What this is: a look at how Anthropic’s own finance team uses Claude, drawn from the company’s published account
The big idea: Claude holds the “integrity layer” underneath the work (consistency, references, first drafts) so the humans spend their time on the narrative and the judgment on top
The payoff: one team member reclaimed 10 to 20 hours a week that used to go to manual consistency checks
What you can copy: start with a recurring task where consistency compounds, give Claude the context, and let it draft and check while you decide
When the company that builds Claude puts its own finance team on it, the workflows are worth a close look. Anthropic published an account of how its finance team uses Claude day to day, and the interesting part is not that they automate finance. It is how they split the work: Claude takes the careful, repetitive checking underneath a report, and the humans keep the narrative, the judgment, and the final call. Here is what they actually do, and what you can borrow even if you never touch a spreadsheet.
What does Anthropic’s finance team actually use Claude for?
Four recurring jobs came up, all of them the kind of work that is slow, detail-heavy, and easy to get subtly wrong:
| Task | What Claude does |
|---|---|
| Board deck validation | Reconciles the numbers across every slide and flags where the story stops matching the data, each quarter |
| Monthly financial reviews | Drafts a first pass of the variance commentary in the team’s established voice, ready to edit |
| Financial model diagnostics | Traces references across spreadsheet tabs and points out structural problems |
| Cross-team context | Pulls the relevant decisions and reasoning out of docs, email, and Slack so nothing gets missed |
Which Claude features power this?
None of this is exotic. It is the everyday Claude toolkit, pointed at finance:
- Claude Cowork for working through documents and decks.
- Claude for Excel for reading and editing the financial models directly.
- Connectors to Google Workspace and Slack, so Claude can see the context instead of being told it. (New to that idea? See our explainer on connectors and MCP.)
- Projects kept separate by audience, so the tone for the board stays distinct from the tone for an internal review.
- Skills tuned for financial-services work.
How much time does it actually save?
Alice Fong on the finance team described reclaiming 10 to 20 hours a week that used to disappear into manual consistency checks. Every time the numbers refresh, Claude re-checks the contradictions and narrative gaps automatically, and it keeps the voice consistent from one month’s report to the next. Her own description of the shift is the clearest summary of the whole approach:
In their words
“Claude does all of this for me now: it holds the integrity layer underneath the work, so my time goes to the narrative on top.”
Why does this matter if you’re not in finance?
The pattern travels well beyond spreadsheets. Any job has a layer of careful, repetitive maintenance underneath the part that actually needs your judgment: checking that a report agrees with itself, keeping tone consistent, tracing where a number or a claim came from, gathering scattered context before you can think. That maintenance layer is exactly what an assistant like Claude is good at, which frees you to spend your attention on the decision, the story, and the call only you can make. It is the same lesson behind how AI is actually showing up in real jobs: it augments the expert, it does not replace them.
Get smarter about AI every morning
Free daily newsletter. Built for people who want to use AI well, not chase every model.
Free forever. Unsubscribe anytime.
How can you start applying this?
- Pick one recurring task where consistency matters and mistakes are easy to miss (a weekly report, a recurring deck, a status doc).
- Give Claude the context it needs. Connect the docs or paste the source material rather than describing it from memory.
- Let it do the first pass: the draft, the cross-check, the “what does not line up here.” Treat that as a starting point, not a final answer.
- You edit and decide. Keep the judgment, the framing, and the sign-off with yourself.
- Reuse the same setup or project each cycle so its memory of your work compounds over time.
Where do humans stay in charge?
This story works because the line is drawn in the right place. Claude handles the integrity layer: reconciling figures, catching contradictions, holding a consistent voice, drafting a first version. The people stay responsible for everything that needs judgment: what the numbers mean, which story to tell, what to do next, and whether the work is right before it goes out. AI carrying the tedious checking is what makes more room for human thinking, not less. If you hand over the judgment too, you have missed the point, and the value.
Common questions
Does Anthropic’s finance team let Claude make decisions?
No. Claude handles the checking, drafting, and consistency work. People keep the analysis, the narrative, and the final sign-off.
What Claude tools does the finance team use?
Claude Cowork for documents and decks, Claude for Excel for models, connectors to Google Workspace and Slack, Projects, and finance-tuned Skills.
How much time can this save?
One team member reported reclaiming 10 to 20 hours a week previously spent on manual consistency checks.
Do I need to be in finance to use this approach?
No. The pattern fits any recurring, detail-heavy work: let AI hold the consistency and first-draft layer while you keep the judgment.
Where should a beginner start?
With one simple recurring task where consistency compounds, like a weekly report or a monthly deck, then build from there.
Sources
- Anthropic: How Anthropic’s finance team uses Claude
- Model Context Protocol (the open standard behind Claude’s connectors)
Last reviewed: June 2026. Based on Anthropic’s published account of its own finance team’s workflows; product features change often, so confirm current details on Claude’s site.