Most of the solutions we need already exist.

Turning public research into companies and products is one of the highest-leverage things an economy can do, and it mostly runs through technology transfer offices (TTOs): the teams that read what researchers publish, judge what has commercial potential, and help patent, license, or spin it out. Closing the gap between what science discovers and what reaches the market is a stated priority of innovation policy across the OECD.

The bottleneck sits at the very first step. TTOs screen disclosures and papers by hand, so the cost of assessment scales with the size of a research portfolio while staff capacity does not. In practice only a small, self-selected fraction of an institution's output is ever evaluated for commercial potential. The rest is lost, not because it was judged unpromising, but because it was never judged at all. This is the mechanism underneath the long-running "European paradox," where world-class science underperforms at becoming companies and value. Well-funded universities can afford professional TTOs; smaller institutions and emerging economies often cannot, so promising work goes unrecognized regardless of merit. The binding constraint is not speed, it is coverage: a system cannot translate what it never assesses.

This is where AI changes the economics. A modern language model can read a technical paper, apply a structured commercialization rubric, and return a calibrated, auditable judgment for a few cents instead of an expert-hour. That turns a first-pass assessment from a rationed resource into an abundant one, so an institution can screen its entire output rather than a sliver of it. The goal is not to replace expert judgment but to direct it: surface the promising work, then let people do the curation, verification, and dealmaking that humans are uniquely good at. Done well, it lowers the activation energy for researchers to become founders and gives funders and policymakers an honest signal of translational potential across every field and region.

I wrote a paper setting out this argument and a method for it, Augmenting the Technology Transfer Office: Large Language Models for Scalable Commercialization Assessment of Research Portfolios. It operationalizes the established Cloverleaf model of technology transfer as an auditable language-model workflow, scoring each paper across thirty criteria spanning market, technology, commercial, and management readiness, then triangulates that with two independent scoring instruments as a built-in validity check, plus a portfolio-analytics layer that rolls per-paper scores up to whole organizational units.

To put the idea into practice, I designed and built Trove end to end. Paste a DOI or an abstract and it returns a full commercialization assessment in seconds, and it has been run on portfolios of hundreds of papers at a fraction of the cost of manual review. The concept, the assessment engine, the scoring instruments, and the interface are all my own.