In cases where a single message contains: A clear feature request Supporting context (training notes, stakeholder alignment) Multiple related pain points Autopilot frequently routes these to “Low confidence” instead of: Extracting the primary idea, and Splitting secondary themes into separate ideas or insights Example pattern: Core request: Display program descriptions in the daily print calendar Supporting pain point: Duplicate manual work (re-entering descriptions via text boxes) Even though both signals are clear, Autopilot struggles when they are embedded in narrative context. Expectation: For Ideas to be the primary workflow, Autopilot should be able to: Identify the dominant request in a message Separate secondary but distinct problems Ignore internal context (e.g., “no action needed”, stakeholder decisions) Right now, this creates friction because: We need to manually restructure messages before sending, or We risk losing signal in Low confidence
