Microsoft has put reconciliations, accruals and close packages inside Copilot and Excel, and thousands of users tried them in a week. The harder problem sits underneath: most leaders say their existing systems, not their AI strategy, are what stop AI from scaling.
Key Takeaways
For two years the pitch for AI in finance has been about what it might do. Last week Microsoft shipped the specifics: an assistant that drafts accruals, reconciles balances and assembles the close review package inside the spreadsheet where most finance work already happens. Early uptake was quick. But a survey released the same week suggests the bottleneck was never the assistant. It is the ledgers, ERPs and data pipes the assistant has to plug into, and the controls that are supposed to watch it once it does.
On October 29 Microsoft announced more than 25 new finance skills and partner-built plugins across Copilot, building on finance skills it put into Copilot in Excel in June. The accounting capabilities cover reconciliations (comparing balances and transactions, flagging discrepancies and preparing reconciling items for review), journal entries (drafting accruals, reclassifications, allocations, reversals and recurring entries, with checks for required inputs and balanced entries) and reporting against plan. A Finance plugin for Cowork assembles close review packages, forecast review briefs and audit evidence packages, while a Capital Markets plugin builds comparables, LBO, DCF and merger models.
The usage numbers are early but not trivial. Over seven days of limited Insider availability, Microsoft says more than 8,000 unique users invoked the accounting skills, logging more than 44,000 skill invocations and more than 18,000 Copilot conversations. On the same day's earnings call, CFO Dive reported, CEO Satya Nadella said automated tasks in Copilot had nearly doubled since July and daily users had risen nearly 50% from the prior quarter. Microsoft also reported $45.2 billion in quarterly capital expenditure, which tells you how hard it intends to push adoption.
Two details matter more to a controller than the skill count. First, the data connections. A Dynamics 365 ERP plugin brings journal entries and supporting schedules into Copilot, and partner plugins connect Intuit QuickBooks, Xero and Zoho Books, but those connectors are only beginning to roll out, and Microsoft notes third-party services may require separate licensing, authentication or admin enablement. Second, the controls. An optional audit trail in Excel records what Copilot changed and why, and individual edits can be reversed. Microsoft's own framing is that finance professionals "retain control of the recommendations, decisions, approvals, and actions that follow."
A week earlier, orchestration vendor Redwood Software published a survey of 300 senior technology, risk and AI leaders in the United States, Canada and the United Kingdom, fielded by Leger Opinion. Its central finding is a rebuke to the idea that AI adoption is a strategy problem: 82% agree that existing systems are bigger obstacles to scaling AI than a lack of strategy. Only 42% report a strong ability to connect AI agents to legacy or core systems, and 54% say they are fully prepared to orchestrate AI across existing systems.
Adoption is running ahead of that plumbing anyway. Some 89% increased agentic deployment in the past year, and 54% describe their adoption as scaling or mature. The oversight has not kept up: only 42% have a centralized registry of deployed agents and 51% have real-time monitoring of agent activity. Charles Crouchman, Redwood's chief product officer, put it bluntly: "The demo is the easy part." The real test, he said, begins when an agent touches a production system.
For finance, a production system is the general ledger. An assistant that drafts a balanced journal entry from a clean trial balance is useful. One that drafts it from a spreadsheet export of a ledger that was reconciled by hand, against a chart of accounts that differs between entities, inherits every one of those gaps and posts them faster. The Copilot connectors make the integration question unavoidable rather than solving it: whichever accounting system sits behind the plugin determines what the assistant can actually see.
The Redwood survey also puts a price on moving too fast. 77% of respondents say their organization has experienced a negative consequence from an ungoverned AI agent, and the most common one was unexpected cloud costs, followed by operational disruption and compliance issues. That is a finance problem twice over: an unbudgeted expense line, and a potential control failure that auditors will ask about.
The cost side is not going away. On Amazon's third-quarter call, CEO Andy Jassy flagged inference costs as a barrier to wider enterprise adoption, saying "we have to keep working on taking the cost of inference down." Fortune 500 companies spent nearly seven times more on Amazon Bedrock than a year earlier, and token volume on the service rose nearly 17-fold. Every skill invocation in a finance workflow is metered somewhere, and usage that doubles in a quarter is a budget line that doubles with it.
Guide
With finance skills now one click away in Excel, the question is where to deploy them first and how to justify the spend. This guide covers where CFOs start with automation and AI and how to build the business case for investment.
Download
Guide
82% of leaders say existing systems, not strategy, are what stall AI. This guide looks at the technology, process design and organizational alignment that move finance and operations teams from intention to execution.
Download
Guide
A Copilot plugin for QuickBooks can only work with what the ledger holds. This guide shows how to spot when your accounting infrastructure has become a constraint on growth, and what moving to a more capable platform involves in practice.
Download
Fully automated accounts payable rose from 7% to 31% in a year, but only 21% are very confident they would catch a fraudulent vendor change.
Hybrid pricing now leads among B2B software companies at 37%, and Intuit has agreed to buy revenue management platform Sequence.
Two-thirds of finance teams now use AI tools, yet 80% still run planning and forecasting mainly in spreadsheets.
