Claude for Finance Teams: How I Use It to Run Close, Cash and Reporting

A finance team gets the most from Claude when it stops treating it as a chat window and gives it three scheduled jobs against the ledger: matching the bank overnight during close, posting one cash runway figure every morning, and drafting variance commentary where every number points to its source line. Write your finance rules into a skill Claude loads every time, keep every connection read only for the first 30 days, and put a named person on every journal and every payment.

Why most finance teams stall after the first month

The pattern I see is the same in a 20 person company and a 400 person one. The team buys seats, everyone opens a chat, and for 4 weeks the board commentary gets better. Then it plateaus. Close still takes 9 or 10 working days, because nothing about close has changed. People are typing better sentences about the same late numbers.

The second problem is that the rules live in people's heads. The controller writes a prompt for accruals, the FP&A analyst writes a different one for variances, and each prompt forgets materiality, the cost centre mapping and the board's preferred format. Twelve people means twelve versions of the finance function. Claude is only as consistent as the rules you load into it, and most teams load none.

Pick the Claude surface by who presses go

Claude comes in several forms, and finance teams waste weeks arguing about them. I use one question: does a person start the job and review it, or does it run while everyone is asleep?

The five steps I run inside a finance team

1. Write the rulebook as a skill before any agent

Anthropic launched skills in October 2025: folders of instructions, scripts and files that Claude loads when a task needs them. The first one I write with a CFO is the finance rulebook. It holds the chart of accounts, the cost centre map, the materiality line (I start at £5,000 or 5% of the budget line, whichever is bigger), the close calendar, and the red list of what never goes in. It takes about 2 hours with the controller in the room. After that, every person and every agent works from the same rules.

2. Close: match the bank overnight, read only

The first scheduled job is bank matching. From day 1 of close, an agent reads the bank feed and the ledger at 02:00, matches what it can, and sends the controller a list of exceptions by 08:00. It can't post anything. A person clears the exceptions and posts the journals. On a clean ledger, I expect close to drop from 10 working days to about 6 within 3 closes. Most of that gain is the team starting each morning on the 30 lines that need a human, instead of the 3,000 that didn't.

3. Cash: one runway figure every morning

The second job reads every bank balance and the 13 week cash forecast each morning and posts one number: weeks of runway at the current burn, with yesterday's figure beside it. If the number moves by more than 1 week overnight, it names the 3 transactions that moved it. The CEO reads it in 10 seconds. Claude reads the accounts. It never moves money, and no bank connection gets write access, ever.

4. Reporting: commentary where every number has a source

The third job drafts the variance commentary for the management pack. The rule in the skill is strict: every figure in the draft carries a tag pointing to the ledger lines it came from. A paragraph with 6 numbers arrives with 6 tags, and the reviewer ticks each one. The draft lands on day 3 or 4 of close instead of day 8, and the FP&A lead spends the saved time on the 2 variances the board will ask about.

5. Controls and cost before you widen anything

For the first 30 days every connector is read only. After that, the bar to add a write permission is a named owner, a written reason and a log of what the agent did. A person signs every journal and every payment, with no exceptions. Each agent also gets a monthly spend cap that stops the run when it's reached, because an agent stuck in a loop at 03:00 will spend whatever you let it.

From my operating seat

I'm Hayat Amin, an AI-native CFO, AI operator and AI coach for CFOs who teaches finance leaders how to build AI into finance operations. I spent twenty years in the C-suite, with three exits and three FT100 listings. Now I build and run AI agent systems inside companies, across finance, sales pipeline, onboarding and compliance evidence, and I coach CFOs one to one on the same work.

I build my own finance agents on Claude because of the rules file. Every run loads the same rulebook, so the agent at 02:00 on a Sunday behaves like the one I tested on Tuesday. Inside the companies I run, the cash runway figure is waiting before the finance team logs in, and the bank exceptions list is the first thing the controller opens. In a sale process, the question a buyer's diligence team keeps asking is where a number came from. A tagged commentary draft answers it every month, long before anyone asks.

Is Claude good for financial analysis?

Yes, for reading, explaining and writing about numbers, and Anthropic has sold a financial services package since July 2025. It's weak as the source of a number. Give it the ledger export or a read only connection, ask it to show which lines each figure came from, and tick those lines before anything leaves the team. Analysis it can trace, you can trust.

What are Claude skills, and should a CFO write them?

Skills are folders of instructions, scripts and files that Claude loads when a task needs them. Anthropic launched them in October 2025. A CFO should write the first one, because it holds the rules only finance knows: the chart of accounts, materiality, the close calendar and what never goes in. It takes about 2 hours, and it stops every person writing their own prompt every month.

Should a finance team use Claude Cowork or Claude Code?

Use Cowork when a person hands off a task and reviews the result, like turning 40 supplier statements in a folder into one reconciliation sheet. Use Claude Code when the job runs on a schedule with nobody watching, like overnight bank matching. Most finance teams I work with start in Cowork for 30 days and move the jobs that repeat every day into scheduled agents.

Where I come in

This is the work I do with CFOs one to one and inside finance teams. In the first session we write your rulebook skill, and by the next close the bank matching agent is running read only. If you're still choosing a tool, read ChatGPT for CFOs first, and for the full build order see the first five finance agents. The map of the whole topic is at AI for CFOs, the end state is at the AI-native CFO, and the sessions are at coaching for CFOs. Which part of your close would you hand to Claude first? Start at meethayat.com and tell me.