Designing for Model Uncertainty: UI Patterns That Reduced User Confusion

AI · 6 min read

The product team needed to show model-driven credit eligibility odds without scaring users or creating false confidence. They prototyped three UI patterns: simple confidence badges (Low/Medium/High), numeric probabilities with short explainers, and interactive sliders that let users adjust sensitivity to see how recommendations change. Usability tests revealed that badges were fast to scan but opaque, sliders improved understanding for power users, and numeric probabilities caused anxiety for some users.

The final implementation used a layered approach: a concise badge for quick scanning plus a tappable detail panel that revealed the probability, contributing factors, and examples of actions that could improve the score. The panel included a toggle to simulate changes (e.g., lowering debt load) so users could explore the model’s sensitivity. Designers also added fallback copy to handle low-confidence cases when the model flagged insufficient data.

After rollout, support tickets related to eligibility confusion dropped 28%, and users who viewed the detail panel were 1.7x more likely to take recommended remedial actions. Product metrics showed better feature adoption without a hit to conversion. This case demonstrates that surfacing uncertainty clearly — but unobtrusively — lets users make informed decisions and preserves trust in AI-driven features.