Investing & underwriting

Write your buy box into a Claude skill so the model stops guessing underwriting assumptions on an offering memorandum

Jake Heller of PLOT Daily underwrote the same eight-unit Los Angeles apartment deal twice in a community workshop, once by handing the offering memorandum to Claude in Excel with no other context and once with the firm's buy box loaded. The cold run built a clean five-year proforma but filled in assumptions on its own: a 5.75% exit cap rate where 6.25% is closer to honest for an older building in a softer submarket, 2% disposition costs where 5% is the working number, insurance near $1,000 a unit where $2,000 a unit is the reality on an old LA building, no interior renovation budget, and the broker's 86.3% loan-to-value. That cold run showed a 27.1% levered IRR; re-run at a realistic 65% loan-to-value, the same deal came back around 17%. The fix is a buy box file (return thresholds, asset criteria such as unit count, vintage and unit mix, target submarkets, deal-killing regulatory flags, and standard assumptions) packaged as a skill with three parts: an instruction file with the steps, an assets folder holding the firm's underwriting template, and a reference folder holding the buy box and standard template inputs. With the skill in place, attaching an offering memorandum triggers a fixed sequence: extract the deal facts, screen them against the criteria, populate the firm's own model, and return go or no-go, negotiate or walk. In the same session the team added base, upside, downside and stress scenario tabs and ran a target-price solver for an 18% levered IRR. Heller says one workshop model still carried a cash-flow error that the summary did not reveal, so every number needs review.

At a glance

Tools used
ClaudeClaude in ExcelClaude Skills
Category
Investing & underwriting
Audience
investor, broker
Difficulty
No-code setup
Source
The Buy Box File That Makes AI Underwriting Reliable
Jake Heller (PLOT Daily) · Article · Jun 23, 2026
First step
Spend a few hours writing your buy box down in one document: required returns, the unit count, vintage and unit mix you buy, target submarkets, the regulatory flags that kill a deal, and the standard exit cap, disposition cost, insurance per unit and loan-to-value you apply to every deal.

Compliance note: Treat the populated model as a first draft and trace the cash flows cell by cell before the model goes to an investment committee, a lender or a limited partner; turn off model training on your data in Claude's settings before loading the buy box, and leave the most confidential inputs out if the firm's policy requires it.

Suggested prompt

Written by this directory, inspired by the source. Replace the [bracketed] parts.

Use my underwriting skill. Read the buy box and the standard template inputs in the reference folder before you open the attached offering memorandum. Then: 1) extract the deal facts from the offering memorandum (price, unit count, unit mix, vintage, in-place rents, expenses, proposed financing); 2) screen those facts against every buy box criterion and list each pass or fail; 3) populate my underwriting template from the assets folder using my standard assumptions, never the broker's, and list every assumption you could not find in the buy box or the offering memorandum instead of filling it in; 4) give me a verdict of go, no-go, negotiate or walk, with the three numbers that drove the verdict; 5) add a scenarios tab with base, upside, downside and stress cases, and solve for the purchase price that reaches my required levered IRR.

Share on X Read the source

← All use cases