Using Claude Projects for CRE Deal Work

The short answer

Claude Projects is a persistent workspace inside Claude that holds files, instructions, and context across multiple chats instead of losing everything when you close a conversation. You upload your reference documents once (buy-box, underwriting template, IC-memo format), write a short set of standing instructions, and every new chat inside that project starts already knowing your standards. For CRE deal work, that is a meaningful upgrade over a plain chat window. Underwriting is repetitive by nature: you are applying the same cap rate logic, the same red flags, and the same memo structure to a new OM every week. Projects removes the part where you re-explain your firm's rules every single time.

It is worth being precise about what this is and is not. Claude Projects is a better way to work with an AI model on documents you feed it by hand. It is not a database, not a pipeline, and not connected to your email, your data room, or your CRM. For a solo analyst doing first-pass reads on a handful of deals a month, that is often plenty. For a team trying to standardize underwriting across five analysts and keep a permanent record of every deal that came through, it usually is not enough on its own. We cover that line later in this post, and in more depth in our guide on when to build your own AI underwriting system.

Setting up a project for a deal

The setup step is what makes or breaks a Claude Project for CRE work. Skip it and you get a generic chatbot. Do it properly and you get something closer to a junior analyst who already knows how your firm underwrites. At minimum, load these into the project's knowledge:

• Your buy-box: asset classes, geography, deal size range, target returns, and the deal-breakers you don't want re-litigated on every file
• Your underwriting template or model structure, described in enough detail that Claude can populate it consistently (rent roll fields, expense line items, exit assumptions)
• Your IC-memo format, including section order and the level of detail your partners expect
• Current market assumptions: cap rate ranges by submarket, rent growth assumptions, expense ratios you're using this quarter

Then write project instructions that tell Claude how to behave: flag anything that falls outside the buy-box, ask for missing data rather than guessing it, and always output in the memo format you provided. Once that is in place, every new chat inside the project inherits it. You do not re-upload the buy-box for the tenth deal of the month. For background on the broader skill-building approach behind this, see our complete guide to Claude skills for real estate and our open claude-skills-cre repo, which has starter prompts and templates you can adapt directly into a project's knowledge base.

Practical deal workflows

With the setup done, here is what actually gets used day to day.

• OM analysis: Drop in a new offering memorandum PDF and ask Claude to pull the rent roll, T-12 financials, and stated assumptions into your standard format, then flag anything inconsistent with the buy-box you already loaded.
• First-pass underwriting: Ask Claude to run the numbers through your underwriting template's logic and produce a rough return profile. Treat this as a first pass an analyst still needs to check, not a finished model.
• Comps summary: Paste in comp data you've pulled elsewhere and have Claude organize it against the subject property, since Claude Projects has no live market data feed of its own.
• IC-memo drafting: Generate a first draft of the memo in your house format, pulling from the OM analysis and underwriting output already in the chat.
• Keeping context across a deal: Because the project retains files and prior chats, you can return to a deal a week later and ask follow-up questions without re-uploading everything. That continuity is the single biggest practical advantage over a one-off chat.

Every one of these still needs a human to check the output against the actual documents. Claude Projects speeds up the first pass. It does not replace the analyst's review.

Where it stops

Three limits matter most for CRE teams considering Projects as their underwriting tool.

• It is not a system of record. There is no queryable database of every deal you've run, no way to pull "show me every multifamily deal we passed on in the last six months," and no guaranteed long-term retention you can rely on for compliance or audit purposes.
• No automation. Nothing arrives in the project on its own. Someone has to open the OM, upload it, and prompt Claude every time. There is no watch-folder, no email intake, no trigger.
• No integrations. Projects does not connect to your CRM, your data room, your email, or a comps API. Everything that goes in, goes in by hand, and everything that comes out has to be manually moved to wherever it needs to live next.

For one analyst underwriting a handful of deals a month, that manual loop is annoying but tolerable. For a team running deal flow at volume, the manual upload step becomes the bottleneck, and the lack of a shared record means nobody can see what anyone else evaluated last quarter.

When to move to an owned system

The signal to graduate is usually volume plus coordination. If you're underwriting a couple of deals a month solo, a well-configured Claude Project is genuinely enough, and building anything more is over-engineering. The moment you're running more than a handful of deals a week, or more than one analyst needs to see the same deal history, or OMs need to flow in automatically from email or a data room instead of being uploaded by hand, you've outgrown what Projects was built to do.

At that point the conversation shifts from "which prompts do we use" to "what system do we own." A custom system means a real database of every deal, automated intake from your inbox or data room, a shared underwriting model your whole team works from, and outputs that land directly in your CRM or IC deck pipeline instead of being copy-pasted there. This is exactly the gap our AI underwriting copilot is built to close, and it's the deeper subject of our guide on building versus buying AI underwriting. NextAutomation builds these systems for CRE firms as something the client owns outright, not a subscription to a black box. If you want the primer on the broader landscape first, start with our pillar guide to Claude for commercial real estate.

Either way, start with Projects. It costs you an afternoon of setup and tells you fast whether the workflow is worth automating further.