The AI Workflow for Mortgage Note Research

AI Belongs in Your Due Diligence Workflow

Due diligence on mortgage notes is data-heavy work. Every loan on a tape arrives as a row in a spreadsheet — columns of unpaid principal balances, property addresses, lien positions, occupancy flags, charge-off dates, and more. Before you can price a single asset, you need to parse that data, research each property and borrower, compare economic indicators across geographies, and model potential outcomes. Multiply that by twenty or fifty loans on a tape and the hours add up fast.

AI tools — specifically large language models like ChatGPT — can compress significant portions of this workflow. Not by replacing your judgment, but by handling the repetitive data manipulation, research aggregation, and document generation that consume the bulk of your analysis time. The result is a faster path from receiving a tape to submitting a letter of intent, with more data points informing your decisions along the way.

What follows is a five-step AI workflow for mortgage note research. Each step builds on the previous one, creating a layered analysis that moves from raw data to actionable intelligence. None of these steps require paid plugins or third-party integrations — they work with the base capabilities of a general-purpose AI assistant.

Step 1: Summarize the Loan Tape at a Glance

The first thing most investors do when they receive a tape is scroll through it horizontally, trying to absorb dozens of columns of data across dozens of rows. It is an inefficient way to get a high-level picture of what you are looking at.

Instead, paste the tape data directly into an AI assistant and prompt it to analyze the assets. A simple instruction — "analyze these assets" — produces a structured summary that surfaces the key characteristics of the pool in seconds:

• Aggregate statistics — total fair market values, highest and lowest UPB balances, average loan size
• Loan characteristics — performing vs. non-performing, lien position breakdown (firsts vs. seconds), occupancy status distribution
• Geographic concentration — which states and metro areas are represented and how the pool is distributed
• Date ranges — charge-off dates, origination dates, and last payment dates that tell you how seasoned the delinquencies are
• Property types — single-family, multi-family, condo, or mixed

This high-level summary serves the same purpose as the first-pass screening you would do manually — identifying obvious patterns, red flags, and concentrations before diving into individual loans. The difference is that the AI produces it in seconds rather than the fifteen to thirty minutes it takes to manually scan and mentally aggregate a spreadsheet.

You can refine this initial analysis by providing more specific prompts. Instead of a generic "analyze," tell the system what you care about: "Summarize these assets with a focus on equity coverage, geographic risk, and lien position." The more context you provide about your investment criteria, the more targeted the output becomes.

Step 2: Reorganize Data at the Loan Level

Spreadsheets are built for horizontal data — one row per loan, one column per attribute. That format works for sorting and filtering, but it is not how most people process information when evaluating individual loans. Reading across thirty columns to understand a single asset is cognitively expensive, especially when you are comparing multiple loans side by side.

The second step in the AI workflow is to ask the system to summarize each asset individually in a vertically oriented format. Instead of one row with thirty columns, you get a compact card for each loan that lists all relevant attributes in a readable, top-to-bottom layout:

This reorganization accomplishes two things. First, it makes each loan digestible as a standalone investment opportunity — you can see all the key parameters at once without scrolling. Second, it creates a format that is easy to copy back into Excel or Google Sheets for further analysis if you prefer to work in a spreadsheet environment for your final pricing.

One practical note: AI systems often truncate long outputs to conserve resources. If you paste a tape with thirty loans and the AI only summarizes six, prompt it to continue with the rest. It will pick up where it left off.

Step 3: Enrich with Census and Demographic Data

Raw tape data tells you about the loan and the property. It tells you almost nothing about the market the property sits in — the local economy, population trends, household income levels, or employment conditions. These macro factors directly influence your resolution probability and timeline. A non-performing loan in a growing metro area with strong employment has a fundamentally different risk profile than the same loan in a shrinking rural market, even if the loan-level numbers look identical.

AI models trained on public data — including U.S. Census data — can layer demographic context onto your loan-level analysis without requiring you to manually research each location. Prompt the system to incorporate census data into the loan-level summaries, and it will append information such as:

• Median household income for the property's ZIP code or metro area
• Unemployment rate relative to state and national averages
• Population growth or decline over recent census periods
• Average household size and demographic composition
• Median age of housing stock in the area

This demographic enrichment adds a dimension of analysis that many note investors skip entirely during the pre-bid phase because the manual research is too time-consuming to justify on every loan. With AI, the marginal cost of pulling this data drops to near zero, which means you can evaluate market fundamentals on every asset in the pool — not just the ones you have already decided to bid on.

Where Census Data Changes Your Analysis

The demographic layer is particularly valuable for three scenarios:

Comparing loans across unfamiliar markets. If you are evaluating a tape with properties in eight different states, you may know three of those markets well and five not at all. Census data gives you a baseline for the unfamiliar markets that prevents blind spots in your pricing.

Identifying collection viability. A market with strong median income and low unemployment suggests a borrower population that is more likely to engage in a loan modification or loss mitigation workout. A market with declining population and weak employment suggests a longer timeline to resolution and a higher probability of foreclosure as your exit.

Supporting or questioning property values. Census trends provide context for whether a broker price opinion or AVM estimate is reasonable. A property valued at $200,000 in a ZIP code with declining population and below-average incomes deserves more scrutiny than the same value in a growing market with strong fundamentals.

A Note on Data Freshness

AI models have training data cutoffs, which means the census data they reference may not reflect the most current figures. As of this writing, most models incorporate data through the 2020 Census and American Community Survey updates through 2022. For pre-bid screening purposes, this is sufficiently current — you are looking for macro trends, not real-time snapshots. For final pricing decisions, supplement AI-generated demographic data with current sources such as the Bureau of Labor Statistics, Census Bureau QuickFacts, or paid data providers.

Step 4: Score Locations for Collection Viability

With loan-level data and census context in hand, the next step is to move from description to evaluation. Ask the AI to score each location represented in the tape based on the viability of collections and resolution.

The system will generate a scoring framework using factors such as:

The AI assigns each location a composite score and provides a rationale for the rating. For example, a property in Englewood, New Jersey might score highly because of affluent demographics, strong property values, and proximity to New York City employment centers. A property in a small rural Michigan market might score lower due to a smaller economy, limited buyer pool, and longer foreclosure timelines.

How to Use Viability Scores

These scores are not definitive rankings — they are screening tools that help you prioritize where to spend your deeper due diligence time and budget. A high viability score does not mean you should bid aggressively. A low score does not mean you should pass. The scores identify which loans deserve closer attention and which ones carry location-level risk that needs to be priced into your model.

The practical workflow is to export the scored data back into your spreadsheet and use the viability scores as an additional column alongside your financial metrics. Sort by score to see your strongest and weakest locations at a glance, then allocate your BPO budget and research time accordingly.

You can also customize the scoring parameters. If your strategy focuses on loan modifications rather than foreclosure, tell the AI to weight employment and income factors more heavily. If you are a foreclosure-oriented investor, weight state legal timelines and property liquidity. The scoring system adapts to whatever criteria you define.

Step 5: Generate Letters, Agreements, and Amortization Schedules

The final step in the AI workflow extends beyond due diligence into the operational side of note investing — using AI to draft the correspondence and financial documents that move a deal forward.

Letters of Intent

Once you have identified the loans you want to bid on, you can prompt the AI to draft a letter of intent using the data from your analysis. A well-structured LOI includes:

• Identification of the assets — loan numbers, property addresses, and borrower references
• Proposed purchase price — your bid on each loan or the pool
• Conditions of purchase — contingencies for title review, collateral file verification, and satisfactory due diligence
• Due diligence timeline — the exclusive period you are requesting to complete your analysis
• Funding timeline — when you expect to wire funds if the deal proceeds

The AI generates a professional draft that you can review, edit, and customize before sending to the seller. This is not a substitute for having your attorney review the final document, but it eliminates the blank-page problem and produces a structured starting point in seconds.

Loan Modification Agreements and Correspondence

On the collections side, AI can draft borrower outreach letters, modification term sheets, and workout proposals. If you have negotiated terms with a borrower — say, a modified balance at a specific interest rate over a defined term — you can prompt the AI to generate the formal agreement language and calculate the payment schedule.

Amortization Schedules

Amortization schedules are a core tool in loan modification negotiations. When you present a borrower with proposed modification terms, showing them exactly how each monthly payment breaks down into principal and interest over the life of the loan builds transparency and trust.

Prompt the AI with three inputs — the modified principal balance, the interest rate, and the term in months — and it will generate a full amortization schedule showing:

• Monthly payment amount
• Principal portion of each payment
• Interest portion of each payment
• Remaining balance after each payment

For example, a $23,000 balance at 9.9% interest over a 20-year term produces a monthly payment of approximately $213. The AI calculates every row of the schedule — all 240 months — and presents it in a table you can copy into a spreadsheet or PDF for the borrower.

This calculation takes the AI seconds. Doing it manually in Excel is straightforward for experienced investors, but the AI approach is faster and eliminates formula errors, especially when you are modeling multiple modification scenarios with different rates and terms.

Prompt Engineering: Getting Better Results

The quality of AI output is directly proportional to the quality of your input. A vague prompt produces a vague response. A specific, context-rich prompt produces targeted, actionable analysis. Here are principles that improve results in the note research context:

Provide your investment criteria up front. Tell the system you are a second-lien non-performing note investor focused on owner-occupied properties in judicial foreclosure states. This context shapes every subsequent response.

Be specific about output format. If you want a table, ask for a table. If you want loan-level summaries, specify that. If you want the data organized for copy-paste into Excel, say so.

Iterate and refine. The first response is rarely the final answer. If the AI summarizes only a subset of your loans, ask it to continue. If the scoring factors do not match your strategy, redefine them. If the census data is too general, ask for ZIP-code-level specifics.

Correct errors directly. AI models occasionally produce inaccurate data points — a wrong unemployment rate, a misidentified state, or a calculation error. When you spot an issue, correct it in your prompt and ask the system to regenerate. Do not assume subsequent outputs are accurate just because the first one looked reasonable.

Chain your prompts. Each step in this workflow builds on the previous one. The AI retains context from earlier in the conversation, so your census-enriched loan summaries carry forward into the scoring step, and the scored data informs the LOI draft. Work sequentially within a single session to maintain that context.

What AI Cannot Do for You

AI accelerates data processing and document generation. It does not replace the judgment calls that define successful note investing. Specifically:

• AI cannot verify property condition. No model can tell you whether a roof is leaking or a foundation is cracked. You still need BPOs, drive-by inspections, and Google Street View.
• AI cannot confirm title status. Chain-of-title issues, undisclosed liens, and recording errors require a professional title search.
• AI cannot assess borrower willingness to engage. Census data and scoring models estimate probability, but whether a specific borrower will pick up the phone and negotiate a modification is unknowable until you try.
• AI cannot replace legal counsel. Drafting a letter of intent or modification agreement with AI is a starting point, not a finished product. Have your attorney review any document before it binds you contractually.
• AI training data has a cutoff. Market conditions, interest rates, and local economic data change. Cross-reference AI-generated data points with current sources before making final pricing decisions.

The value of AI in note research is not that it makes decisions for you. The value is that it handles the data manipulation, formatting, enrichment, and drafting work that previously consumed hours of your time — freeing you to focus on the analysis, negotiation, and relationship-building that actually drive returns.

Putting the Workflow Into Practice

Here is the complete five-step workflow summarized as a repeatable process:

Each step takes minutes. The full workflow — from raw tape to scored, enriched analysis with draft correspondence — can be completed in under an hour for a typical pool of twenty to thirty loans. Without AI, the same depth of analysis would take a full day or more.

The investors who integrate AI into their research process are not cutting corners. They are covering more ground, evaluating more data points, and arriving at better-informed bids — all while spending less time on the mechanical work that used to define the due diligence process. The competitive advantage is not the AI itself. It is the additional analysis you can perform because the AI freed up your time to do it.