How I use AI to analyze judicial reorganizations

A judicial reorganization case can run to thousands of pages once you add up the plan, the amendments, the creditors' meeting minutes and the judge's rulings. What decides whether a credit is worth buying usually fits in one paragraph of one of those documents. My job is to find that paragraph before making an offer.

I am a corporate transactions lawyer, and my routine has always differed from my litigation colleagues': I do not track procedural deadlines or a docket with daily filings. I analyze judicial reorganizations to support investment decisions, such as buying or extending credit and structuring the deal. Much of my time went into gathering scattered information and separating what changes the investment thesis. That is the part of the work where AI started to make a difference.

Three rules I follow today:

1. Every answer from the AI goes back to the original document before it enters the numbers.
2. Instead of asking for a summary, I ask which document defines each condition and how it changed.
3. I write down my criteria before delegating any step.

What I need to know before buying a credit

When I evaluate buying a credit in judicial reorganization, I want to know how much I will receive, and when. The answer is in the reorganization plan, which sets out for each credit:

• the class: the group of creditors the credit belongs to, and each group has its own payment terms;
• the recognized amount: how much of the credit was officially accepted in the case;
• the haircut: the discount the plan applies to that amount;
• the grace period and the term: how long until the first payment and how many installments follow;
• the adjustment index: how the amount is updated while payment is pending;
• the collateral: whether any asset, such as a property, still secures payment.

The hard part is that this information is almost never in one place. Imagine an unsecured credit offered at a discount. The approved plan pays that class with a 50% haircut over ten years. An amendment voted months later raises the haircut to 70% and adds a two-year grace period. Later, the judge strikes down the clause that set the adjustment index. Anyone who reads only the plan calculates a return that no longer exists.

How I use Solon and Claude

I use two tools, at different moments. Solon reads the entire case file and shows where each piece of information came from. Claude, Anthropic's AI assistant, helps me think through the deal.

In Solon I find the information I am looking for and cross-check one document against another. In the unsecured credit example, that is where I rebuild the sequence: the plan, the amendment and the ruling, each with a reference to the document it came from.

With that base in place, I move to Claude and test acquisition prices, terms and payment hypotheses. I see how much each variable moves the expected return. When an assumption starts to weigh on the decision, I go back to the case file and check where it came from before adding it to the scenario.

The legal department at Century Communities worked in a similar way on an M&A deal with 87 land contracts. The team used AI for the first read of the contracts and to organize due diligence, without losing access to the source documents. According to Jarret Coleman, the company's general counsel, in a Thomson Reuters case study, a task that took an hour came down to five minutes or less.

The original source validates the answer

The first rule I adopted was to treat every answer from the AI as a starting point and check each one against the original document.

In an insolvency proceeding I read the plan, amendments, creditors' meeting minutes and court rulings. A misreading in any of them hits the investment decision directly: in the example above, missing the amendment would be enough. AI finds and organizes the information spread across those documents well. Deciding what it means for the viability of the case is still my job.

That is why I only work with AI that shows where it got each piece of information. If a fact goes into an investment assumption, I need to reach the document it came from, read the passage in context and check whether anything changed later.

Summarizing the case file is rarely enough

Over time I realized that summarizing documents solved little of my problem.

In a judicial reorganization I usually look for one specific fact that confirms or breaks an assumption: the payment terms of a class, a piece of collateral, a change in the new version of the plan or a ruling that affects the structure planned for the deal. In long proceedings I spent much of my time just finding that fact, before I could start analyzing it.

So I changed my questions. Side by side, the difference looks like this:

Question that yields little	Question that works
"Summarize the reorganization plan."	"Which document sets the haircut for the unsecured class, and on what page?"
"What are the risks of this credit?"	"Did any amendment or ruling change the payment terms for that class after the plan was approved?"
"Explain the payment terms."	"Which index adjusts this credit today, and which document does that rule come from?"

The questions on the right point to a document, and that is why I can check the answer.

Delegating requires explicit criteria

The second rule was to write down the criteria I used to apply by intuition.

To build an investment scenario from the case file, I need to know which assumptions call for more attention, which clauses tend to generate litigation and what care a bid needs to reduce successor liability risk. I also need to recognize when a piece of information stops being routine and calls for review.

A generic request such as "analyze this plan" yields little in this work. Delegation started to work once I told the AI what to look for, what to connect, what to flag and when to stop and call me.

Writing those criteria down also changed my method. I now keep apart three stages that used to blur together: establishing the facts, doing the legal analysis and building the economic scenarios.

What changed in my routine

The most visible gain was time. Gathering information, comparing plans and sketching the first scenarios used to take much of the analysis window. Now they take a fraction of it.

Large firms report the same effect. According to Microsoft, Husch Blackwell, with more than twenty offices across the United States, saved 8,800 lawyer hours using Copilot, Microsoft's AI assistant, to summarize meetings, analyze documents and draft correspondence. At DLA Piper, the operational and administrative teams saved up to 36 hours per week on content production and data analysis. Those routines differ from mine, but there too the time came out of reading, summarizing and drafting.

With the time I get back, I test more payment hypotheses before concluding whether a deal is viable. When buying a credit, I compare entry prices, payment conditions and terms, and assumptions about compliance with the plan, before forming an opinion.

Writing down the criteria is what let me delegate the search and organization of documents. I still read the paragraph that decides the purchase myself, in the original document. I just get to it faster.

References

• DLA Piper, Microsoft
• Husch Blackwell, Microsoft
• Century Communities, Thomson Reuters

Frequently asked questions

Does AI replace the lawyer in a judicial reorganization analysis?

No. AI retrieves and organizes information scattered across hundreds of documents, and the interpretation of what that means for the viability of the case remains with the lawyer. The AI's answer is a starting point to be verified against the original source, never a conclusion.

What should be checked before acquiring a distressed credit?

How the plan treats that credit: class, recognized amount, haircut, grace period, term, adjustment index and the collateral that still holds. That information is rarely in one place. A condition can originate in the plan, change in an amendment and later be affected by a court ruling.

Why is summarizing the case file rarely enough?

Because what decides an investment thesis is usually one specific point rather than the overview: a condition applying to a given class, a piece of collateral, a change in a new version of the plan. Instead of asking for a summary, the useful question is which document defines that condition and how it changed over the course of the case.