Closing the DEI Policy-Practice Gap: A Real-Time, Fairness-Aware AI Framework Using Organizational Justice Theory

Abstract

Although 78 percent of institutions have formal policies on DEI, in only 42 percent of companies do employees claim to feel included. We combined the Organizational Justice Theory with a real-time and fairness conscious artificial intelligence pipeline to examine a total of more than 10,000 employee responses (n = 984 surveys, n = 45 interviews), in the technology, healthcare, and financial fields. We found incongruent policy implementation (40%), absence of leadership commitment (30%), and cultural resistance (20%) to be fundamental obstacles to implementing policies using fine-tuned DistilBERT to sentiment analyze, LDA model, and constrained logistic regression. The strongest predictor of the perceived inclusion was leadership commitment (b = 0.45, p < 0.001; OR = 2.51). The bias-reduced pipeline (equal opportunity difference = 0.07) consumed the data in an enterprise scale (1.8 hours) and allowed a real-time DEI Pulse Dashboard which extrapolated up to 30% inclusion benefits with target interventions. The research contributes to the literature on organizational justice by defining the scale of fairness perceptions and providing a bias-reduced, replicable, and open-source-
friendly framework to implement a real-time, bias-based, and DEI monitoring system.