Football Recruitment Intelligence

The problem

A similar player is not automatically a suitable replacement. I wanted to make football recruitment questions inspectable: which observed characteristics matter, what does the available data actually cover, and how sensitive is a comparison to the chosen requirements?

The approach

I combined a reproducible Python research pipeline with a Next.js and TypeScript interface. StatsBomb and Pappalardo-Wyscout observations retain their source definitions and coverage labels. The product connects searchable player-season profiles, role-based Player DNA, explicit recruitment requirements, club context and shareable comparison scenarios.

The challenge

Football datasets differ in definitions, seasons and coverage. Apparent similarity can reflect the data source or a small comparison pool. I therefore kept metadata separate from performance observations, compared models with simple baselines and made unsupported comparisons and uncertainty visible in the product.

Outcome & value

A live English/Dutch research product with public source code, documented methods and reproducible release checks. The evaluation did not support broad cross-league translation or establish improved transfer decisions. Rankings describe observed similarity under selected criteria; they are not transfer recommendations or predictions of success.

Lessons learned

A useful data product must explain where a comparison stops being reliable. Source provenance, simple baselines and negative findings belong in the interface as well as the research documentation.