A free playbook by The Aigent Lab

Underwriting & Analysis

Underwriting & Analysis · Playbook 09

Your due-diligence analyst. Before you sign.

Buyers hand a seller or developer years of payments based on a brochure. Build a Claude Project that pulls market reports, the contract, the payment terms, and the counterparty's actual track record into one grounded analyst, then read the bull case and the bear case back to back before anyone signs.


Why this matters.

Buyers pay in stages over years before the deal fully closes, against materials designed to sell, not to inform. The ways a buyer actually loses money are well known: delay, spec or scope downgrade, oversupply at completion, and counterparty default. None of them appear in the marketing.
The brochure is the marketing. The numbers are the investment.

Lock it down. Then split the case in two.

01

Create the project and lock it

Paste instructions telling the model to work only from uploaded documents and refuse to speculate about numbers or dates it hasn't been given.

02

Upload market context

Recent quarterly market reports from real research houses, so the model has real comparables instead of memory.

03

Upload the counterparty paper trail

The contract, payment schedule, and the counterparty's last five completed projects with promised vs actual delivery dates.

04

Ask for the bull case, then the bear case

Run both memos back to back. If the bear case is stronger than the bull case, you've already saved the deposit.

The bear case prompt
Using only the documents in this project, write a bear case memo covering how a buyer typically loses money on a deal like this: delivery delay, spec downgrade, oversupply at completion, and counterparty default. Cite the source document and page for every claim. End with what you cannot verify from these documents.

What it still gets wrong.

i.

It doesn't visit the site

Construction quality, access, and the finished feel of a place have no document substitute. A site visit still matters.

ii.

It can't see what isn't disclosed

Undisclosed issues or quiet changes won't show up. Treat absence of a risk as inconclusive, not reassurance.

iii.

Bad inputs poison the analysis

Upload only the marketing brochure and you get the brochure's narrative back in memo form.

iv.

It's not financial advice

A client's lending, tax position, and existing exposure aren't in the project. The memos are research; the decision is still theirs.


Next Playbook

The five stages of AI, and where most agents are stuck

Read Next ↗