Technology Information Law Blog by Charles Gideon Korrell

In Recentive Analytics, Inc. v. Fox Corp., No. 23-2437 (Fed. Cir. Apr. 18, 2025), the Federal Circuit affirmed the dismissal of a patent infringement suit on § 101 grounds, holding that Recentive’s asserted patents were directed to ineligible subject matter. The court concluded that the patents merely applied well-known machine learning techniques to the new context of scheduling live events and generating television network maps, without disclosing any specific improvements to the machine learning models themselves.

Background: Scheduling and Broadcasting with ML

Recentive sued Fox Corp., alleging infringement of four patents that described the use of machine learning to optimize the scheduling of live events and dynamically generate television broadcast “network maps.” These network maps determined which programs would air in which local markets at specific times, a particularly complex task in the context of live sports broadcasts like NFL games.

Historically, networks made these scheduling decisions manually, based on general market heuristics or static planning tools. Recentive claimed to have pioneered a method that applied machine learning to this domain, using models trained on historical data—such as ticket sales, regional viewer preferences, and event logistics—to create optimized schedules and automatically update them in response to real-time data. According to Recentive, Fox deployed similar technology in its broadcast operations without a license, including tools that tailored game broadcasts for maximum ratings across different affiliates and time slots.

The four asserted patents fell into two categories:

• Machine Learning Training Patents ('367 and ‘960 patents): Directed to generating optimized event schedules using machine learning models trained on historical event data.
• Network Map Patents ('811 and '957 patents): Concerned with using machine learning to optimize network broadcast schedules and dynamically update network maps in response to changing conditions.

While the patents invoked modern ML techniques—such as neural networks, support vector machines, and gradient-boosted forests—they required only “any suitable machine learning technique” and did not claim new models, architectures, or training methods.

The Alice Framework

Applying the Alice two-step test, the court evaluated whether the patents were directed to patent-eligible subject matter under § 101.

Step One: Directed to an Abstract Idea

At step one, the court found that the patents were directed to the abstract idea of applying generic machine learning methods to new domains (event scheduling and network mapping). Key to the court’s analysis:

• Generic ML Use: The court emphasized that Recentive’s claims merely described using conventional ML techniques to solve known scheduling and broadcast optimization problems.
• Field of Use Limitation: Applying ML to a previously “unsophisticated” domain like television scheduling was not enough. As the court reiterated, restricting an abstract idea to a specific environment does not render it eligible.
• No Technological Improvement: Although the patents touted “real-time” and “dynamic” scheduling, the court found these features to be inherent to ML itself and lacking any claimed improvement to the underlying technology.

The court distinguished the claims from those in Enfish, McRO, and Koninklijke, where specific improvements to computer functionality or data processing methods were claimed. Instead, it likened the case to SAP Am. v. InvestPic and Electric Power Group, where courts rejected abstract data-processing claims lacking technical specificity.

Step Two: No Inventive Concept

At step two, the court found that the claims lacked any “inventive concept” sufficient to transform the abstract idea into a patent-eligible application:

• Generic computing environment: The patents were implemented on general-purpose computing hardware.
• No technical solution: Even features like real-time updating and iterative training were inherent aspects of machine learning, not technological innovations.
• Efficiency gains irrelevant: The court reiterated that merely achieving results faster or more efficiently using a computer does not make a claim patent-eligible (Content Extraction, Customedia, Trinity).

A Clear Line on ML and § 101

In holding that “patents that do no more than claim the application of generic machine learning to new data environments” are ineligible, the panel clarified how existing § 101 jurisprudence applies to ML-related patents:

• To survive Alice, claims must go beyond stating the use of ML in a novel context. They must describe how the ML models themselves are improved or implemented in a technically meaningful way.
• This decision aligns with recent decisions like SAP Am. v. InvestPic, Electric Power Group, and Stanford, which stress that invoking known tools on new datasets is not enough.

Litigation Strategy: The Algorithm That Wasn’t Claimed

A particularly revealing moment came during oral argument when Recentive’s counsel admitted that they deliberately avoided claiming a new ML algorithm. The rationale? Fear of falling into another § 101 trap: the prohibition on claiming natural laws or mathematical formulas.

This litigation strategy underscores the dilemma facing patent drafters in AI and ML. Claims that are too abstract fail at Alice step one; claims that are too technical risk being characterized as mathematical laws or unpatentable subject matter at step two. In seeking to thread that needle, Recentive ended up with patents that described no technical implementation at all—just an invocation of ML on a new dataset.

This signals a growing tension in ML patent prosecution: balancing § 101 eligibility with disclosure sufficiency, without accidentally triggering disqualification under the guise of abstraction or mathematical formalism.

Takeaway for Practitioners

The Federal Circuit’s decision in Recentive sends a clear and cautionary message to those drafting or litigating AI-related patents:

• ML Must Be More Than a Buzzword: Simply applying a standard ML model—without claiming how the model is improved or adapted—will almost certainly fail under § 101.
• Implementation Details Matter: Vague references to “dynamically optimized schedules” or “real-time adjustment” will not suffice without concrete technical means to achieve those goals.
• Disclosure Strategy Is Crucial: Avoiding algorithmic detail to sidestep one § 101 problem may backfire by triggering another. Patent claims must walk a fine line: detailed enough to show inventive application, but not so mathematical as to seem like an unpatentable law of nature.
• Field-of-Use Limits Won’t Rescue Abstract Claims: No matter how novel the application, courts will disregard “do it with AI” claims that offer no technological advance in the computing or modeling process itself.
• Expect Closer Scrutiny in Litigation: Plaintiffs asserting AI-driven patents will need to show not just commercial similarity but also a clearly delineated inventive step in how ML is implemented. As Recentive illustrates, merely identifying an infringing product that uses AI will not save a claim that lacks specificity in what the invention actually improves.

Machine learning may revolutionize many industries, but for now, it does not exempt patentees from the requirements of § 101. To claim such innovations successfully, applicants must do more than dress up abstract ideas in predictive analytics—they must show how their inventions improve the underlying technology itself.

By Charles Gideon Korrell