A complete guide to Claude Projects — what it is, how it works, and exactly how investment bankers and analysts should be using it on live deals today.
The Problem With How Most Analysts Use AI
Every new chat felt like Groundhog Day.
- Re-upload the 10-K. Explain the company from scratch.
- Re-upload the CIM. Set the deal context again.
- Re-upload the model assumptions. Re-explain the thesis.
- Start from zero — every single session, every single time.
This is not a minor inconvenience. When you are mid-deal — cross-referencing a CIM against management accounts, building a comparables set, drafting the investment thesis — having to re-establish context with your AI tool every session kills the compounding effect that makes AI valuable in the first place.
Most analysts treat AI like a search engine: ask a question, get an answer, close the tab. That is the lowest-leverage way to use it. The real advantage comes from persistent, accumulated context — and that is exactly what Claude Projects delivers.
What Is Claude Projects, Exactly
Claude Projects is a feature inside Claude.ai that lets you create a persistent workspace — a named container that holds your uploaded documents, your instructions to Claude, and your conversation history across multiple sessions.
The core mechanic
You upload documents to a Project once. Every conversation you open inside that Project automatically inherits those documents as full context — without re-uploading, without re-explaining, without wasting tokens on setup. Claude sees everything from the moment you type your first word.
Think of it as giving Claude a permanent deal room for a specific engagement. The 10-K, CIM, financial model, industry report, and internal deal memo are all loaded. Every prompt you send runs against the complete picture — not a fragment of it.
A Project has three core components:
How to Set Up Your First Project
Setup takes under ten minutes. Click each step to expand the detail on what to do and why it matters.
Create a new Project + expand
In Claude.ai, click Projects in the left sidebar, then New Project.
Acme Corp — Series B or Target Co — M&A Sell-Side. Use a naming convention you will recognise three months from now. You can run multiple Projects simultaneously — one per live deal is the cleanest structure. Projects do not share documents or context with each other, which keeps your deal information clean and segregated.Upload your core documents + expand
Drop in every document Claude will need to reference across the engagement.
Write your Project instructions + expand
The most underused feature. Tell Claude who you are and exactly how to respond.
Open your first conversation + expand
Start working immediately. Claude already knows everything in the Project.
Keep the Project current as the deal evolves + expand
Update documents when new information arrives. A live Project is a living deal room.
The Before vs. After Workflow
This is what the actual workflow change looks like in practice. Toggle between the two to see the difference.
- Open a new chat. No context. Start from zero.
- Upload the 10-K. Wait for processing. Re-explain the company.
- Run your analysis. Chat closes or context fills up.
- Open another chat. Re-upload the CIM. Re-explain the deal.
- Manually stitch outputs from 3–4 separate chats together.
- Token budget consumed by re-uploads and context-setting, not analysis.
- No thread between sessions — no compounding context across the deal.
- Open a conversation inside your deal Project. All documents loaded automatically.
- Type your first analytical question immediately. No setup required.
- Claude cross-references 10-K, CIM, and model assumptions in a single response.
- Open a new conversation next session. Same Project. Full context retained.
- All sessions are part of one coherent workflow — outputs build on each other.
- Token budget spent entirely on analysis, not re-uploads or re-explanation.
- Context compounds across the entire engagement lifecycle.
IB-Specific Use Cases
The generic use case is “stop re-uploading documents.” The IB-specific value is deeper. Here is how the feature maps to real deal workflows. Click each to expand.
- Draft and iterate on the CIM executive summary across multiple sessions with full continuity
- Cross-reference management accounts against CIM claims to surface inconsistencies
- Build and refine the comparables set with running context on peer exclusion rationale
- Prepare management Q&A responses against anticipated buyer due diligence questions
- Prepare for investor objections with a running, evolving list of anticipated questions
- Cross-reference pitch deck narrative against model assumptions for internal consistency
- Draft tailored follow-up materials in response to specific investor feedback without re-explaining company context
- Screen 15–20 comparables and document inclusion/exclusion rationale in a single session
- Extract EV/EBITDA, P/E, EV/Revenue across the set and flag statistical outliers
- Run sensitivity analysis on the implied valuation range with persistent underlying assumptions
- Flag inconsistencies between the CIM and audited accounts automatically
- Cross-reference revenue recognition policies against reported numbers
- Maintain a running tracker of open items and identified risks as diligence progresses
What I Produced in One Single Session
To make this concrete — here is what I ran in a single Project session on a live deal. One conversation. No re-uploads. No context-setting.
What used to be four separate ChatGPT chats stitched together manually became one coherent, high-quality workflow. The output is sharper because Claude is reasoning across the full picture simultaneously — not a fragment of it.
AI doesn’t just save time when you use it right. It compounds. Persistent context means every prompt builds on the last — instead of starting from zero every single time.
Common Setup Mistakes to Avoid
Uploading everything indiscriminately
More documents does not equal better output. Each document consumes context window space. Upload only what Claude needs to reason across. A bloated Project produces diffuse, unfocused responses — the opposite of what you want.
Skipping the Project instructions
The instructions field is the highest-leverage setup step and the most commonly skipped one. Without instructions, Claude has no standing context on your role, the deal type, or your output preferences. Write at least 3–4 sentences. It takes two minutes and changes every response you get.
Creating one Project for all your deals
One Project per deal is the right structure. Mixing multiple deals into a single Project produces confused cross-referencing. Claude will pull context from the wrong engagement. Keep the knowledge base clean and deal-specific.
Treating Projects like a filing cabinet
The value is not storage — it is synthesis. Claude’s ability to cross-reference documents against each other in real time is the core advantage. Ask cross-document questions. Force the synthesis. That is where the output quality step-change comes from.
Not updating documents as the deal evolves
A Project with a six-month-old CIM and a current set of management accounts will produce inconsistent, unreliable analysis. When documents are revised during the engagement, replace them in the Project immediately. A live Project should be a living deal room, not a static archive.
Your Setup Checklist
Run through this before you start working in any new Project. Click each item to check it off.
Build the context once. Let every session inherit it.
The real productivity gain in AI is not the first prompt. It is the tenth.
Bhagyesh
I run an independent investment banking practice offering offshore IB analyst services — financial modeling, valuation, pitch decks, CIMs, and deal support for startups, PE/VC firms, and boutique IB firms. I share what I actually use in live deal workflows.