Claude Just Got 10 Jobs in Finance — and It Now Lives Inside Microsoft 365
Anthropic’s finance agent release is more than a product announcement. It’s a preview of how production AI stacks are actually going to be built.
There’s a pattern in how Anthropic releases things.
First: research. Second: model capability. Third: the infrastructure that makes the capability useful at scale.
The finance agent release — ten ready-to-run agent templates, Microsoft 365 native integration, and a Managed Agent deployment mode — is the third thing. This isn’t a demo. It’s a production stack.
Here’s what it actually is and why it matters.
The Architecture First
Before the individual agents, the architecture is worth understanding.
Each template packages three components:
Skills — the domain knowledge and task instructions. This is the “how to do pitchbooks” layer, the “what to look for in KYC files” layer. The accumulated expertise encoded as instructions.
Connectors — governed, real-time access to the data sources the task runs on. Not “go search the internet.” Controlled, permissioned access to FactSet, PitchBook, S&P Capital IQ, and the other data sources financial teams already pay for.
Subagents — additional Claude models called for specific sub-tasks. The Pitch Builder’s main agent might delegate comparables selection to a subagent. The Valuation Reviewer might hand off a specific methodology check. The work is parallelized and specialized at the sub-task level.
This three-component structure is the interesting part. It means teams can adapt any template to their own conventions, risk policies, and approval flows without rebuilding from scratch. The template is a reference architecture, not a black box.
The Ten Templates
The templates split cleanly into two categories.
Research and Client Coverage
Pitch Builder creates target lists, runs comparables analysis, and drafts pitchbooks. The end-to-end workflow that currently takes days of analyst time.
Meeting Preparer assembles client and counterparty briefs ahead of calls — background, recent activity, key metrics. The prep work that happens the night before.
Earnings Reviewer reads transcripts and filings, updates financial models, and flags items that are relevant to the current investment thesis. It’s not just summarizing — it’s filtering against what you’ve already decided you care about.
Model Builder creates and maintains financial models from filings, data feeds, and analyst inputs. This one interfaces directly with Excel — the model lives in the spreadsheet, not in a chat window.
Market Researcher tracks sector and issuer developments, synthesizes news, filings, and broker research, and flags items for credit and risk review. Continuous coverage, not point-in-time searches.
Finance and Operations
Valuation Reviewer checks valuations against comparable companies, methodology, and the firm’s own review standards. It knows your standards — that’s the skills layer.
General Ledger Reconciler reconciles GL accounts and runs net asset value calculations against the books of record. This is back-office work that currently requires significant human time and produces audit risk when rushed.
Month-End Closer runs the close checklist, prepares journal entries, and produces close reports. It can run autonomously overnight — which is significant, covered below.
Statement Auditor reviews financial statements for consistency, completeness, and audit-readiness. Pre-audit QA at machine speed.
KYC Screener assembles entity files, reviews source documents, and packages escalations for compliance review. KYC is one of the most labor-intensive compliance workflows in financial services — high volume, repetitive structure, severe consequences for errors.
Microsoft 365: Context Continuity
The Microsoft 365 integration is more significant than it looks on the surface.
Claude now works natively inside Excel, PowerPoint, Word, and Outlook. The individual capabilities per app are useful: building financial models in Excel, drafting decks in PowerPoint, editing credit memos in Word against the firm’s templates. But the more important feature is that context carries between applications.
What that means in practice: an analyst who builds a financial model in Excel doesn’t need to re-explain it when the work moves to PowerPoint. The deck can update automatically when the underlying numbers change. Claude carries what it knows across the whole workflow.
For most AI tools, each application is a fresh context. You explain the situation in app A, finish something, open app B, and explain it again. Context continuity eliminates that. The work is a workflow, not a series of disconnected conversations.
There’s also Dispatch — a feature in Claude Cowork that lets analysts assign Claude work by text or voice. Claude keeps working on local files while the analyst is away from their desk. The finished work is ready for review when they’re back.
Two Deployment Modes: What They Actually Mean
The same template can deploy two ways, and the distinction matters.
Plugin Mode
The agent runs alongside the analyst inside Claude Code or Claude Cowork. The analyst is in the loop. They hand the agent a task, review the output, iterate. The interaction is collaborative.
Plugin mode is for the workflow where the analyst wants Claude’s output to feed into their own judgment. Hand the Pitch Builder a target list, review the pitchbook it produces, refine it. The agent handles the compilation and first draft; the analyst handles the judgment call.
Managed Agent Mode
The agent runs autonomously on the Claude Platform. Long-running sessions — hours if needed. Per-tool permissions so each agent can only touch what it’s authorized to touch. Managed credential vaults. A full audit log of every tool call and decision, visible in the Claude Console to compliance and engineering teams.
The Month-End Closer is the clearest example. The close checklist runs overnight. By the time the finance team arrives in the morning, the journal entries are prepared, the close report is drafted, and compliance has a timestamped record of every step Claude took to get there.
This is what production AI looks like in regulated industries. Not a chatbot you ask questions. An autonomous process with audit trails.
The Data Ecosystem
AI agents are only as capable as the data they can access. The data layer here is already substantial.
Existing connectors include FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, Chronograph, LSEG, and Daloopa. New connectors added this week include Dun & Bradstreet (business identity and verified entity data), Fiscal AI (real-time fundamentals across public equities), Financial Modeling Prep (quotes, filings, transcripts across asset classes), Guidepoint (100,000+ expert interview transcripts), IBISWorld (industry-level financials and forecasts), SS&C Intralinks (DealCenter AI data rooms for diligence), Third Bridge (primary-source expert interviews), and Verisk (insurance data for underwriting and claims).
And Moody’s launched an MCP app — a deeper integration than a connector. An MCP app embeds the provider’s own tools inside Claude, not just their data. Moody’s app gives Claude access to proprietary credit ratings and data on more than 600 million public and private companies for compliance, credit analysis, and business development.
The pattern: the data layer already existed. Finance teams have been paying for these sources for years. The connectors give Claude governed, permissioned access to data the firm already owns. The intelligence isn’t finding new data — it’s synthesizing the existing data faster and at greater depth.
The Benchmark
Claude Opus 4.7 leads Vals AI’s Finance Agent benchmark at 64.37%. That’s the industry benchmark specifically for financial tasks — not the generic reasoning benchmarks where most AI comparisons happen.
Vals AI’s finance benchmark tests tasks that financial professionals actually care about: financial analysis, model building, document synthesis, and compliance tasks. 64.37% isn’t perfect, but it’s the highest score on the leaderboard, and it’s on domain-specific tasks rather than general capability tests.
Anthropic recommends Claude Opus 4.7 for finance work. The platform supports routing lighter tasks to smaller models, with complex analysis tasks going to Opus 4.7.
What This Means for Builders
If you’re a developer working in fintech or building tools for financial firms, a few things are worth paying attention to.
The template pattern is adaptable. The reference architectures are starting points. A firm building on the Claude Platform can take the KYC Screener template and adapt the skills layer to their specific jurisdictional requirements, the connectors layer to their specific data sources, and the subagent structure to their approval flows. The bones are there.
The audit log is a product requirement, not a nice-to-have. If you’re building AI workflows for regulated industries, the compliance trail isn’t optional. Managed Agents ships with this built in. Any custom agent built on the Claude Platform should account for this from the start.
The MCP app model is worth watching. Moody’s didn’t just add a connector — they embedded their own tools inside Claude. That’s a deeper integration pattern. Expect more data providers to follow this direction. The distinction between “Claude accesses provider data” and “provider tools run inside Claude” will matter for how you architect integrations.
Microsoft 365 is now the interface layer. For enterprise users, the most important integration isn’t an API — it’s where the work already happens. Excel, PowerPoint, Word, Outlook. If your AI product doesn’t meet users where their workflows already live, you’re asking them to change their behavior to use AI. The Microsoft 365 integrations remove that friction.
How to Actually Get Started (Non-Technical Guide)
You don’t need to be an engineer to use any of this. If you work in finance — as an analyst, accountant, CFO, compliance officer, or portfolio manager — here’s how to go from zero to running your first Claude finance task.
Step 1: Get a Claude Account
Go to claude.ai and sign up. You need a paid plan (Pro or above) to access the advanced models used for finance work. Claude Opus 4.7 — the one Anthropic recommends for finance tasks — is available on the Max plan.
If your firm is deploying this at scale, your IT or operations team will set up a Claude for Work or Claude Platform account. Ask them to enable the Microsoft 365 integration while they’re at it.
Cost: Pro plan is $20/month. Max plan is $100/month. Enterprise pricing is custom.
Step 2: Connect It to Microsoft 365 (Optional but Recommended)
If your firm uses Excel, Word, PowerPoint, or Outlook — and most finance teams do — install the Claude add-in from the Microsoft 365 marketplace or the Outlook marketplace.
Once installed, Claude appears in the ribbon inside each Office app. No browser tab switching. No copy-pasting between tools. You’re working directly inside your existing documents.
What this unlocks: Build a financial model in Excel → open PowerPoint → Claude already knows the numbers. You don’t re-explain anything.
Step 3: Pick the Right Agent for Your Task
Use this as a quick reference:
+------------------------------------------------------+-----------------------------+
| What you need to do | Agent to use |
+......................................................+.............................+
| Prepare for a client meeting | Meeting Preparer |
+......................................................+.............................+
| Build a pitchbook or target list | Pitch Builder |
+......................................................+.............................+
| Review earnings transcripts or filings | Earnings Reviewer |
+......................................................+.............................+
| Build or update a financial model in Excel | Model Builder |
+......................................................+.............................+
| Reconcile GL accounts or run NAV calcs | General Ledger Reconciler |
+......................................................+.............................+
| Run month-end close overnight | Month-End Closer |
+......................................................+.............................+
| QA financial statements before audit | Statement Auditor |
+......................................................+.............................+
| Process KYC files and compliance docs | KYC Screener |
+......................................................+.............................+
| Track sector news and credit developments | Market Researcher |
+......................................................+.............................+
| Check a valuation against comps | Valuation Reviewer |
+------------------------------------------------------+-----------------------------+If you’re not sure which one to start with, Meeting Preparer is the easiest entry point. It’s a familiar task — prep for a call — and the output is immediately useful.
Step 4: Run Your First Task
Here’s a concrete example using Meeting Preparer:
Open Claude (in the browser or inside Outlook)
Type: “Prepare a meeting brief for a call with [Company Name] on [date]. I need their recent financial performance, key executives, and any news in the last 30 days.”
If your firm has connected data sources (FactSet, PitchBook, etc.), Claude pulls from those. If not, it uses what it can access and flags gaps.
Review the output. It will give you a structured brief: background, key metrics, recent activity, talking points.
Edit anything that needs local context — relationships, internal history, deal-specific notes.
That’s it. The whole process takes under 10 minutes versus 1–2 hours of manual prep.
Step 5: For Overnight / Automated Work — Ask Your IT Team
Some agents (Month-End Closer, General Ledger Reconciler) are designed to run autonomously overnight without anyone in the loop. This is the Managed Agent mode.
Setting this up requires your IT or operations team to configure the Claude Platform, set permissions, and connect it to your data systems. You don’t configure this yourself — but you do need to ask for it.
What to tell IT:
You want Managed Agent mode enabled for [specific agent]
You need it connected to [your ERP / GL system / data source]
You want the audit log visible to compliance in the Claude Console
Once configured, you assign the task before you leave for the day, and the output is waiting for you in the morning.
Step 6: Review, Don’t Rubber-Stamp
Claude produces drafts. The analyst, accountant, or compliance officer still makes the final call.
A useful frame: treat Claude’s output like work from a very fast junior analyst who has read everything but doesn’t know your firm’s specific judgment calls. Review it the same way. Catch what needs local context. Sign off on what’s right. Build your own sense of where it needs the most supervision.
Over time, the templates get better as your firm adapts the skills layer to your own standards. The first month is the steepest part of the learning curve.
The Bottom Line
This release is a production AI stack, not a product announcement.
Ten templates with real domain knowledge. A Microsoft 365 layer that carries context between applications. A Managed Agent mode with audit trails for compliance. A data ecosystem with governed access to the sources financial teams already use.
The headline is ten AI agents for finance. The actual story is that Anthropic is building the production infrastructure for AI to run in regulated industries — the permissions layer, the audit layer, the data governance layer, and the workflow integration layer — all at once.
The teams who figure out how to deploy this well in the next twelve months will be operating with a significantly different capability level than the ones who don’t.