Can AI Predict the Commercial Value of the Next Scientific Breakthrough?

A new AI tool developed at Duke University’s Fuqua School of Business may change how the world identifies promising research, before it ever earns a citation or patent.

By predicting value before citations or commercial activity occur, the tool could reshape how firms scout innovation and how universities manage their research portfolios.

Scientifiq.ai, a publicly available tool, “provides a window inside universities to find the most fundamental innovations that can be used to create products and services that will improve the lives of humankind,” said Fuqua professor Sharique Hasan. 

In their article published in the Strategic Management Journal, Measuring the Commercial Potential of Science, Hasan and his colleagues Roger Masclans, a Fuqua PhD, and Wes Cohen, the Snow Family Professor of Business Administration at Fuqua, explain how they used large language models and deep neural networks to predict the probability that a scientific paper will generate economic value for firms.

This research develops a new method that would allow companies and investors to identify promising research early, help scientists gauge the value of their work before publication, and enable universities to better plan commercial applications of their findings. 

This is especially important, Hasan said, because the model identifies potential commercialization before any actual citations or commercial activity occur — distinguishing it from traditional backward-looking metrics.

How to measure the commercial potential of scientific discoveries

“Citations reflect the past,” Hasan said. “But what if you're a company trying to assess brand-new research that hasn’t been cited yet? You want to know whether it has potential to drive future innovation.”

For that, the researchers built upon the assumption that a citation in a “renewed patent” application would be a credible proxy for commercial potential. Patent applications are expensive, and a citation in a renewed patent “represents the likelihood that a firm perceives an article as contributing to its economic gain,” the researchers write.

An AI model to predict commercial potential

The researchers used large language models (LLMs) and deep neural networks to develop a measure of commercial potential.

They trained their model on a database of 139 million academic papers published between 2000 and 2020, covering eleven fields, primarily natural and applied sciences and engineering, excluding social sciences.

First, they used the LLM to translate the text of the papers’ abstracts into numerical representations of different text features. Then, they instructed their neural network to analyze those numbers and identify patterns in the papers that were eventually cited in renewed patents.

“What combinations of these numbers predict a higher likelihood that a paper will be cited by a company?” Hasan asked.

The researchers found that their measure is very accurate in predicting commercial value. By validating their “commercial potential score” against a “holdout sample” of papers excluded from training their model, they found that, “an article in the top quartile of our measure is more than 20 times more likely to be cited by a renewed patent than an article in the bottom quartile.”

Practical uses of scientifiq.ai

Decades ago, large companies financed and conducted basic research, and also turned that research into products (the R&D developments originating from the Bell Labs are a frequently cited example, as shown in research from Fuqua’s Ashish Arora and Sharon Belenzon).

But in the last few decades, companies have delegated basic science to universities. “What we have now,” Hasan said, “is a system where companies are really good at commercializing and producing, while scientists at universities are really good at discovering.”

“Universities became the engines of these more fundamental innovations, funded through large grants from the government and private sector,” he said.

In this context, a tool such as scientifiq.ai could give universities an early signal of the commercial potential of their discoveries, measuring the chance an academic paper will, for example, lead to a license agreement with a company that will generate revenue for a university, he said.

This tool will also help companies find the experts they need to solve their technical bottlenecks.

“Let’s say a company has developed a molecule and they need a drug delivery mechanism — something that would allow the molecule to go inside the body — but they don’t have the internal capabilities to produce this delivery mechanism,” Hasan said.

Typically, companies would task an expert to search scientific literature, consult journals, and often contact universities to find researchers working on that specific topic. “It’s a hard process,” Hasan said.

With scientifiq.ai, however, “you can find the experts anywhere without making dozens of phone calls,” he said.

“Scientifiq.ai is a matching tool,” Hasan said. “On one side, you have someone with resources and a pressing problem. On the other, a scientist in a lab who may already be working on the solution. Our tool helps bring them together.” 

The researchers believe their tool will be useful for companies in innovation-intensive industries to stay ahead of academic advancements, and for universities' Technology Transfer Offices to monitor internal discoveries for licensing, patenting, and startup creation. Government agencies funding scientific research can also use this model to assess the potential economic impact of their investments. 

It can also help level the playing field by highlighting valuable research at less prominent institutions that might otherwise be overlooked.

In addition to the public tool, scientifiq.ai, the researchers have also made their code publicly available on GitHub, enabling other researchers to build on their model. 

“We made this work public because progress depends on others being able to build on it,” Hasan said

“The model offers researchers and institutions a practical tool and a chance to rethink how we evaluate and advance science with economic potential. By reducing bias in R&D selection and spotlighting hidden opportunities, this approach opens the door to smarter, more inclusive innovation strategies.”