Prompting for the Hard Side: AI Workflows for Distressed Assets

By W. Kyle Tangney, Britt Gillespie & Lacey Willard | The New CRE Edge

‎‎

In most of commercial real estate, moving fast is an advantage. In distressed work, it’s the whole game.

Timing is often the difference between recovering value and watching it evaporate, and every call has to hold up to scrutiny later.

That mix, speed under pressure and decisions you can defend, is where structured AI earns its keep. Here are three of the moments we lean on it most.

One note before those three, because it matters more here than anywhere else in this series. Everything below runs inside a permissioned environment, on files we are already authorized to handle. Borrower correspondence and loan documents do not go into public tools. In distressed work the file is the evidence, and how you handled it is part of what has to hold up later.

Deciding Which Asset Is Actually on Fire | W. Kyle Tangney

The hardest part of a distressed book is deciding which asset gets your attention this week. You’ve got forty-plus situations, a couple of analysts, and a natural pull toward whoever emailed last. The loudest borrower is rarely the biggest risk to recovery.

So I run the whole book through a triage prompt before I decide where anybody spends their time. It ranks every asset on default severity, likely recovery range, market liquidity, and time to resolution. I still make the calls. The ranking just points me at the six assets where attention actually moves the recovery math, while the six that happen to be shouting wait their turn.

Advanced Prompt Example

You are a senior special servicer with deep experience in distressed CRE loan resolution. Using the attached loan summaries, property financials, appraisals, market reports, and borrower correspondence for a portfolio of assets, triage each asset on five dimensions: default severity and legal standing, estimated recovery range under REO, note sale, and modification, market liquidity for disposition, and estimated time to resolution. Return a prioritized action matrix with the recommended resolution path, urgency, next step, and responsible party. Rank by where focused attention would most improve recovery, not by which borrower is loudest.

Run inside a permissioned workspace on servicer-held files. The same applies to the two prompts below. Nothing in these belongs in a public tool.

‎ ‎

Choosing the Resolution Path Without Guessing | Britt Gillespie

REO, note sale, or modification. Get that call wrong and real money comes off the trust’s recovery. The paths are hard to compare because they carry different timelines, costs, and risks, and the right answer is rarely obvious from the file.

I model all of them against one framework before I commit. It stress-tests pricing and exit timing, then hands back a ranked recommendation with the conditions that would change the ranking. The output gives me the foundation to make a call I can put in front of the trust and stand behind. That standard covers the inputs too. The loan file, the appraisals, and the borrower’s financials stay inside our environment, handled the way the trust would expect them to be handled.

Advanced Prompt Example

You are a distressed asset resolution analyst. Using the attached appraisals, loan documents, operating statements, market comparables, and estimated holding and legal cost schedules, model these resolution paths for the asset: REO disposition at 6, 12, and 18 month exits, note sale at par, 85 cents, and 75 cents on the dollar, loan modification with an extended term and rate relief, and deed-in-lieu with an agreed disposition timeline. For each, calculate estimated net recovery to the trust, time to resolution, execution risk, and the key assumptions. Conclude with a ranked recommendation and the market or borrower conditions that would change the ranking. Be conservative and show your reasoning

‎ ‎

Writing the Decision Down While You Still Remember Why | Lacey Willard

In distressed work, you have to be able to show how you got there. Almost any resolution decision can get second-guessed later, by the trust, by investors, sometimes by a regulator, usually months after the fact when the context has faded and only the outcome is left. Reviewers are starting to ask how the analysis was produced, and where the file went while you produced it. That answer belongs in the same record.

So we use AI to turn the analysis into a clean decision memo the moment the call is made: what we considered, what we assumed, why this path, and what would have changed our minds. The value here is durability. When someone questions the decision nine months from now, the reasoning is already on paper, and it holds.

Advanced Prompt Example

You are preparing a resolution decision memo for a distressed CRE asset that may be reviewed later by the trust, investors, or regulators. Using the attached scenario analysis, appraisals, financials, and correspondence, document the recommended resolution path as a clear, defensible narrative. Include the paths considered, the key assumptions behind each, the basis for the recommendation, the primary risks, and the conditions that would have changed the decision. Write it so someone reviewing it months from now, without the original context, can follow and pressure-test the logic. Keep the tone neutral and factual.

The same structure carries over to the other pressure points in a resolution, borrower capacity before a workout and disposition timing in a moving market. Different inputs, same discipline.

‎ ‎

A Few Things You Can Start Doing

• Triage by recovery impact. Let the loudest borrower of the week wait their turn.
• Make every scenario show its assumptions and what would change the ranking.
• Write the decision memo while you still remember why. Defensibility lives in a document.
• Keep the servicer’s judgment in charge. The workflow organizes the file and you resolve the asset.
• Run it where the file already lives. If a document needs permission to sit in the servicing file, it needs the same permission before it goes near a model

That’s the through-line of this series: AI to win and move deals, to protect the value of what you own, and here, to claw back value when the stakes are highest and the clock is running.

‎ ‎

This is the kind of work we teach in Greysteel’s CRE MBA program, built from real resolutions. The first cohort opens the second week of June. If you resolve distressed assets for a living, this is the highest-stakes place to get AI right. Getting it right includes knowing what never leaves your environment, which is a large part of what we teach. Reach out or keep an eye on the site.