I built a program from scratch to turn overseas labs’ advanced research into products, and the commercialization rate rose year after year to about half
This page is a translation of the Korean original.
Background
Think of it as choosing seeds to plant for a harvest years from now, not this year’s crop. Advanced technology means technology that goes into products two to five years out, not what sells today.
To pick good seeds you first gather many, then filter at each step — which is why this is called a pipeline. Candidates enter wide and pass through gates (Stage-Gate) — discovery → review → kickoff → midpoint check → hand-off to the product team — and only a few reach the end.
There are two ways to find candidates: an open track that looks outside, at universities, start-ups and research institutes (open innovation), and a strategic track that works backward from the product roadmap to name the technologies the company must have. The open track brings unexpected opportunities; the strategic track brings certain usefulness.
Why it matters to a company
- Keep money from scattering — Advanced projects fail often by nature. Gates let you stop hopeless projects early and concentrate the remaining budget on the ones that can work.
- Don’t learn about technology gaps too late — If you find out late that a competitor already has a technology, catching up can take years. Regular discovery works as an early warning.
- Get research all the way into products — The stretch where research results never reach a product team is called the “valley of death.” You can only cross it if the hand-off stage is designed in from the start.
What I did
Headquarters was focused on products it could sell now, which made it hard to invest in technology years ahead. Each overseas lab did advanced research, but there was no path from that work into headquarters’ products, and each site used different criteria to choose projects.
I planned “Global No. 1” from scratch — an advanced-technology commercialization program covering every overseas lab — and ran it, reporting to the CTO at kickoff, review and completion.
We surveyed each lab’s advanced research, selected projects that could be commercialized within two to three years, incubated them, and connected them to business units. Projects were split into an open track (found outside) and a strategic track (worked back from the product roadmap), and each was assigned to a lab based on fit with its existing projects, the level of its people, and the external environment (regulation, standards, security constraints).
Of the 10 projects found in the first year (2 open, 8 strategic), half — five — reached products over time as the technologies matured. Early on only a few became products, but as the number of projects and funding grew and operations matured, the commercialization rate rose to about half. One public example: federated learning technology from the UK lab was applied to Bixby.
Key decisions
- Scoped the program to technology that could be commercialized within two to three years — Technology too far out can’t be absorbed by product teams, and near-term technology is handled by the business units themselves. That middle band — left open while headquarters focused on products — was where overseas labs could add value.
- Chose the lab for each project on three criteria — Fit with existing projects brings speed, the level of the people decides the outcome, and regulation, standards and security constraints differ by country (for example, the US favored work tied to US regulation and standards, while some countries could not host security-technology projects).
- Argued for investing in federated learning despite the short-term cost — For a device company, on-device AI matters most, and federated learning is at its core: key values learned on devices retrain the main model, which is then redistributed, so quality keeps improving.