Predicting Salesforce Product Ideas 90% Faster with AI

Challenge

Salesforce's IdeaExchange platform collects thousands of feature requests from its community, but evaluating which ideas have real potential often stalled in a years-long backlog under the existing process. The team needed a faster, more reliable way to surface high-potential ideas before they stalled.

Action

I designed a predictive Rising Score model using logistic regression, trained on historical user forum data, to identify ideas likely to gain traction the next year. I also applied rubric-based local LLM extraction on unstructured text to pull out strategic product insights without moving sensitive data outside the organization.

Result

The model cut idea evaluation wait time by 90% at 93% accuracy, and identified 3 strategic product focus areas for Salesforce stakeholders across product and roadmap teams.