I recently received early access to Claude for Financial Services 2.0, and I think I just got a sneak peek into the future of finance workflows.
I ran a quality of earnings review in about 30 minutes. Same files, same logic, just without the usual scramble. One window stayed open while the rest of the noise fell away, and the work moved in a straight line instead of the usual zigzag through tabs and footnotes.
Claude moved through the 10-K, checked the debt schedule, surfaced a revenue recognition shift and organized the findings in a way I could hand directly to the deal team. The first pass read like a second.
This was not a parlor trick or polished autocomplete. It was a system doing real analytical work.
Meet Claude for Financial Services 2.0
Claude for Financial Services 2.0 is Anthropic’s finance-tuned model built to work inside the tools analysts already use. Through the Model Context Protocol, it connects to Egnyte, Chronograph, LSEG, Moody’s, S&P Global, Snowflake and Databricks. It reads documents, spreadsheets and financial statements as a unified information source.
It moves through data rooms and footnotes the way an experienced analyst would, reconciles numbers across files and produces structured outputs with citations. Agent Skills reflect common workflows like DCFs, comps, diligence summaries and earnings analysis. With the Excel add-in, the model now operates inside the tool where most finance work actually takes place. And the Skill set goes further than modeling and diligence, with capabilities for initiating coverage reports, one-page company profiles, structured diligence data packs and earnings call summaries generated directly from raw documents or data room files.
The result is an analytical system capable of handling data access, commentary, reconciliation and documentation at the pace modern finance requires.
The Integration Layer
Earlier versions of Claude for Financial Services sat outside the workflow. You pushed data in and pulled results out. This release moves into the middle of the work.
The connectors pull from the same systems analysts already use, which is why the model can sit at the center. Snowflake for actuals. Egnyte for diligence. S&P for comps. No file shuffling.
The reasoning engine has been tested against financial benchmarks, and the Agent Skills track closely with institutional workflows. The outputs carry citations and traceability, which makes review straightforward. With the Excel add-in tying everything back to the environment where analysts already work, the system lands in a place that feels cohesive rather than bolted on.
Excel Is Where Finance Lives
Finance happens in spreadsheets. Claude now works there too.
Inside Excel, it reads formulas and dependencies. It explains calculations with cell-level references. It fixes errors without breaking logic. It tests scenarios and highlights affected cells. It builds schedules and models from prompts.
Each session starts clean. Chat history does not persist between sessions. Anthropic’s financial-services deployments do not use customer data to train models. For teams working under SOX, ICFR or PCAOB oversight, this is baseline governance.
Claude behaves like a senior analyst you can question, a junior analyst who drafts quickly and a documentation system that shows its work.
What Early Adopters Report
The first institutions using Claude are the ones least inclined to chase trends, which makes their early adoption notable.
Per Anthropic’s release on the financial services expansion, NBIM reports roughly 20 percent productivity gains. AIG reduced underwriting cycles by more than 5x with accuracy climbing past 90 percent. RBC Capital Markets streamlined workflows that once demanded days of manual data gathering. Chronograph sees its integration opening new ground for private-markets teams.
Each example points to the same pattern: early gains coming from the elimination of friction, not a change in the underlying work.
Current Limitations
The model moves quickly, but it doesn’t clear every hurdle.
Very large spreadsheets can slow it down. Pivot tables, data tables, conditional formatting, macros and VBA remain outside its reach. Multi-entity or cross-year reconciliation still calls for an analyst who understands the story behind the numbers.
Shared spreadsheets also introduce security risks, and prompt injection is one of them.
Claude delivers a strong first draft. Your judgment closes the loop.
How CFOs Should Pilot This
Start with workflows where traceability is essential. Quality of earnings. Monthly close. Variance commentary. Board materials.
Use Claude as both a senior and junior analyst. Ask it to explain logic. Let it draft schedules and summaries. Review the work for consistency and accuracy.
Build governance as you adopt. File access controls. Spreadsheet security settings. Reviewer sign-offs. Clear usage policies.
Finance teams that adopt early will set the pattern others follow, just as the 30-minute quality of earnings review hinted at what becomes possible when the work starts to move this quickly. Infrastructure settles around whoever moves first.
