I Ran A Claude Cowork Month-End Close (Here’s the Full Workflow)

Last month I sat down on a Monday morning with the same list I have every month. Build the variance dashboard. Write the commentary for each location. Update the forecast model. Pull together the CFO deck content. Most of it was going to take the morning, maybe longer if any numbers looked weird and needed digging.

I did something different this time. I opened Claude Cowork, pointed it at my finance files folder, and just started asking questions. Four prompts and one conversation later, I had all four deliverables. The dashboard filtered by location. The commentary was saved as a Word doc. The forecast model had live sliders. The CFO deck speaker notes flagged a risk I would have missed.

This is the full workflow to run a Claude Cowork month-end close. Every step, every prompt, what the output looked like, and where you still need to apply your own judgment.

https://youtu.be/-yeTf5YFlmw

What Claude Cowork Can Do

Before the prompts, one thing to understand about how Cowork works: it reads your files directly from your computer. You connect a folder, and Claude can see everything in it. No copy/paste, no reformatting, no wrestling with paste limits.

It reads your files directly

Most people’s experience with AI and financial data looks like: open Excel, select the data, copy, switch to the browser, paste, then carefully watch the formatting fall apart. Cowork skips all of that. You point it at a folder and every prompt you write has direct access to the actual files.

This changes the workflow more than it sounds like it should. When your data isn’t the bottleneck, the only question is what you want to do with it.

Outputs go back to your folder

When Claude produces something worth keeping — a Word doc, a dashboard, a set of speaker notes — you can ask it to save the file directly to your connected folder. The output appears alongside the source files, ready to use. No copy/paste from the chat window, no manual save-as.

Setup — connect your folder

This takes 30 seconds.

1. Open Claude Cowork on your desktop.
2. Click “Select Folder” and navigate to wherever your financial files live.
3. Confirm Claude can see the folder (you can ask it to list what’s there).
4. Start prompting.

Note: You do this once per session. After that, every prompt you write has your actual data behind it. The setup is the whole setup.

Sequence 1 — build the P&L and POS dashboard

The prompt

Prompt: "I have a P&L file and a point-of-sale transactions file in this folder. Read both and build me an interactive dashboard that shows: (1) monthly revenue vs. budget by location as a bar chart, (2) sales by product category as a donut chart, (3) a table of the top 5 performing months by location. Include a location filter dropdown."

What the output looks like

Claude finds both files without you naming them, reads the data, and builds a working HTML dashboard that renders directly inside the Cowork conversation panel. There’s a dropdown at the top. Click Astoria and the bar chart updates. Click Hell’s Kitchen and the donut shifts to show that location’s product mix.

Case Study: June 2023, Three Locations

When I ran this with the Coffee Shop dataset, the dashboard showed Astoria’s June revenue sitting above budget while Lower Manhattan ran below. Clicking between locations made the gap immediately obvious — the kind of thing that takes five minutes to see in a pivot table took three seconds with the filter. In a close review meeting where someone asks “which location had the strongest month?” the answer is visible before the question is finished.

How to interact with the artifact

1. Use the location dropdown to switch between locations.
2. Hover over any bar or segment for the exact value.
3. Screenshot the view you want for the CFO deck or team update.
4. Ask Claude to add a new chart type or metric mid-conversation if the initial output is missing something.

Sequence 2 — variance commentary in one prompt

The prompt

Prompt: "Read the P&L file in my folder. Compare June Actuals to Budget for all three locations across every line item. Write 3 to 5 bullet points of variance commentary per location, like a senior finance analyst explaining results to a CFO. Flag anything over 5% variance and use the actual dollar amounts — don't round."

The executive paragraph follow-up

Follow-up: "Combine that into a single executive paragraph. Most important story first."

Case Study: Hell’s Kitchen, June 2023

The commentary Claude produced for Hell’s Kitchen named the labor variance above plan ($14,200 unfavorable), traced it to the additional barista coverage added in response to May’s scheduling issues, and noted that revenue ran in line with budget. That’s three connected observations that required reading both the GL data and the location commentary to produce. The output was ready to paste into a deck without editing.

Saving as a Word doc

Save instruction: "Save this as a Word document in my folder."

The file appears in your connected folder within seconds. From there it’s one click to open in Word, one more to attach to an email.

Sequence 3 — driver-based forecast app

The prompt

Prompt: "Using the June actuals from the P&L file, build me an interactive forecast tool. I want to be able to adjust these three drivers with sliders: revenue growth percentage per location, COGS as a percentage of revenue, and labor cost adjustment percentage. As I move the sliders, show the projected next-month P&L updating in real time. Include a total row across all three locations."

Working with the sliders

The artifact renders with three sliders and a live P&L table. Moving Astoria’s revenue growth slider up 10% updates every downstream line — gross margin, EBITDA, the total row — in real time.

Case Study: EBITDA sensitivity across all three locations

When I moved the Astoria revenue growth slider to +10%, total EBITDA across all three locations increased by $26,400 projected. When I then increased Hell’s Kitchen COGS by 2 percentage points, the combined EBITDA impact was a net $8,700 improvement — Astoria’s revenue upside more than offset Hell’s Kitchen’s cost pressure. That kind of scenario comparison used to mean rebuilding a sensitivity table in Excel.

Adding features mid-conversation

Follow-up: "Add a scenario save button so I can bookmark different combinations."

Claude updated the existing artifact in place — no rebuild, no new prompt chain. The save button appeared, I bookmarked the base case and the Astoria upside scenario, and they stayed accessible for the rest of the session.

Sequence 4 — month-end CFO deck

Feeding Claude two files at once

Reference both files in a single prompt — Cowork reads them simultaneously.

The prompt

Prompt: "Read two files in my folder: the P&L file and the commentary file. Use both to write the content for a 5-slide CFO deck for June 2023. Slide 1: Executive summary, 4 to 5 bullets, most important story first. Slide 2: Revenue performance by location, key callouts only. Slide 3: Cost and margin story. Slide 4: One specific risk and one specific opportunity heading into July, both data-backed. Slide 5: 3 action items with suggested owners and due dates. Write everything as speaker notes."

What Slide 4 actually said

The risk Claude identified was Lower Manhattan’s dependence on commuter-driven afternoon traffic — a pattern visible in the GL data that the manager commentary confirmed but didn’t name as a structural risk. The opportunity was Astoria’s mobile order growth, pulled from the commentary file’s May and June notes. Neither of these was generic. Both required reading across two different data sources. I changed the wording on one bullet. Everything else went straight into the deck.

Saving and using the output

Save as a Word doc using the same instruction as Sequence 2. Review slides 1 and 4 most carefully — the executive summary sometimes orders stories by magnitude rather than strategic importance, and Slide 4’s framing benefits from a human check on whether the risk is actually the right risk to lead with.

What Cowork can’t do yet

Works well	Still needs human judgment
Writing commentary from structured data	Complex multi-year Excel models
Building dashboard artifacts	Live ERP or GL system connections
Synthesizing across multiple files	Audit trail documentation
Forecast tools with simple driver logic	GAAP-specific accounting judgments
Drafting CFO deck content	Materiality assessments
Saving outputs to your workspace folder	Regulatory filing language

How to run this in your next close

1. Connect your close folder in Cowork before anything else in the session.
2. Run the commentary first (Sequence 2) — the output feeds directly into Sequence 4.
3. Build the dashboard second (Sequence 1) — use the artifact view to screenshot what you need for the team review.
4. Run the forecast app third (Sequence 3) if you have a scenario conversation coming up.
5. Run the CFO deck last (Sequence 4) — it gets better when the commentary already exists in the folder for Claude to reference.

What The Claude Cowork Month-End Changed For Me

Four hours of close work came down to under an hour. The dashboard took 90 seconds. The commentary was saved before my second coffee. The forecast tool was ready by the time my first meeting started. The CFO deck content needed two edits.

What still needed me: the Slide 4 risk framing, the commentary QC check, and the final judgment call on whether the Astoria mobile order opportunity was ready to put in front of the board yet.

Cowork did most of the building. I did the thinking about what to do with what it built.