AI Patentability Tool: Can It Identify Patentable Science for Investors?

Investors and technology-transfer organizations dedicate considerable time to identifying scientific research with commercial potential before it becomes patentable. Now, an artificial-intelligence tool aims to accelerate that process by estimating how closely a research paper resembles patent-linked work—potentially months or years before a patent application, investment deal or university spin-off reveals its commercial value.

Called the Translation Readiness Index (TRI), the machine-learning tool analyzes the language used in scientific titles and abstracts. It compares a paper’s vocabulary with that of publications previously associated with patents, generating a score that indicates how “patent-like” the research may be. Researchers at the data-analytics company League of Scholars in Sydney, Australia, developed the system. Their findings were published as a preprint on arXiv¹ and have not yet been peer reviewed.

“This is a new way to triage and rank research,” says Paul McCarthy, a computational social scientist, League of Scholars co-founder and co-author of the preprint. TRI estimates the likelihood that a paper contains language commonly found in patent-related research, he says.

Identifying commercially promising research

The researchers trained TRI using 20,610 scientific papers, including 9,431 that had been linked to patents. They fed the titles and abstracts into five machine-learning classifiers. The best-performing model ranked patent-linked papers above comparable papers without patent links 78% of the time.

Research papers later cited by patents were more likely to contain words such as “prototype,” “device” and “design” than papers that were not cited in patents. However, TRI evaluates only titles and abstracts. It does not assess the quality of a study’s underlying data, methods or results.

To determine whether TRI’s highest-ranked papers were associated with other signs of commercial activity, the researchers examined the 100 highest-scoring papers by authors at the University of Western Australia (UWA) in Perth. The papers, published between 2019 and 2026, were more likely than a random sample to include industry collaborators or researchers with previous patent experience. According to McCarthy, 83 of the 100 papers had industry co-authors, while 34 included at least one UWA-affiliated author who had previously patented research. The team is now testing TRI at several universities.

McCarthy cautions against using TRI as the sole basis for investment decisions because it provides a probabilistic ranking rather than a definitive prediction. Nevertheless, he says the tool could help researchers, funders and investors discover “unexpected gems” that might otherwise be overlooked.

Ben Miles, co-founder of Empirical Ventures, an early-stage deep-tech investment firm in London, says TRI could provide an additional signal for academics and funders assessing which ideas deserve further support. Universities, governments and philanthropic organizations could use the tool to identify research with potential before it is sufficiently developed to attract private investment, he says.

However, patent potential does not necessarily translate into commercial success. A technology can be patentable yet fail to meet market needs, attract customers or generate a viable business model. Any AI-based assessment of commercialization potential must therefore be combined with expert review, market analysis and technical validation.

AI tools for finding university spin-offs

TRI is one of several research-discovery tools designed to identify promising science and potential commercial opportunities. Some have already been adopted by research institutions to find discoveries that could eventually support new products, licenses or university spin-off companies.

One example is Haystack, a tool developed for Cornell University’s technology-transfer team in Ithaca, New York. Haystack can scan as many as 13,000 papers annually—far more than the team could manually review, says Matt Marx, Cornell’s vice president for entrepreneurship, innovation and external engagement, who developed the system.