I don't manage my rentals. A Project does.

I don't manage my rentals. A Project does.

How I turned monthly property-manager statements into a Claude Project that watches the portfolio for me, and the ten-minute build you can copy.

Paul Cobb
August 17, 2026

For years, staying on top of my rental portfolio meant reading every monthly statement from my property manager myself, trying to remember what looked off last month, and hoping I'd catch a maintenance charge creeping up before it turned into a new roof. Now a Claude Project does that watching. It reads the statements, trends the numbers, flags what I should ask about, and keeps a running list of open issues. I review and decide. It remembers everything else.

In the last issue I said the real jump isn't which AI you use, it's giving a whole business function its own Project. This is the one I promised to open up.

Why a Project, and not just a chat.
Most people use AI as a chat they start over every time. For recurring operator work, that restart is the whole problem. A Project is the opposite: one place that keeps its instructions, its files, and its memory, so it builds on last month instead of meeting you fresh every time. That is what turns AI from a smart intern you re-brief daily into a team member who already knows the account.

The ten-minute build.
Brandon Gaille at Claude Unleashed wrote a clean ten-minute version of this, and it is worth crediting because the bones are right. Here they are, then I'll show you the one I actually run.

1. Pick a use case that repeats. If the work happens again next month, and there's reference material that should ride along with every chat, it's a Project. Skip the giant "everything I do" project and pick one job.

2. Name it and write seed instructions. Act as [role], for [audience], produce [this], follow [these rules], and never do [these things]. Don't perfect it on day one. It sharpens over the first week.

3. Keep the knowledge base under six files. One that sets the standard, a couple of real examples that hit the bar. Retrieval value, not file count.

4. Run a test prompt that forces it to open a file and answer in your format. If it quotes a real line and follows your rules, it works. If not, the fix is almost always in the instructions.

The one I run.
Mine is a Property Operations Manager for the rental side. Its standing job: read each monthly owner statement, log rent collected and vacancies, trend maintenance across the portfolio, and flag when a pattern says a system (roof, HVAC, plumbing) is heading toward end of life, so I can plan the capital instead of getting surprised by it. It drafts the emails to my property manager in my voice, engaged and expecting accountability, and it reminds me to follow up when something goes quiet. Month over month, it is the rent-roll read and the maintenance dashboard I never kept consistently by hand.

Here is the part that makes it a system instead of a chatbot. At the end of each monthly review it produces a one-page Ops Summary, and that summary is the input to a second Project: my CFO. One feeds the other. The CFO never has to go digging for what happened in the portfolio, because the Ops Manager already handed it over. Two Projects, one seam between them, and the work flows.

The play: give one function a Project this week. Paste this skeleton into a new Project's instructions and fill in your own function:

"You are my [role] for [my business or entity]. Each month I upload [the documents]. Your standing job: [three to five responsibilities]. Keep a running log in this Project of [the things to track], and update it each month. At the end of each review, produce a one-page summary I can hand to [the next person or Project]. Speak plainly. Rules: [your negative rules]."

Then test it: "Read the [standard doc] in this Project, quote one line from it, and produce this month's summary in the format above." If it quotes a real line and holds the format, you're live.

Where it bites. Two honest limits. First, it is only as good as the context you give it and the review you keep doing. It can misread a statement or miscategorize a charge, so it flags and drafts, and I decide and send. Never let it move money or set a number unsupervised. Second, the memory is the whole point, so do not start a fresh chat every month. Same Project, same thread, and it compounds. Start over and you have rebuilt a form instead of a system. One more: resist dumping thirty files in. Under six, chosen for what they actually help retrieve, or you can't tell what's helping from what's hurting.

Same recipe, different function. I'm building the rehab side the same way: a Project that helps turn a walkthrough into a scope of work and a materials list. The function changes, the recipe doesn't. Once you've built one, you see them everywhere.

If you want a running start, I can share the anonymized instructions behind my Property Manager and CFO Projects, with my specifics swapped for placeholders you can drop your own into.

Know an operator drowning in monthly statements? Forward it. That's how this list grows. And hit reply: what's the first function you'd hand to a Project? I read every one.

Written by Paul Cobb, real estate operator (Louisville / Kentuckiana), forVEX builder. I run the plays I write about. More depth: forvex.app/journal