Perceived fairness, accountability and transparency toward ML-based AI service delivery: conceptualization and measurement

Purpose This study conceptualizes and validates a multidimensional framework of consumers' perceived fairness, accountability and transparency (FAT) in machine learning (ML)-based AI service delivery. Addressing inconsistent conceptualizations and limited service-oriented measures, it examines how FAT perceptions emerge across goal-setting, coding and developing and implementation stages. A formative FAT scale is developed and tested for its predictive validity on trust and usage intention. Design/methodology/approach A multi-stage scale development process was used. Items were generated through literature review and 20 interviews, and face validity was assessed via an expert focus group. Two surveys (n = 448; n = 535) supported exploratory factor analysis (EFA), confirmatory factor analysis (CFA) and partial least squares structural equation modeling (PLS-SEM) validation of a reflective–formative FAT structure. A third survey (n = 455) established convergent and nomological validity with established FAT scales, trust and intention to use. Findings Results confirm FAT as a multidimensional construct shaped across goal-setting, coding and developing and implementation phases. All nine first-order factors significantly form the three higher-order FAT dimensions. The scale demonstrates strong reliability and validity, and fairness and transparency exhibit the strongest effects on trust and intention to use. Findings support a stage-based view of ethical perception in AI service delivery. Originality/value This research provides the first stage-based conceptualization and validated scale of perceived FAT in ML-based service AI. It integrates attribution theory with a socio-technical lens to disentangle how ethical judgments form throughout the AI lifecycle. The study advances theoretical clarity and offers actionable guidance for designing, developing and deploying ethically aligned AI services.