Cloud360
Back to blog
Microsoft Copilot

Copilot for finance: what IT must prepare first

Microsoft brought AI to the accounting close and financial analysis. What defines success isn't the model: it's your permissions, your ERP and your data governance.

November 4, 20268 min read
Ivan LimaIvan Lima · Digital Transformation Expert
Copilot for finance: what IT must prepare first

Copilot for finance demands more from IT than from finance.

Microsoft has just turned Copilot into an assistant that can put together reconciliations, journal entries and investment memos. The question for IT isn’t whether it works, but whether your environment is ready for an agent to read, cross-reference and draft with the company’s financial data.

Three days to month-end close. The controller asks IT to enable “that new Copilot thing” because a peer at another company says it prepares the close package in hours. Finance has already decided. It falls to you to find out what that agent will be able to see.

That scene will play out at many companies across the region in the coming months. On October 29, Microsoft announced more than 25 new finance skills and partner plugins for Copilot, built for accounting, capital markets and corporate finance.

What matters to you isn’t the feature list. It’s that, for the first time, AI stops working on a single file and starts working across entire workflows that span files, conversations, the ERP and external sources. That shifts the risk, and the responsibility, straight to IT.

Figure What it means
25+ skills and plugins The offering already covers reconciliation, journal entries, FP&A, valuation and reporting
8,000+ users in 7 days Real usage in limited access, ahead of broad availability
44,000+ invocations Finance’s demand won’t wait for IT’s calendar
26 external partners Each plugin is one more integration to govern

What Microsoft announced, without the brochure

The announcement has three layers, and it’s worth separating them.

Accounting skills in Excel and Cowork

They automate the most manual accounting tasks.

  • Account reconciliation that identifies differences and reconciling items.
  • Draft journal entries (accruals, reclassifications, reversals) with balance validation.
  • Budget-versus-actual reports that explain what changed and why.

They’re in Frontier, Microsoft’s early access program. The Excel finance skills, by contrast, are already generally available.

Cowork plugins for finance and capital markets

Cowork is Copilot’s agent for multi-step tasks. With these plugins, an analyst can delegate anything from preparing a close review package to testing an acquisition thesis with an LBO model. Both plugins are generally available.

Connection to systems

A Dynamics 365 ERP plugin, Fabric IQ over Power BI models and plugins from providers such as QuickBooks, Xero, Moody’s, FactSet, S&P Global and LSEG. Several are in gradual rollout.

From assistant on a sheet to agent in a workflow

Until now, AI in finance was a companion inside Excel. Useful, but limited to the open file.

The underlying change is something else. With Work IQ, Copilot draws on organizational context: the close calendar, the deal team’s conversations, emails with vendors. With Office integrated into the Copilot app, users no longer even need to start in Excel.

For the business, that means fewer hours consolidating information. For IT, it means the agent sees what the user can see, everywhere at once.

An agent working across entire workflows doesn’t create new risks: it exposes the ones your organization already had.

The permissions nobody reviewed now have a reader

Microsoft is clear on one point: Copilot operates within the Microsoft 365 boundary, honors existing permissions and sensitivity labels, and doesn’t use your data to train the foundation models.

That’s good news and, at the same time, the problem.

Honoring permissions only protects you if the permissions are right. Think of the SharePoint site where someone shared the payroll folder “with the entire organization” three years ago. Or the Teams channel for the acquisition project that nobody archived. Until today, finding that information meant knowing where to look. An agent building a close package doesn’t need to know.

  • Before: excess permissions were a latent risk that depended on someone searching.
  • After: excess permissions are an active risk, because the agent searches for you.

That’s the argument you can take to the committee: AI doesn’t reduce the need for data governance, it makes it urgent.

Answer quality is decided in the ERP

An AI-generated reconciliation is only as good as its source data. If the chart of accounts has duplicates, if cost centers aren’t standardized across subsidiaries or if half the adjustments live in parallel spreadsheets, the agent will produce fast, wrong drafts.

The Dynamics 365 plugin makes it possible to prepare journal entries and schedules using data from the system of record. Fabric IQ does the same with governed data in Power BI and Fabric. Both reward companies that already have a tidy semantic layer.

If your company runs another ERP, the question changes: which connector exists, who maintains it and with what credentials does it operate? The announcement doesn’t mention connectors for SAP or Oracle, so it’s worth confirming with Microsoft before committing to a plan.

AI doesn’t fix a messy data model. It makes it fail faster and with better prose.

Traceability: the argument that opens the door with audit

Here’s the most interesting part for anyone who has to defend adoption before internal audit or the risk committee.

When invoking an accounting skill in Excel, users can start an audit trail that records what Copilot changed and why. Copilot can also reconstruct the workbook’s history: who changed what, when, and how it affected the analysis. It can even revert specific edits without touching the rest.

To an auditor, that’s worth more than any productivity promise. But there’s a condition: the audit trail is optional. If you don’t make it mandatory by policy, each user will decide on their own.

Third-party plugins: a perimeter IT didn’t choose

The partner list is long: accounting platforms, credit risk agencies, market data providers and business banking. Microsoft notes that these services may require separate licensing, subscriptions, authentication or administrator enablement.

Translation: each plugin is a new vendor in your data chain. Before enabling one, it’s worth answering:

  1. What company data goes out to that vendor?
  2. Under what contract and in which jurisdiction is it processed?
  3. Who approves activation: the user, the department or IT?

In Latin America, where data protection rules and financial regulators’ requirements vary from country to country, this isn’t a formality. It’s a legal review.

A 90-day plan so you’re not late

Finance is going to push. Better to arrive with a plan than with a no.

  1. Exposure inventory. Identify sites, teams and folders holding financial data and check who actually has access. Start with whatever is shared with the entire organization.
  2. Minimum viable labeling. Apply sensitivity labels to payroll, projections, transactions and unpublished statements. Don’t wait for perfect classification.
  3. Scoped pilot. Pick one process, such as a subsidiary’s bank reconciliation, and a small group of users. Measure hours saved and errors caught.
  4. Traceability policy. Define when the audit trail is mandatory and who reviews drafts before they’re posted.
  5. Plugin committee. Set up an approval process for third-party connectors with IT, legal and security at the table.
  6. Scale with data. Take the pilot results to the CFO and the board, licensing costs included. The announcement doesn’t publish prices, so request a quote for Copilot and each plugin for your region.

What a skeptical CIO will ask

“We already have Copilot enabled. What changes?” The scope. Moving from an assistant inside a document to an agent that carries out multi-step tasks with organizational context multiplies what it can find. Controls that were enough for chat may fall short for Cowork.

“Several features are in early access. Isn’t it better to wait?” Wait to deploy, yes. Wait to prepare, no. Cleaning up permissions and labeling takes months, and you’ll need them either way. If finance adopts on its own while IT waits, you’ll lose control of the how.

“What if the AI makes a mistake in a journal entry?” It will. Microsoft’s design assumes the professional reviews and approves: Copilot prepares, the person decides. Your job is to make sure that review point exists in the process and doesn’t depend on each user’s goodwill.

The bottleneck won’t be AI

The capabilities are already on the table and early usage shows finance wants them. What will separate the companies that capture value from those that pile up incidents isn’t the model or the license.

It’s the state of their permissions, the quality of their ERP and the clarity of their rules on who reviews what.

That work belongs to IT. And it starts before the controller calls again.

AI in finance is a data governance project with a better interface.

Your first step this week: ask your team for a report of the sites and folders with financial data shared with the entire organization. With that single data point you’ll have the starting point for your pilot and the argument for your next meeting with the CFO.

Sources

Sharein

Want to dig deeper into this?

Let's talk about how to apply it in your business.

Talk to an expert

You might also like

Copilot Now Executes, Verifies, and Governs Itself: What Changed in 2026
Microsoft Copilot

Copilot Now Executes, Verifies, and Governs Itself: What Changed in 2026

September 22, 2026

Read More
The New Copilot: What Changes for SMBs in Ecuador
Microsoft Copilot

The New Copilot: What Changes for SMBs in Ecuador

September 28, 2026

Read More
An AI agent on the service desk: the ROI Forrester measured
AI Agents

An AI agent on the service desk: the ROI Forrester measured

October 14, 2026

Read More