Neural network gave Sberbank a 67% chance of winning a 1.44 billion ruble court case

The specialized AI tool "neShemyaka!" (neshemyaka.ru) put Sberbank's chances of overturning, on cassation, a ruling to recover 1.44 billion rubles at 67%, according to figures cited by Habr in a piece on case No. A40-166729/2024, heard by the Intellectual Property Court. While the legal community debated the high-profile twists in the case, the authors decided to "change the optics" and add tensor computations to the legal argumentation.

What is known about the case and the assessment tool

• Case number: No. A40-166729/2024
• Court: Intellectual Property Court
• Amount in dispute: 1.44 billion rubles
• Estimated chance of Sberbank overturning the recovery on cassation: 67%
• Tool: the neural service "neShemyaka!" (neshemyaka.ru)

The tool's name, "neShemyaka!", alludes to the expression "Shemyaka's court" (shemyakin sud), a historical symbol in Russian culture of an unjust, biased ruling. Giving such a name to a neural service that assesses court prospects sets an ironic tone for the piece, playing on the contrast between the folkloric reputation of an unfair court and the tool's claim to objective calculation. The authors describe how the tool "coolly" assessed the prospects of appeal, meaning without the reputational or emotional factors that matter to lawyers, from the defendant's side, that is, Sberbank's, and arrived at a 67% probability of overturning the billion-ruble recovery at the cassation level.

How does the AI assessment of the case's prospects compare with reality?

The text gives away the result as a "spoiler": according to the authors, the neural network "hit the bullseye," meaning its forecast was confirmed by how the case actually unfolded. The piece also describes a broader trend in legal practice: the use of predictive neural tools to put a number on the chances of success in a dispute, alongside traditional lawyers' analysis rather than in place of it. Such tools are trained on large bodies of court decisions and, from formal features of a case (the court instance, the category of dispute, the composition of the panel, earlier rulings in similar cases), produce a probabilistic estimate of the outcome.

That estimate can serve as an extra argument when deciding, for example, whether to appeal further, accept a settlement, or prepare for the worst-case scenario in advance. Importantly, such an assessment does not replace substantive legal analysis: it does not read the case file the way a lawyer does, but works with statistical patterns that are hard to catch by hand, especially when it comes to how a particular court instance behaves across many similar disputes.

Why this matters for the legal technology market

The appearance of such forecasting services in public discussion of large corporate disputes, such as the case involving Sberbank and 1.44 billion rubles, shows that AI-based predictive analytics is moving out of the category of experimental legal-tech and becoming a tool that a wider audience, including observers outside the legal field who follow case law, is interested in. For the legal market, this is another signal: a quantitative assessment of a dispute's prospects, which used to be the intuition of individual expert lawyers, is gradually being formalized and turning into a reproducible metric that can be cited when choosing a procedural strategy.

The story of case No. A40-166729/2024 itself may become one of the first widely discussed precedents of this kind in the Russian-speaking legal community.

It is also notable that the case was drawn specifically from the practice of the Intellectual Property Court, a specialized instance that hears disputes where the body of precedent and the terminology are especially hard for a non-specialist. Showing that a neural tool can give a quantitatively verifiable forecast even in such a narrowly specialized case works as an argument for wider use of such solutions, not only in disputes involving large banks but across the rest of corporate litigation, where the stakes and the complexity of argumentation are comparable to case No. A40-166729/2024.

The ironic tone of the piece, from the service's name "neShemyaka!" to the phrase about "changing the optics" and adding tensor computations to legal argumentation, does not negate the substance of the publication. The authors effectively invite readers to look at a high-profile case not only through classical legal analysis but also through a quantitative probabilistic estimate produced by an algorithm independent of the parties to the dispute. For readers following case No. A40-166729/2024, this double lens, lawyers' expert opinion plus the 67% figure calculated by the neural network, gives a fuller picture of what is happening than either source alone.