UX in the Age of AI: Rethinking Evaluation Metrics Through a Statistical Lens

Traditional UX metrics like SUS, NPS, and task completion rates were built for deterministic, rule-based interfaces where identical inputs yield identical outputs. But AI-powered products—chatbots, generative interfaces, recommendation engines—produce stochastic, context-sensitive, and temporally variable results. This renders legacy metrics structurally insufficient for capturing real user experience. Harish Vijayakumar’s paper, published on arXiv, tackles this gap by introducing ADUX-Stat (Adaptive Dynamic UX Statistical Framework), which reconceptualizes usability as a probabilistic signal distribution rather than a static score.

ADUX-Stat integrates three original constructs: Interaction Entropy Index (IEI) to measure the unpredictability of AI responses from the user’s perspective; Temporal Drift Coefficient (TDC) to gauge how perceived usability changes over repeated sessions; and Bayesian Usability Confidence Score (BUCS) to output credible interval estimates of usability quality under uncertainty. The framework is validated conceptually across five established AI product categories. It provides a reproducible, field-deployable methodology for researchers and practitioners, bridging HCI, statistical modeling, and AI product evaluation.