This case-study presentation examines how speaking assessments can better serve multilingual test takers by separating intelligibility, comprehensibility and accentedness in scoring. Using a case study from Arabic L1 candidates in an English proficiency speaking exam, we show why accent is a weak proxy for communicative success and which pronunciation features more reliably affect understanding. We then translate the findings for assessment providers whose speaking scores shape admissions, employment and mobility decisions: how to reduce construct-irrelevant bias, improve score interpretation, strengthen fairness for diverse candidate populations, and design CEFR-aligned rubrics that reward being understood rather than sounding native-like. Framed around E-ATP’s theme of Building Assessment for Good, the session offers practical principles for speaking-test design, rater training and test-taker experience.