Copilot in Excel Finance Connectors, Skills, and Audit Controls Explained

Reusable skills, eight named finance connectors, and new audit controls arrived this week. Here's what FP&A and corporate finance teams need to evaluate before building on any of it.

Microsoft launched a broad set of Copilot in Excel finance connectors and workflow features yesterday, targeting FP&A analysts, corporate finance teams, and investment professionals who run critical work inside spreadsheets. The release addresses three things at once: repeatability, trusted data access, and auditability. Taken together, it's arguably Microsoft's most finance-focused Excel Copilot update to date, though the company has not published benchmarks or production case studies to substantiate the capabilities.

The release is organized around three pillars Microsoft calls repeatable skills, trusted data, and traceability. Skills let teams encode their own workflows so Copilot runs them consistently rather than improvising each time. Six new Copilot in Excel finance connectors from CB Insights, Daloopa, FactSet, Morningstar, PitchBook, and S&P Global extend a roster that already includes LSEG and Moody's, bringing the total named finance connectors to eight. A new attribution feature logs what Copilot changed in a workbook, giving reviewers a paper trail (Microsoft 365 Blog, yesterday).

The rollout is staggered. Pre-built skills and connectors are generally available today for Microsoft 365 Copilot customers. Custom skills are in Insider preview, with broad rollout expected next month according to Microsoft's timeline. Partner-built skills via the Microsoft Marketplace are slated for Q3 2026 (Thurrott, yesterday).

Reusable skills: the operating model shift

The most consequential part of this release isn't a feature. It's a change in what Copilot is for.

Until now, Copilot in Excel functioned primarily as a prompt responder: describe what you want, get a result, repeat from scratch next time. The Microsoft Excel Copilot reusable skills system changes that model. A skill is a structured instruction set, written as a standard markdown file, that tells Copilot exactly how to execute a defined process which steps to follow, what formatting to apply, how to structure the output. Save a SKILL.md file to OneDrive and Copilot picks it up automatically, using it to build a three-statement model or board package according to the firm's own methodology (Microsoft 365 Blog, yesterday).

Think of it as the difference between briefing a capable contractor from scratch every week versus training a staff member on your firm's process once and having them run it reliably on demand.

The goal, as Microsoft put it, is to stop "starting from scratch each time," producing outputs that are easier to review, reuse, and trust (Thurrott, yesterday). A starter library of pre-built finance skills is live today across Excel for Web, Windows, and Mac. Custom skills are in Insider preview on Windows and Mac, with general availability expected next month per Microsoft's stated rollout plan. Third-party skills distributed through the Microsoft Marketplace are slated for Q3 2026 (Microsoft 365 Blog, yesterday).

The practical upside is standardization: encode a DCF methodology once and every analyst runs the same process, with the same formatting and structure. Real value for consistency, review, and onboarding. What Microsoft's announcement does not provide is any evidence that Copilot executes those instructions reliably under real conditions no benchmarks, no accuracy data, no case studies from production environments. That's the gap finance teams need to close before treating skills as anything more than a promising structure.

Copilot in Excel finance connectors: what's live and what's still unclear

The connector pillar addresses a friction point every finance professional knows. Getting vendor data into Excel typically means exporting from a platform, reformatting, pasting, checking for errors, and starting over when the underlying data refreshes. Federated Copilot connectors eliminate at least one handoff by pulling external data directly into the Excel environment, and the same connections extend into Copilot Chat and the Researcher agent (Microsoft Cloud Blog, yesterday).

Six new providers go live today: CB Insights, Daloopa, FactSet (in preview, generally available in July), Morningstar, PitchBook, and S&P Global. They follow LSEG and Moody's connectors released last month, bringing the total to eight named finance connectors and giving Microsoft 365 Copilot coverage across public equity, private market, credit, and company intelligence data (Microsoft 365 Blog, yesterday).

The coverage list is meaningful. The documentation behind it is thin. Microsoft's announcement names providers but describes datasets only at the category level. What historical depth each connector offers, what coverage universe it draws from, what refresh cadence it runs on, and what query limits apply none of that is confirmed in publicly available materials. Whether users need separate vendor entitlements on top of a Microsoft 365 Copilot subscription, and what data residency or compliance constraints apply, also remains unconfirmed. Finance teams should contact each data provider directly before treating any of these as production-ready infrastructure, particularly for workflows that touch regulated data or external reporting. The Excel Copilot FactSet, Morningstar, and PitchBook connectors may be the most immediately relevant for investment teams, but the same documentation gap applies across all six.

Traceability and controls: the governance test

Skills and connectors only matter if finance teams can stand behind the outputs. That's where traceability becomes the critical question, and where Microsoft's announcement is simultaneously most important and most incomplete.

The new "Copilot attribution in Show Changes" feature, now generally available, logs what Copilot modified in a workbook. Microsoft describes the intended result as Copilot working "more like a trusted analyst" proposing a path forward, explaining its approach, and keeping every change visible and reviewable (Microsoft 365 Blog, yesterday). Workbook rules and personalization features, also generally available, round out the control layer.

A practitioner benchmark published last month by RowSpeak, an AI-in-finance commentary, specifies what finance-grade traceability actually requires: identification of which workbook version was used, which sheets and data ranges were analyzed, which calculations were deterministic versus model-generated, what data-quality warnings were surfaced, and a complete record of what was reviewed, edited, exported, or shared. That's the checklist a regulated finance team would apply to any AI output entering a formal deliverable.

Microsoft's "Show Changes" attribution addresses part of that list, particularly the change log itself. Whether it satisfies the full scope is not confirmed in current documentation. For internal reporting and reconciliation workflows, the current controls may be sufficient to justify a pilot. For regulated external deliverables or complex multi-step models, the documentation doesn't yet provide the evidence needed to proceed with confidence.

What to verify before you pilot this

The three pillars are coherent as a framework: skills give Copilot a repeatable structure, connectors give it trusted inputs, traceability gives reviewers a record. Whether the execution matches the framework is what a pilot has to answer.

Four checks to run before taking any of this into production:

Access and licensing. Confirm which features are available at your Microsoft 365 Copilot license tier today versus next month. Verify whether each data connector requires a separate vendor agreement or whether data access is bundled.

Rollout stage. Pre-built skills and federated connectors are GA. Custom skills remain in Insider preview until next month per Microsoft's timeline. Partner skills don't arrive until Q3. Match the feature to the actual availability window before building a workflow that depends on it.

Auditability scope. Test what "Show Changes" actually captures in your environment. Compare it against your internal review and sign-off requirements, especially for any output entering a regulated report, audit deliverable, or board presentation.

Use-case fit. The features are most defensible today for internal repeatable workflows: monthly reporting packs, variance analysis, close-the-books runs, and reconciliation where a human reviews before anything goes out. Higher-stakes applications, such as external-facing or regulated deliverables, carry more risk until Microsoft provides more detailed documentation and independent validation exists.

The skills system ships today with sample templates and a markdown format any analyst can pick up quickly. That low barrier to entry means pilot costs are modest. So is the cost of discovering early, in a controlled setting, exactly what these features can and can't reliably do.

The broader play

Microsoft has made a clear architectural choice. Copilot in Excel is no longer positioned as a smart assistant for ad hoc queries. It's being built as a platform for encoding and running institutional finance workflows, with a data layer and an audit trail designed to match the stakes of that work.

The Q3 arrival of partner-built skills on the Microsoft Marketplace signals where this is headed. That's not an incremental feature update it's the beginning of an ecosystem play, with Microsoft positioning Excel Copilot as a host for finance-specific tools from third-party developers (Thurrott, yesterday).

What's still missing is evidence. No benchmarks, no independent testing, no production case studies. The RowSpeak verification framework is useful precisely because it specifies what finance teams should demand from any AI system before trusting its outputs in high-stakes work, and Microsoft's public documentation only partially satisfies it (RowSpeak, last month). Run a controlled pilot on an internal repeatable workflow, use it to stress-test the audit controls, then decide how far to expand scope. The architecture gives finance teams more to work with than they had a week ago. The proof has to come from the work itself.