Claude Cowork for Finance – Automate Your Work Step-By-Step

Every finance team I’ve worked with has some version of the same problem. The data exists. It’s sitting in a folder somewhere, usually a shared drive that nobody has organized since the last reorg, in files named things like “GL_Export_FINAL_v3_USE THIS ONE.xlsx.” Before anyone can do actual analysis, someone has to figure out what they’re working with, get it into a usable shape, and then start building the outputs that leadership actually needs.

That three-step sequence, orient, assemble, package, is where most of the hours go. Not the thinking. The assembly.

I’ve seen analysts spend the first two days of a close cycle just getting files organized and reconciled before a single formula gets written. I’ve seen FP&A managers stay late not because the analysis was hard but because the PowerPoint took three hours. I’ve watched teams rebuild the same variance summary from scratch every single month because nobody ever made it repeatable.

The analysis itself is usually the fastest part. It’s everything around it that grinds people down.

That’s the problem Claude Cowork is actually designed to solve. Not AI as a chat tool you ask questions. AI as something that sits down, opens your files, and does the work.

https://www.youtube.com/watch?v=bku9DGSpWBc

What Is Claude Cowork

If you’ve used Claude in the browser, you already know what it can do with a prompt. You paste in some data, ask a question, get an answer. It’s useful. But when you close the tab, nothing happened to your files. You still have to take the output and do something with it yourself.

Claude Cowork is a different mode entirely. It runs inside the Claude desktop app, you give it access to a specific folder on your computer, and it works through tasks the same way a capable analyst would. It reads your files, figures out what’s in them, executes the steps, and saves finished outputs directly to your folder. You can watch it work or walk away and come back when it’s done.

The distinction that matters for finance teams is this: regular Claude tells you how. Cowork does it.

It’s worth being clear about what that means in practice, because the use case for finance work is almost immediately obvious once you see it. Cowork can read a folder full of inconsistently named GL exports and organize them. It can pull daily cash balances, identify unusual movements, and write a narrative summary you could put in front of a CFO. It can read your GL and POS data together and produce a memo and a slide deck from a single prompt. All of that happens in your folder, on your machine, without you touching a formula or building a pivot table.

One thing to know before we get into the setup: Cowork stores conversation history locally on your device, not on Anthropic’s servers. For finance teams with concerns about where their data lives, that distinction matters and it’s worth knowing upfront.

How to Set Up Claude Cowork

Before we get into what Cowork can do with your finance data, you need to get it running. This is not a complicated setup, but there are a few steps that trip people up if they try to skip ahead.

Step 1: Install the Claude Desktop App

Go to claude.ai/download and download the desktop app for your operating system. Cowork is available on both Mac and Windows. This is a separate install from the browser version of Claude, and it’s required. You cannot run Cowork from claude.ai in your browser.

Once it’s downloaded, install it and sign in with your existing Claude account. If you don’t have one, you’ll need to create one and subscribe to a paid plan. Cowork is available on Pro ($20/month), Max ($100/month), Team, and Enterprise. It is not available on the free tier.

Step 2: Switch Into Cowork Mode

Open Claude Desktop. At the top of the interface you’ll see two tabs: Chat and Tasks. Click Tasks. That puts you in Cowork mode. The input field now expects a task description rather than a conversational prompt, which is a small but meaningful difference we’ll cover when we get to the actual prompts.

Step 3: Grant Access to a Folder

Click the folder icon in the Cowork interface and select the local folder you want Cowork to work with. This is the permission step. Cowork can only read, edit, and create files inside folders you explicitly grant access to. Nothing outside that folder is touched.

For the walkthroughs in this article, I’d recommend creating a dedicated working folder before you start. Drop in your source files, give Cowork access to that folder, and you’re ready to go. Don’t grant access to your entire desktop or documents folder on the first run. Keep it scoped.

Step 4: Prepare Your Source Files

For the three sequences I’m walking through here, you need three things in that folder:

• A GL export with transaction-level detail covering at least two comparable periods
• A daily bank balance file or a cash account summary pulled from your GL
• POS transaction data broken out by location

They don’t need to be clean or consistently named before you start. That’s actually the first thing we’re going to let Cowork handle.

Step 5: Understand the Plan Approval Step

Before Cowork takes any action, it shows you its plan. It lists the files it found, describes what it intends to do, and waits for your approval. This step exists for a reason. Review it every time. If Cowork misread a filename or misunderstood the task, this is where you catch it before anything gets changed. I treat it the same way I’d treat reviewing a staff member’s work plan before telling them to go execute. Thirty seconds now saves you from cleaning up a mess later.

Once you approve, Cowork gets to work. You’ll see a progress indicator showing what it’s doing at each step. You can watch, you can redirect mid-task if something looks off, or you can go do something else and come back when it’s finished.

Walkthrough Step One: Getting the Files Under Control

Most finance analysis workflows have a dirty secret. Before anyone runs a single formula, someone has to spend an hour figuring out what files they’re actually working with. Which export is current. Whether the budget file in the folder is the approved version or the one from the scenario that got killed in March. Whether there are duplicates sitting in there from the last time someone grabbed a quick pull and forgot to delete it.

I’ve seen this cost teams two to three hours at the start of every close cycle. It’s not glamorous work and it doesn’t require much skill, but it requires attention and it cannot be skipped, because if you start your analysis on the wrong file, everything downstream is wrong.

This is where Cowork earns its place immediately.

The Prompt to Use

Drop this into the Cowork task input exactly as written, adjusting the dataset name for your own files:

“I have a folder of F9 Finance Coffee Shop files. They include general ledger exports, point-of-sale data, and some other documents. The naming and organization is inconsistent. Please review all the files, identify what each one contains, rename them with a consistent naming convention that includes the file type and date range, create subfolders to organize them by category, and move each file into the right subfolder. Give me a summary of what you found and what you changed.”

A few things worth noting about how this prompt is written. It tells Cowork what the files contain so it has context before it starts reading. It specifies what the naming convention should include, not just that it should be consistent. And it asks for a written summary at the end, which gives you a record of what changed and flags anything Cowork wasn’t sure about.

What Cowork Does With It

After you approve the plan, Cowork opens each file, reads the contents, and starts working through the folder. What you get at the end is a reorganized folder with subfolders by category, files renamed with consistent conventions that include the data type and period, and a written summary that tells you exactly what it found.

On a typical Coffee Shop dataset, that summary might tell you it found two files covering the same period and flagged them as potential duplicates, that one file didn’t match the expected format and was placed in a review subfolder, and that the renaming convention it used was GL_[Period][Location] for the GL files and POS[Period]_[Location] for the transaction data.

That last part matters as much as the renaming itself. Cowork isn’t silently making decisions. It tells you what it did and what it wasn’t sure about, which means you’re not inheriting a black box. You’re inheriting a clean folder with a clear record of how it got that way.

The other thing to understand is that this organized folder becomes the foundation for everything else in the session. Cowork now knows what’s in it. When you run the cash analysis in the next step, it’s not starting from scratch. It’s working from context it already has, which makes every subsequent task faster and more accurate.

Case Study: A Finance Team’s Shared Drive Before Month-End Close

I was working with a finance team at a mid-size company that was three days out from close. Their shared drive had GL exports from two different system pulls, a revised budget file with a name nearly identical to the original, and a cash summary that one analyst had updated and another had also updated separately without knowing the first one had done it. Nobody was sure which version of anything was current.

The manual process for sorting that out was someone opening every file, checking the timestamps and the data inside, and making judgment calls about what to keep. That took the better part of a morning.

Running the Cowork file organization sequence on that same folder took a few minutes. It flagged the duplicate cash files immediately, identified which GL pull was more recent based on the data inside rather than the filename, and organized everything into a structure the whole team could navigate without a guided tour. The analyst who would have spent that morning doing the cleanup spent it doing the actual analysis instead. The close still happened on the same timeline. The difference was what the team did with their time inside it.

Walkthrough Step Two: Reviewing Daily Bank Balances and Surfacing Cash Risks

Most cash reviews I’ve seen happen by accident. Someone is building a report, they notice a number looks off, they dig in, and they find something worth flagging. That’s not a cash review process. That’s luck wearing a spreadsheet.

The problem is that a real cash review, one where someone actually sits with the daily balance data and looks for patterns, takes time that most finance teams don’t protect. It gets compressed into whatever is left after the close package is built, which means it either doesn’t happen or it happens at 7pm when nobody is at their sharpest.

What Cowork does here is make the review deliberate instead of incidental. You point it at your bank balance data, tell it what you want, and it comes back with a narrative that names what it found. Not a table. A paragraph with a point of view. That’s the part that changes how this information gets used.

The Prompt to Use

“Using the bank balance data in this folder, please review the daily ending balances for the F9 Finance Coffee Shop. Summarize how balances trended over the period, identify any days with significant drops or unusual movements, and flag any patterns that would warrant a closer look from a cash management perspective. Write your findings as a brief narrative summary I could share with a CFO.”

The phrase “share with a CFO” is doing real work in that prompt. It tells Cowork the output needs to be clean, concise, and written for someone who wants the conclusion first and the supporting detail second. When you leave that framing out, the output tends to read more like a data description than an analysis. Put it in and the difference is immediate.

What the Summary Actually Said

The output from this sequence on the Coffee Shop dataset came back as a tight three-paragraph narrative. It opened with the overall trend across the period, moved into the specific days where balance movement was outside the normal range, and closed with a plain-language observation about what the pattern suggested.

It identified a single-day drop midway through the period that was roughly 30 percent larger than the average daily movement, noted that the timing corresponded with what appeared to be an above-average disbursement cycle, and flagged that while no sustained deterioration was visible, the variance in daily movements was wider than the first half of the period suggested it would be.

That is a finished paragraph. Not a starting point for one. The kind of thing you used to write yourself after staring at the data for twenty minutes, then editing for another ten because the first draft had too many numbers in it.

The Follow-Up That Adds the Risk Call

After the summary comes back, ask this:

“If you had to identify one risk worth mentioning to leadership based on this data, what would it be and why?”

This is where the interaction shifts from reporting to reasoning. Cowork doesn’t just pull the biggest number and call it a risk. It looks at the pattern across the period and makes a judgment call about what warrants attention. On the Coffee Shop data it came back with a specific observation about timing risk between disbursements and revenue deposits, and framed it in terms of what it could mean for short-term liquidity if revenue softened in a given week.

You still need to validate that. You know your business and Cowork doesn’t. But the first draft of the thinking is done, and in my experience that’s where most of the time goes. Getting from the data to a sentence that says something is harder than it looks, and Cowork gets you there in seconds instead of twenty minutes.

Walkthrough Step Three: From Raw Data to Memo and Slides in One Prompt

This is the sequence I use to close the case with finance teams who are still on the fence. Not because it’s the most technically impressive thing Cowork can do, but because it produces something you can hold. A memo you could send. A slide deck you could present. Both of them built from the same data, in the same session, from one prompt.

If you’ve run the first two sequences before this one, Cowork already has context on your files. It knows what’s in the GL export, it’s seen the cash data, and it’s working from a folder it already organized. That continuity matters. The analysis in Step Three is better because of the work done in Steps One and Two.

The Prompt to Use

Using the general ledger and point-of-sale data in this folder, please analyze the financial performance of the F9 Finance Coffee Shop for the most recent full period available. The analysis should cover revenue by location, key expense categories, gross margin, and any notable variances. Then do two things: first, write a one-page executive memo summarizing the findings and any areas that need attention, formatted professionally and ready to send. Second, create a PowerPoint presentation with slides that cover the same content visually, including a summary slide, a revenue slide, an expense slide, and a key observations slide. Save both files in this folder.”

Name the slides explicitly. When you leave the structure open-ended, Cowork makes reasonable choices, but they may not match what your organization expects. When you specify the slides, you get the structure you want without a revision cycle.

The Memo Output

The memo came back formatted with a header, a brief context paragraph, and findings organized by category. Revenue by location was clean. The expense section called out the two categories with the most meaningful movement. The closing paragraph named what warranted follow-up without overstating it.

What needed a pass: a few sentences were longer than they needed to be, and one section repeated a number that had already appeared earlier in the memo. That’s a five-minute edit, not a rewrite. The structure, the findings, and the framing were all there. I was cleaning up, not rebuilding.

The PowerPoint Output

The slide content was solid. Each slide had the right information on it, organized in a logical sequence, with the key figures visible without hunting for them. The revenue slide broke out performance by location. The expense slide called out the variance drivers. The key observations slide landed the conclusions without burying them in supporting detail.

The design was functional and not much more than that. Default fonts, no formatting polish, the kind of slide you’d show an internal team but wouldn’t put in front of a board without cleaning it up first. If design matters for your use case, plan for a pass in PowerPoint before the deck goes anywhere. The content gives you a strong foundation. The formatting is your job, or Claude in PowerPoint can handle it as a follow-up step.

The Follow-Up on Recommendations

After the deck is built, add this prompt:

“Can you revise the key observations slide to add a recommendation section with two or three specific actions the team should take based on the data?”

What comes back is not generic advice. It’s grounded in what Cowork found in the analysis. On the Coffee Shop data, the recommendations referenced the specific locations showing margin pressure, the expense categories with the widest variance, and the cash timing pattern flagged in Step Two. It connected the dots across all three sequences and put action items on a slide.

That is the moment where most finance professionals stop and recalibrate what they thought this tool was capable of. It’s not auto-completing your sentences. It’s reading your data, forming a view, and telling you what to do about it. You still own the judgment call on whether it’s right. But the first draft of the recommendation is done, and it’s built on your actual numbers.

Case Study: An FP&A Manager Running Monthly Reporting Alone

I talked to an FP&A manager at a company where the finance function was essentially her and a part-time analyst who was also supporting accounting. Monthly reporting was her responsibility start to finish. Close took twelve to fifteen hours spread across four days, and most of that time was not analysis. It was finding the right files, building the comparison views, writing the variance commentary, and then formatting everything into a package that looked professional enough to send to the leadership team.

She ran the three-sequence Cowork workflow on her next close cycle. File organization took a few minutes instead of an hour. The cash summary came back as a draft she edited rather than a blank page she filled. The memo and slide deck needed cleanup but the structure and content were done. Her total time on the close dropped by roughly five hours that first cycle, and she expected it to improve as she refined her prompts.

The thing she said that stuck with me: the analysis she was doing before wasn’t bad. It was just buried under all the assembly work. When the assembly got faster, the analysis got more attention. The deliverable got better at the same time the hours went down. That combination is what the tool is actually for.

Where Claude Cowork Falls Short Right Now

I want to be straight with you on this because the limitations matter and you’ll hit them faster than you expect.

Cowork is in research preview. That’s not a marketing disclaimer, it’s a description of where the product actually is. The core capability is real and it’s useful, but there are rough edges and you should know what they are before you build a workflow around this tool.

Here’s what I’ve run into and what I’ve heard from others using it on finance work:

• The desktop app requirement is a real constraint. Cowork does not run in the browser. If your work machine is locked down by IT and you can’t install software without approval, you’re either waiting on that approval or running this on a personal machine. For enterprise teams, this is the first conversation you’ll need to have before Cowork becomes part of anyone’s regular workflow.
• Complex spreadsheets can trip it up. Files with heavy formatting, merged cells, non-standard structures, or multiple embedded tables are harder for Cowork to parse cleanly. A straightforward GL export with consistent columns works well. A heavily formatted management report with color-coded sections and summary rows mixed into the detail is a different story. Know your files before you commit to a workflow that depends on them being readable.
• The PowerPoint output needs a design pass. I said this in the walkthrough and I’ll say it again here because it’s the expectation gap I see most often. The content is strong. The design is not presentation-ready without cleanup. If you’re handing this deck to a CFO tomorrow morning, plan time to format it.
• Scheduled tasks only run when your machine is awake and the app is open. If you set up a Monday morning report to run at 7am and your laptop went to sleep Sunday night, the task doesn’t run. This is a meaningful limitation for recurring automation workflows. It works well if you have a desktop that stays on. It’s less reliable on a laptop that travels.
• Behavior can vary between sessions. Because Cowork is still in research preview, the same prompt doesn’t always produce identical output. For most tasks that’s fine. For anything where consistency really matters, build in a review step rather than assuming the output will match what you got last time.

None of these are reasons to ignore the tool. They’re reasons to go in with accurate expectations and design your workflows around the constraints rather than pretending they don’t exist.

How to Run This Yourself

The fastest way to get a feel for what Cowork can actually do is to run it on a real folder with real files. Not a test. Not a demo dataset you’ve never worked with before. Something from your actual close process where you already know what the answer should look like, so you can evaluate the output with a critical eye.

Here’s how to start:

Install the app and get into Cowork mode. Go to claude.ai/download, install the desktop app, sign in with a paid account, and click the Tasks tab. That’s your starting point.

Create a working folder with three files. One GL export covering at least two comparable periods. One daily cash or bank balance summary. One POS or revenue file broken out by location or category. Drop them in a dedicated folder, grant Cowork access to that folder, and you’re ready.

Run the three prompts in sequence. Start with file organization, move to the cash review, finish with the memo and deck. Run them in the same session so Cowork carries context forward. The third sequence is meaningfully better when it’s building on what happened in the first two.

The Three Prompts to Copy

Prompt 1: File Organization

“I have a folder of finance files. They include general ledger exports, point-of-sale data, and some other documents. The naming and organization is inconsistent. Please review all the files, identify what each one contains, rename them with a consistent naming convention that includes the file type and date range, create subfolders to organize them by category, and move each file into the right subfolder. Give me a summary of what you found and what you changed.”

Prompt 2: Cash Balance Review

“Using the bank balance data in this folder, please review the daily ending balances. Summarize how balances trended over the period, identify any days with significant drops or unusual movements, and flag any patterns that would warrant a closer look from a cash management perspective. Write your findings as a brief narrative summary I could share with a CFO.”

Prompt 3: Memo and Slide Deck

“Using the general ledger and point-of-sale data in this folder, please analyze the financial performance for the most recent full period available. The analysis should cover revenue by location, key expense categories, gross margin, and any notable variances. Then do two things: first, write a one-page executive memo summarizing the findings and any areas that need attention, formatted professionally and ready to send. Second, create a PowerPoint presentation with slides that cover the same content visually, including a summary slide, a revenue slide, an expense slide, and a key observations slide. Save both files in this folder.”

How to Correct Course Mid-Task

If Cowork’s plan looks off when it shows you the approval step, don’t approve it and don’t cancel it. Tell it what’s wrong. You can type directly in the task interface to redirect before it starts executing. If it’s already running and you see it heading somewhere you don’t want, you can interrupt mid-task and give it new direction. You’re not locked in once you hit approve. Think of it less like submitting a form and more like working with someone where you can tap them on the shoulder and say “actually, do it this way instead.”