Designing AI Explanations: Startup Rewrites Model Feedback UI to Cut Support Tickets

AI · 5 min read

Moderato, a startup offering automated content moderation APIs, found customers overwhelmed by opaque model decisions and noisy false positives. Support tickets spiked because moderators could not see why the model flagged content or how to correct it for future runs.

Design and ML teams collaborated on a new UI that presents concise model rationale, confidence intervals, and an easy correction workflow that captures the corrected label and a short reason. Designers tested three explanation styles—textual rationale, highlighted evidence, and examples from similar cases—and settled on a hybrid card that prioritized clarity and actionability.

After rolling the redesign to power users, Moderato saw a 27% decrease in correction-related tickets and improved retraining data quality. The product team also established a design pattern library for model explanations, specifying when to show confidence scores versus simple trust signals, which helped scale the pattern across other AI-driven features.