This CFO’s 7-step method handled 1,500 invoice lines without her (steal it)

Anna Tiomina is a Fractional CFO.

She's smart, detail-oriented, and can solve hard financial puzzles.

Last year, a client came to her with a BIG problem.

The task was to verify the claims and find the evidence.

The problem was there were 1,500 lines of invoice data.

So Anna turned to Claude to automate that workflow. And now, that task that would have taken days, runs as a Skill in Claude Cowork.

She didn't have to spend her weekend stuck in manual work. She just had a folder, a set of instructions and a workflow she used (that runs in the background).

Plus, with that time she’s saved, she can help as one of our Experts in the AI Finance Club 😉

The Agent Buzz

Last year, true 'Agents' were a buzzword. And, they mostly still are.

In my previous newsletter here, I told you to ignore Agents.

That advice mostly still stands. True AI agents or fully autonomous systems that can run themselves with zero supervision aren't ready for finance.

But something in between has arrived.

What teams are actually building are semi-autonomous AI workflows, tools that can connect to your apps and carry out tasks with much less instruction than before, but still with you in the loop.

Claude Cowork is probably the closest thing to an agent right now. You give it a folder, a set of instructions, and a goal. It works through the task, checks in with you, and delivers the output.

It's not truly autonomous (you're still supervising) but it's a big step beyond a basic workflow.

Think of it as a junior employee who shows up every Monday morning and follows a task description. They need oversight. But they get the work done.

And your job? To start thinking about how many of these you can create to support your team.

Anna did this. And now, instead of spending her weekends on reviewing documents and identifying discrepancies, she spends them on advice.

How to Build Your First AI Workflow in Claude Cowork

To build a semi-autonomous AI workflow, all you need is a clear brief, and a problem to solve.

You need to pick one repetitive task. Write down the exact process. Then let AI help you with the rest.

Step 1 – Find a task

Look for tasks you do every week that follow similar steps. They should be:

• The same input format (download from your system, folder of files, spreadsheet)
• The same process (map, filter, calculate, format)
• The same output (dashboard, summary, reconciliation)

To begin with, pick one that is easy, but that takes a lot of time.

The one that makes you think "I can't believe I'm still doing this manually."

Step 2 – Enable the Skill Creator

If you haven't used Claude Desktop yet, download it. Open Settings. Go to Customize. Select Skills. Turn on Skill Creator.

A Skill is basically a repeatable activity that the workflow can run to produce consistent outputs.

You don't HAVE to use Skills, but it is a good idea to build them so that your results are the same every time.

Plus, in Claude, you can run multiple Skills in one Chat, or in one Cowork session.

Claude already has some built in skills. E.g ones for creating PowerPoints, others for creating Docs etc. But you can create as many of your own as you like.

Think of them as activity building blocks.

Step 3 – Build your first Skill

Open Claude Desktop. In the conversation, describe your task step by step. Here's what to include:

• Input format:

• What to do with it:

• Output format:

Now, feed Claude an example. E.g one real month's files. Paste the data into the chat. Show Claude the desired output.

Then ask:

Claude will generate a Skill and will give you the option to 'Save Skill' so that you can then use it.

NOTE – You can create Skills anywhere, it doesn't have to be in Claude Desktop, we are just using Desktop as this is also where Cowork runs.

Step 4 – Set up your Cowork Folder

Cowork is Claude's workflow automation tool. You give it access to a folder and a goal, and it works through the task whilst checking in with you. When you add new files, Cowork can pick them up, run according to your instructions, and save the output in an output folder.

Here's how:

1. Create a folder on your desktop e.g 'Monthly Consolidation' with 2 sub-folders for: "Invoice_Input" and "Invoice_Output"
2. Open Claude Desktop
3. Go to Cowork (the middle option)
4. Click "Work in a Project"
5. Select the folder you created

Step 5 – Create Your Instructions

Just like when you create a Custom GPT in ChatGPT, Gemini Gem, or Claude Project, you can specify base instructions.

These tell the AI how you want it to act.

There are 2 ways you can do this:

1. Create a README file that sits alongside the folders you've created on your desktop.
2. Create a new Project in Cowork, and enter your instructions in the settings there.

Both achieve the same results.

If you have created a skill, your instructions could be super simple:

You can also evolve your instructions for Cowork to run through multiple steps, deploying multiple skills, or connecting to multiple apps, but it's always better to start simple.

Step 6 – Run It

Now, next month when your files arrive, put them in "Invoice_Input.", open Cowork and type "Carry out your instructions."

You come back 10 minutes later. The consolidated file is ready in "Invoice_Output."

Anna uses this for month-end:

Pro Tip — You can also schedule Cowork to run automatically, by going to 'Scheduled' in the side bar, and creating a recurring task following the same method as above. Just make sure your files are in the right folders before it runs.

Step 7 – Iterate

This is important. Treat your workflow like you'd treat a team member. Check in with it, refine it and expand the scope after it's working.

Anna does this every quarter.

Finance use cases you can automate this week:

• Month-end close: Files download automatically. Cowork consolidates, reconciles, creates your close summary, flags variances over 5%. Output goes to your review folder. You approve and distribute.
• Budget vs Actuals: Your actuals download. Cowork maps them to your budget template using a lookup table. Calculates variances. Highlights any over 10%. Exports to your dashboard feed.
• Expense categorisation: You have a folder of credit card exports. Cowork reads the descriptions, categorises them against your chart of accounts, flags anything unclear, exports a pre-mapped expense report.
• Invoice dispute resolution: Folders of POs, invoices, and contracts. Cowork maps them together, calculates totals by vendor, flags discrepancies, creates a summary report. (This is basically what Anna automated.)

The One Thing to Remember

The finance team of the future will have a blend of human skills, and AI skills. So you need to start up skilling now.

You might spend 2 hours building your first Skills and Cowork Projects. And yes, it takes effort. But once it's running, that task stops taking your time.

Anna could've kept spending weekends on dispute analysis. Instead, she built a Skill once that runs in Cowork whenever she needs it.

It doesn't need to be perfect (or complicated). It just needs to be a good-enough workflow that saves you 6 hours a month. That's it.

So pick a task that makes you angry because you're still doing it manually and describe it to Claude.

Then, build the Skill and set up Cowork. And by Friday, you'll have your first AI workflow running.

So tell me, now I’ve separated the buzz from what you can build right now.

What are you going to build first?

​

Best,

Your AI Finance Expert,

– Nicolas

​

P.S. — Which task are struggling to automate right now? Hit reply and let me know.

P.P.S. — If you wish to see more of what Claude Cowork could do, here's my tutorial on How to Use Claude Cowork For Finance Like a PRO. BONUS, learn how to use it with PowerPoint.

​

​