MLLMs automate configurator UI usability testing with 18 criteria

A new paper from researchers at Graz University of Technology and other institutions explores using multimodal large language models (MLLMs) for semi-automated usability analysis of configurator user interfaces. Configuration tools—used to tailor complex software systems, services, and products—often suffer from poor usability that existing heuristics don't fully address. The team synthesized 18 configurator-specific usability criteria from literature, then tested MLLMs on 16 real-world configurators, asking the models to rate each criterion and suggest fixes.

The results are promising: MLLMs reliably flagged usability issues and offered actionable, domain-aware feedback. While human validation is still required, the approach could cut the effort for usability analysis dramatically. The paper provides a structured framework for integrating MLLMs into UX workflows, making it easier to scale configurator testing without sacrificing depth. This could be a game-changer for product configurators in e-commerce, software setup wizards, and industrial design tools.