSEG and Moody's in May, Microsoft added connectors for CB Insights, Daloopa, FactSet, Morningstar, PitchBook, and S&P Global via Kensho's Deterministic Retrieval.

Analysts can now pull SEC-filed operating metrics or private transaction multiples straight into formulas without manual CSV downloads. FactSet is slated for general availability in July, though users will still need active enterprise subscriptions with each data vendor.

To address the risk of AI wrecking live workbooks, Microsoft introduced Plan with Copilot.

Before executing multi-step updates, Copilot presents a blueprint detailing which ranges, sheets, and assumptions it intends to tweak, pausing for user approval. Once applied, every edit is attributed directly to Copilot inside Excel's native Show Changes pane, complete with direct cell links.

Why it matters

We think this is one of Microsoft's sharpest enterprise AI updates to date. The real barrier to AI adoption in finance was never a lack of raw math ability; it was the utter lack of defensibility.

For investment bankers, corporate FP&A teams, and accounting departments, the value proposition is immediate. Junior staff can automate recurring month-end grunt work and comparable company tables while maintaining full lineage. That said, the real cost of this setup is steep: on top of the Microsoft 365 Copilot license, extracting value from live connectors requires already holding brutal enterprise subscriptions from FactSet or PitchBook.

For the AI ecosystem, this update puts immense pressure on standalone financial AI startups. Many new ventures spent the last two years building web-based canvas tools to generate balance sheets. By turning Excel into an automated orchestrator connected to institutional feeds, Microsoft meets finance teams where they already live, making third-party wrappers a tough sell.

The lingering challenge lies in trust. An attributed change list in the sidebar is helpful, but when multi-sheet formulas interact with complex macro assumptions, human review remains non-negotiable. If teams treat these markdown skills as autonomous magic rather than rigid drafts, catching subtle hallucinations deep within an operating model could actually become harder, not easier.