What is AI distressed note purchase underwriting? AI distressed note purchase underwriting is the use of large language models such as Claude and ChatGPT to evaluate a non-performing or sub-performing commercial mortgage you are buying at a discount to its unpaid principal balance, modeling every path to recovery, from a reinstated loan to taking title of the collateral. A loan-to-own buyer is not underwriting a building the way a fee-simple acquirer does. You are underwriting a debt instrument, the borrower behind it, and the legal road to control. This guide sits inside our broader work on AI deal analysis and goes deep on the note-buyer workflow specifically.
Key Takeaways
- Note buying underwrites a debt instrument, not a stabilized asset, so the analysis centers on the discount to unpaid principal balance and the cost and time to control the collateral.
- AI can ingest the full loan file, the note, mortgage, assignment chain, and guaranty, in minutes and build a structured summary that a human reviewer then verifies against source documents.
- The core output is a recovery waterfall that prices three exit paths: a reinstated or paid-off loan, a consensual deed-in-lieu, and a foreclosure that ends in ownership.
- Loan-to-own only pencils when the all-in basis after legal and carry costs is well below the stabilized value of the underlying property, with margin for execution risk.
- AI surfaces and organizes the math; state foreclosure law, bankruptcy risk, and title defects still require specialist counsel before you bid.
What Loan-to-Own and Note Purchase Underwriting Actually Requires
Note purchase underwriting requires you to answer one question before price: what do I own if the borrower never pays again? When you buy a distressed commercial mortgage, you acquire the lender's rights, the right to collect, to enforce default remedies, and ultimately to foreclose and take the collateral. The discount to par only matters in relation to the recovery you can actually realize and how long it takes. A note at 70 cents on the dollar of a $10 million unpaid principal balance costs $7 million, but your real basis is that figure plus legal fees, default interest you may never collect, property carry during the workout, and the opportunity cost of capital tied up for 12 to 24 months.
This is a different exercise from buying the asset outright. Our guide on AI distressed commercial real estate acquisition analysis covers sourcing distress signals and underwriting a broken capital stack at the asset level. Note buying narrows the lens to the loan position and the path to control, which is why it deserves its own underwriting workflow.
Building the AI Note Diligence Workflow
Start by having AI structure the loan file, because the document set, not a marketing flyer, is the asset. A disciplined workflow ingests the promissory note, the mortgage or deed of trust, the assignment chain, the loan agreement, any guaranties, the most recent servicing comments, and the borrower's last reported operating statements. Claude Opus 4.8 can read a multi-hundred-page loan file and produce a one-page term summary: original principal, current unpaid principal balance, note rate, default rate, maturity date, recourse versus non-recourse, and the specific default that has occurred.
The highest-value extraction tasks are the ones humans rush. Ask the model to confirm the assignment chain is unbroken from the originating lender to the current seller, to flag whether the guaranty is full recourse or a limited bad-boy carve-out guaranty, and to identify any lockbox, cash management, or reserve provisions that change who controls cash flow today. Every flagged item routes to counsel. The model accelerates the read; it does not render the legal opinion. For a structured scoring layer on top of these findings, pair this with our AI deal scoring frameworks for CRE investors that blend quantitative and qualitative inputs.
Pricing the Note: Discount to Par and the Recovery Waterfall
Price a distressed note by working backward from realizable recovery, not forward from a percentage of par. The recovery waterfall starts with the current value of the collateral as-is, applies the costs and timeline to convert your loan position into either cash or ownership, and discounts the result to present value. If the underlying property is worth $9 million as-is on a 7 percent cap rate against $630,000 of net operating income, and the unpaid principal balance is $10 million, you are already lending into a position where the collateral does not cover par. Your maximum bid is a function of that $9 million floor, less the cost to get there, less your required return.
AI helps by running the waterfall across a range of assumptions at once: collateral value plus or minus 10 percent, foreclosure timelines of 6, 12, and 18 months, and legal budgets that scale with whether the borrower contests. The model can compute your all-in basis and the implied yield-to-recovery for each branch, then present the distribution rather than a single point estimate. The discipline matters because distressed buyers who anchor on a headline discount, "I bought at 65 cents," routinely ignore that default interest and protective advances are rarely recovered in full. Express the answer as a basis per square foot or basis per unit so it compares cleanly against stabilized value.
Modeling the Three Exit Paths with AI
Every note has three realistic exits, and you must price all three before you bid. The first is a cure: the borrower reinstates the loan or refinances and pays you off at or near par. This is the fastest and highest-return outcome, but you do not control it, so you cannot underwrite to it. The second is a consensual resolution, typically a deed-in-lieu of foreclosure or a negotiated discounted payoff, where the borrower hands over the asset or a partial payment to avoid a fight. The third is a contested foreclosure that ends with you owning the property, the true loan-to-own path.
AI is well suited to model the time and cost of each branch because the variables are knowable. Have the model build a timeline-and-cost table: a cure path of 3 to 6 months with minimal legal cost, a deed-in-lieu path of 4 to 9 months with moderate cost and the risk of subordinate liens that survive, and a judicial foreclosure path that can run 12 to 24 months or longer in some states with full litigation budgets. The borrower-side mirror of this analysis, how a sponsor models a workout, is covered in our guide on using Claude for CRE loan modification and workout analysis on distressed debt; reading both sides sharpens your bid. Industry data from the Mortgage Bankers Association on commercial mortgage maturities and delinquency helps frame how many borrowers in a given vintage are likely to cure versus default.
The Loan-to-Own Decision: Do You Actually Want the Asset?
The loan-to-own decision turns on a question many note buyers skip: if foreclosure succeeds, do you actually want to own and operate this property? Taking title means inheriting the business plan, the capital needs, the tenant rollover, and the local market. A note that looks cheap on a recovery basis is a poor loan-to-own candidate if the asset is functionally obsolete, sits in a declining submarket, or needs capital you are not prepared to deploy. AI can pull the collateral's stabilized value, model the cost to reposition, and compare your all-in note basis against the value you could create as an owner.
The clean test is whether your basis after taking control, the purchase price of the note plus legal and carry costs, lands far enough below the property's achievable stabilized value to reward the execution risk you are accepting. CRE investors structuring these plays often work with The AI Consulting Network to stand up a repeatable note-underwriting model so every opportunity is scored the same way rather than evaluated on instinct. When the spread is thin, the discipline to pass is the most valuable output of the model.
What AI Cannot Do in Note Buying
AI cannot replace the legal and title work that decides whether a loan-to-own play succeeds or fails. State foreclosure procedure varies widely between judicial and non-judicial states, redemption rights differ, and a single defect in the assignment chain can stall enforcement for months. Bankruptcy is the other wildcard: a borrower filing for Chapter 11 the day before a foreclosure sale can impose an automatic stay that resets your entire timeline. No model prices the litigation behavior of a motivated, well-counseled borrower.
Treat AI as the analyst that compresses a week of file review into an afternoon and keeps your recovery math consistent, then route every flagged legal issue, title exception, and enforcement question to specialist counsel before capital is committed. Investors who want help building that human-plus-AI note workflow can reach out to Avi Hacker, J.D. at The AI Consulting Network, where the focus is exactly this kind of disciplined, repeatable underwriting.
Frequently Asked Questions
Q: How is buying a distressed note different from buying the property directly?
A: When you buy a note you acquire the lender's rights and remedies, not the real estate. You only reach the property through a cure, a deed-in-lieu, or a completed foreclosure, so your underwriting must price the time, cost, and legal risk of converting a loan position into either cash or ownership.
Q: What discount to par should I expect on a distressed CRE note?
A: There is no universal number. The right price is whatever leaves your all-in basis, the purchase price plus legal and carry costs, far enough below realizable collateral value to reward the risk. Headline discounts like 65 or 70 cents on the dollar are meaningless without the recovery waterfall behind them.
Q: Can AI tell me whether a borrower will fight foreclosure?
A: No. AI can model the cost and timeline of a contested versus consensual outcome, but it cannot predict litigation behavior. Treat the contested foreclosure timeline as your base case and any faster resolution as upside, and confirm enforcement strategy with counsel in the relevant state.
Q: Which AI tools are best for reading a loan file?
A: Long-context models such as Claude Opus 4.8 handle full loan files well because they can read the note, mortgage, and guaranty together and cross-reference terms. Always verify extracted terms against the source documents before relying on them for a bid.