Making spreadsheets smarter: empowering users to detect errors with explainable AI.
AI-Powered Anomaly Detection in Spreadsheets- Workiva
ROLE
UX Researcher • Interaction Designer
EXPERTISE
User Research · Thematic Analysis · Affinity Mapping · Data Visualization · Human-AI Interaction · Interaction Design · System Explainability · Privacy UX
YEAR
2025
Description
Designing Intuitive AI for Error Detection in Spreadsheets
Workiva challenged us to design a tool that makes AI-powered anomaly detection accessible to non-technical users. Our goal was to improve how everyday spreadsheet users identify errors without overwhelming them.
Project Details
Conducted semi-structured interviews with analysts, small business owners, and accountants
Mapped user pain points into themes like error visibility, trust in AI, and workflow integration
Designed a color-coded, sidebar-integrated anomaly tool with natural language explanations and feedback loops
Process
Research-Led, Trust-Built
Our UX research focused on building explainable and usable AI interactions. We applied thematic analysis to user interviews and shaped features based on cognitive load, workflow fit, and visual clarity.
Interviewed 5 users across varying tech backgrounds
Used affinity mapping to categorize findings into actionable insights
Developed 2 detailed personas and mapped their current vs. ideal user journeys
Prioritized explainability, feedback control, and role-based privacy in the design
Solution
Simple, Smart, and Transparent Error Detection
Our interface integrates seamlessly into existing spreadsheet workflows while guiding users with color highlights, smart tooltips, and a side panel of AI-generated insights.
One-Click Anomaly Detection via top-bar tool
Color-Coded Highlights: Red for critical, yellow for warnings
AI Sidebar Panel: Explanation of flagged cells, grouped alerts, and guided next steps
Natural Language Feedback: Short insights like “+42% vs. average”
Role-Based Visibility: Share alerts with collaborators or keep private
Confirm/Reject Feedback Loop: Train the AI to get smarter with every use
Results
Built for Trust, Designed for Real Workflows
User research led to a system that users trust and understand no code required. Our solution made error detection faster, smarter, and more collaborative.
Reduced user-reported error identification time by 40% (based on concept testing feedback)
All users preferred the “explainable AI” feature over black-box alerts
Clear UI and role-based sharing improved cross-team collaboration and trust
Judges praised the project for its accessibility, clarity, and research-driven design