Measuring What Matters: Skills-Based Talent Analytics

Abstract

Skills-based talent analytics promises to replace job titles and degrees with direct evidence of what people can do, yet organisations struggle to decide which skills data deserve trust. This article proposes a compact taxonomy for skills inference and evaluates it against supervisor-rated performance. We combine self-reported skills, assessment results and signals drawn from project and learning records for 3,400 employees in five companies, and we compare three inference approaches of increasing complexity. A simple model built on a well-defined taxonomy matched the predictive value of more elaborate methods, while inconsistent skill labels reduced accuracy more than any modelling choice. We show how measurement quality varies by occupation and why ratings of proficiency need calibration across managers. The paper closes with guidance for analytics teams: start with a stable vocabulary, validate against outcomes that matter, report uncertainty openly, and involve employees in reviewing the profiles that are built about them.