SMART FILL
& CLEANUP.
Applying Machine Learning to the grid: Automating repetitive data entry and formatting tasks for millions of users.
ITERATIVE RESEARCH
USABILITY MECHANISMS
HUMAN-AI INTERACTION
Patterns in the Noise
Spreadsheet users spend a disproportionate amount of time on "janitorial work"—formatting, splitting columns, and cleaning data. The challenge was to leverage Google's Machine Learning capabilities to detect these patterns and offer one-click automations without being intrusive.
This required defining a new interaction model for "Suggested Actions" inside the grid—balancing the intelligence of the model with user control and trust.
-
01
TRIGGER CONFIDENCE
Defined the "Confidence Thresholds" for when the AI should intervene. We found that users preferred high-precision suggestions over high-recall frequency.
-
02
PREVIEW UI
Designed the "Ghost Preview" pattern, allowing users to see the result of the transformation (e.g., splitting First/Last name) before committing to it.
-
03
SMART CLEANUP
Expanded the model to identify data inconsistencies (duplicates, whitespace) and created a dedicated side panel for batch review.
ITERATIVE RESEARCH
Conducted 7 rounds of research to test user reactions to different levels of proactive suggestions and fine-tune the UI copy.
USABILITY STUDIES
Validated the "Ghost Preview" pattern and sidebar interactions to ensure users understood why changes were being suggested.
PATENT DEVELOPMENT
Research findings directly contributed to 2 patents on the interaction models for displaying and accepting probabilistic data suggestions.