The Moment Before Bankruptcy
Bankruptcy law has a timing problem. A company may still be paying its debts, no petition may have been filed, and no court may be involved. Yet a lender’s credit model may already have detected deterioration in that company’s risk profile. A supplier may be reassessing payment terms. An insurer may be repricing exposure, while a major customer may be quietly considering whether the company remains a reliable counterparty. None of these decisions requires a bankruptcy filing or any formal legal recognition of distress.
The gap matters because prediction can change the condition it predicts. A lender that reduces exposure affects liquidity. A supplier that shortens payment terms does the same. A customer that begins shifting orders elsewhere changes expected revenue. Each actor may simply be managing its own risk. But a company that eventually reaches Chapter 11 may arrive there with its commercial position already altered by decisions made well before bankruptcy law entered the picture.
In earlier work, I used the term Algorithmic Shadow Insolvency to describe the predictive gap in which financial distress begins shaping restructuring decisions before those decisions become legally visible.[1] This article looks at that gap from a practitioner’s perspective: What changes when commercially important actors can see, price, and act upon financial deterioration before a bankruptcy case begins? The concept does not suggest that an algorithm can declare a company insolvent, nor does it propose another legal test for insolvency.
From Prediction to Commercial Consequence
Credit models have long influenced lending decisions. What is changing is not simply their sophistication. Risk can increasingly be reassessed through model-based processes and translated into commercial decisions before the events that traditionally make distress unmistakable. That question is becoming more immediate as model-based risk assessment moves deeper into ordinary financial decision-making. U.S. banking regulators’ revised 2026 model risk guidance reflects how embedded quantitative models have become in significant banking activities.[2] Bankruptcy-prediction research likewise shows that machine-learning methods can improve predictive performance over traditional approaches.[3]
For restructuring lawyers, however, prediction accuracy is only half the question. The harder problem begins when prediction changes behavior. A lender may reduce exposure without terminating a facility. A supplier may shorten terms without ending the relationship. An insurer may reprice risk. A customer may diversify its supply chain before canceling a contract. Each decision may be rational in isolation. Taken together, they can progressively reduce the company’s room to maneuver.
This is where Algorithmic Shadow Insolvency differs from the familiar idea of an early warning system. Early warning generally asks whether distress can be identified soon enough for the company, its advisers, its creditors, or public authorities to respond. Predictive systems operating throughout a company’s commercial network create a different possibility.[4] The company may not control those systems, know which signals they are detecting, or even realize that counterparties have begun to react. The warning may no longer be delivered to the company in distress. It may be delivered to everyone around it.
That distinction matters in Chapter 11. Filing activates legal mechanisms that can protect the debtor and support continued operations, including the automatic stay, debtor-in-possession authority, and access to postpetition financing subject to statutory requirements.[5] But formal protection does not recreate every commercial condition that existed before filing. If credit has tightened, suppliers have reduced exposure, customers have begun diversifying, or strategic capital has become hesitant, the legal process inherits those changes. The filing date remains legally decisive. Commercially, it may no longer be the beginning of the story.
Protecting the Restructuring Window
The practical response is not necessarily to file earlier. That would confuse predictive information with the decision to invoke a formal bankruptcy process and, in many cases, miss the problem entirely. The more immediate question is whether the company is losing viable restructuring options while management and counsel are still measuring distress principally through liquidity, covenant compliance, debt maturities, defaults, and other familiar indicators.
A different approach begins with the reactions surrounding the company. Shorter payment terms, reduced credit availability, additional collateral demands, insurance repricing, customer diversification, or hesitation by strategic investors may each be ordinary commercial events. They should not automatically be treated as evidence of impending bankruptcy. But when several appear together around a financially pressured company, treating each as an unrelated inconvenience may obscure something more important: The company’s risk is being reassessed outside the company itself.
A restructuring lawyer does not need access to a lender’s proprietary model or a supplier’s risk system to recognize that shift. The algorithm may be invisible; its commercial footprint is not. Counsel can observe changes in counterparty behavior and ask whether they form a broader pattern. The relevant inquiry is whether those changes are beginning to affect liquidity, operating continuity, or the range of restructuring choices still available.
That distinction matters because restructuring options do not disappear at once. They narrow. A financing source becomes more expensive. A critical supplier becomes less patient. A customer becomes less willing to depend on the company. Strategic capital becomes harder to secure. None of these developments necessarily requires Chapter 11. But each can make a restructuring that later becomes necessary more difficult to execute.
The appropriate response is therefore preparation rather than panic. Counsel and management can identify the relationships most capable of affecting continuity, determine whether adverse changes are isolated or cumulative, preserve liquidity, engage critical counterparties while confidence can still be stabilized, test alternative financing, and prepare restructuring options before those options become materially narrower.
The objective is not to file earlier. It is to recognize earlier when the ability to restructure is beginning to erode.
In practical terms, that means three things: detect the reaction, map the pattern, and preserve optionality. This does not turn every adverse commercial signal into a restructuring event, and it does not require lawyers to become data scientists. It asks restructuring counsel to do something much more familiar: understand the company as a business operating within a network of relationships and recognize when those relationships begin changing faster than the traditional indicators of distress.
The Question for Chapter 11
Chapter 11 will continue to have a legally identifiable beginning. Algorithmic Shadow Insolvency concerns the period before it—that is, the weeks or months during which commercially consequential responses may accumulate without a filing, judicial supervision, or a formal restructuring process. That period matters precisely because meaningful choices may still exist within it.
The point is not to give predictive models legal authority, nor to treat every algorithmic warning as evidence of insolvency. The more immediate implication is for restructuring practice. If the commercial environment surrounding a company can recognize changing risk earlier, counsel may need to ask not only whether a filing or other restructuring step is approaching but also whether important counterparties have already begun behaving as though the company’s risk has changed.
That suggests a different form of early intervention. It does not begin with a petition. It begins with preserving the conditions under which a restructuring can still work. By the time distress is obvious enough to force the filing question, some of those choices may already be gone.
Esin Civelek, Algorithmic Shadow Insolvency: The Predictive Turn in European Restructuring Law, SSRN (May 5, 2026). ↑
Off. of the Comptroller of the Currency, Bd. of Governors of the Fed. Rsrv. Sys. & Fed. Deposit Ins. Corp., OCC Bulletin 2026-13: Model Risk Management: Revised Guidance (Apr. 17, 2026). ↑
Flavio Barboza, Herbert Kimura & Edward Altman, Machine Learning Models and Bankruptcy Prediction, 83 Expert Sys. with Applications 405, 405–17 (2017). ↑
Jing Wu, Zhaocheng Zhang & Sean X. Zhou, Credit Rating Prediction Through Supply Chains: A Machine Learning Approach, 31 Prod. & Operations Mgmt. 1613, 1613–29 (2022). ↑
11 U.S.C. §§ 362, 364, 1107–1108; U.S. Courts, Chapter 11—Bankruptcy Basics (last visited Sept. 30, 2026). ↑
