If you use Claude for finance work and you start every chat by pasting the same files and explaining your company all over again, Claude Projects fix that. This post explains what a Project is, how to set one up, and three Projects a finance team can build this week.

Most of the room had never heard of them

I taught Claude Projects to a room of finance professionals the other day. What stood out?

How few people knew they even existed.

Most were opening a new chat every single time. Pasting the same files, explaining the company again, retyping the same instructions about tone and format. Does this ring a bell?

It works, sort of. But it’s slow, and the day you forget to paste one file the answers quietly get worse, right?

So what is a Project, in plain words?

A Project is a workspace. You upload your documents once, write your instructions once, and every chat you start inside it already has that background. Nothing fancy, just very practical.

Anthropic describes Claude Projects as self-contained workspaces with their own chat histories and knowledge bases. Two parts matter for us. The project knowledge, which is the files you upload, and the project instructions, which tell Claude how to respond in every chat inside that Project.

They’re available on free accounts too, capped at five Projects. On Team and Enterprise plans you can share a Project with colleagues, and that’s where it gets super interesting for a finance team, because the whole team works from the same files and the same rules.

One thing worth knowing. On paid plans, when a Project holds a lot of material, Claude switches to a retrieval mode and pulls the most relevant parts instead of reading everything every time. So keep the files focused and clean. The cleaner the inputs, the more reliable the outputs, same as everywhere else in finance.

Setting one up is the easy part

Anthropic’s guide on how to create and manage projects has the full detail. Here’s the short version:

1. Create it. Open Projects from the left sidebar (or go to claude.ai/projects) and click “+ New Project”.
2. Name it for humans. Give it a name and a description. Claude can’t see these, they’re for you and your team, so call it something people will recognise, like “Month-end close”.
3. Add the files. In the project knowledge area, use the “+” button to upload your documents.
4. Write the instructions. Click “Set project instructions”, explain who Claude is helping, what format you want and what it should always flag, then click “Save instructions”.
5. Start a chat inside the Project. Every new chat there starts with the files and the instructions already in place.

That’s it!

My advice? Spend more time on the instructions than on the uploads. Prompting is clarity of communication. Write them the way you’d brief a new junior on day one: what we do, how we write, what never goes out without a second pair of eyes.

Three Projects a finance team could set up this week

Start with the boring, repetitive stuff. That’s where the hours are hiding.

Month-end close. Upload your close checklist, last quarter’s variance commentary and the chart of accounts. Then give it this month’s numbers and ask it to draft the commentary in the same style. You still check every explanation against what actually happened in the business. AI drafts the words, you own the reasons.

CySEC reporting. The relevant circulars, your templates, last quarter’s submission. When a new circular lands, add it to the Project and ask “what changed since last time?”. One catch, it can only compare what’s actually in there. If the new circular isn’t uploaded, there’s nothing to compare against. And treat the answer as a first read, then go back to the circular itself before anything changes in a submission.

Audit queries. Your accounting policies and last year’s management letter. When the auditors send their long list of questions, plenty of the answers are already sitting in there, right? Let Claude find them and draft the replies, then you check each one and attach the evidence. If you’re on the other side of the table, I wrote about using AI for audit materiality and sampling without losing the audit trail.

Think before you upload

One honest warning. Think about what you upload.

A Project feels like a private filing cabinet, but it’s still a third-party tool. Client data needs the same governance in a Project as anywhere else. Check what your plan and your firm’s policy allow before personal data or client files go in, and if you share Projects on a Team plan, check who can see what.

Do yourself a favour and start with documents that are already public or internal-only: the chart of accounts, your close checklist, published circulars. Add the sensitive material only once someone with the authority has signed off on it.

Pick the one that hurts the most

You don’t need all three. Pick the task that eats the most hours every month and build one Project around it. Use it for a full cycle, fix the instructions wherever the answers drifted, then build the next one.

Small wins, stacked. That’s how AI actually sticks in a finance team.

Which one would save your team the most time?

If you’d like your team to build these hands-on, with your own documents, that’s exactly what we do in the AI for Finance training.