In Ex Parte Szostak, Appeal 2025-001931, the Patent Trial and Appeal Board (PTAB) of the U.S. Patent and Trademark Office (USPTO) issued a decision reversing a rejection under Section 101 of claims directed to making predictions using machine learning. The PTAB decision underscores practical pathways to overcome 35 U.S.C. § 101 rejections.
The PTAB found the claims patent-eligible because recited data-transform and training steps improved the operation of the machine-learning model and therefore integrated any abstract idea into a practical application. The reversal of the Section 101 rejection reflects the expansion of patent eligibility championed by Under Secretary of Commerce for Intellectual Property and Director of the USPTO John Squires and U.S. Secretary of Commerce Howard Lutnick.
The digital gaming platform Skillz Platform Inc. owns the patent application subject to the appeal. The appealed claims were directed to determining and predicting user lifetime value (LTV) over different time intervals for a client application, with an exemplary context of digital gaming. The PTAB identified the invention as using short-term LTV data and cohort historical data to predict long-term LTV, in part through selecting an appropriate mathematical transform, pre-processing historical LTV data in a transform space, developing a machine-learning model with the pre-processed data, and then generating predictions.
The USPTO Examiner had rejected the claims under Section 101 as directed to patent-ineligible subject matter and under Section 103 as obvious over combinations of multiple prior art references. The PTAB reversed both rejections. This article focuses on the PTAB’s Section 101 analysis.
Applying the U.S. Supreme Court’s Alice two-step framework and the USPTO’s 2019 Revised Subject Matter Eligibility Guidance, the PTAB determined that the claims integrate an abstract idea into a practical application by improving the training and operation of a machine-learning model. Importantly, the PTAB found that the improvement was tethered to specific claim recitations. The appealed claims required selecting a transform based on characteristics of historical LTV data, pre-processing that data in a transform space, training the model on the transformed data, and applying an inverse transform to generated distributions. The specification explained that these elements reduce the effect of skewed LTV distributions and improve predictive accuracy without biasing the machine-learning model toward particular user groups.
The PTAB contrasted its precedential Ex parte Desjardins decision in which claims were found eligible because they improved continual learning and model efficiency with the Federal Circuit’s decision in Recentive Analytics v. Fox, which found claims ineligible where machine learning was used in a field-of-use manner without improving machine learning itself. The PTAB viewed the appealed claims as more analogous to Ex parte Desjardins than Recentive Analytics because the claimed transform-selection and transform-space training improved the model’s functioning per the specification and claim language.
The PTAB’s decision highlights a path to eligibility during examination at the USPTO for data analytics and machine-learning claims, namely emphasizing claim elements that concretely improve the training and operation of a machine-learning model and ensuring that the specification explains the technical problem and how the claimed invention solves it.
This advisory bulletin may be reproduced, in whole or in part, with the prior permission of Thompson Hine LLP and acknowledgment of its source and copyright. This publication is intended to inform clients about legal matters of current interest. It is not intended as legal advice. Readers should not act upon the information contained in it without professional counsel.
This document may be considered attorney advertising in some jurisdictions.
© 2026 THOMPSON HINE LLP. ALL RIGHTS RESERVED.
