Salesforce Decode
Salesforcedecode
Back to questions
AgentforceAdvancedfeedbacklearningquality

Design feedback loop from human agent corrections to agent improvement

Real World Scenario

Supervisors correct wrong agent answers in Case comments but corrections never reach agent training pipeline; same mistakes repeat.

Expected Answer

• Structured dispositions: agent wrong, knowledge gap, policy change • Workflow creating knowledge article or topic utterance from repeated corrections • Weekly review queue of low-CSAT transcripts assigned to content owners • Track correction themes metrics not anecdotal fixes • Close loop notification when fix deployed to agent version • Avoid auto-training on uncorrected noisy feedback • Governance council prioritizes fix backlog

Follow-Up Questions & Answers

Click to expand — each follow-up includes a direct, interview-ready answer

Direct answer: Structured dispositions: agent wrong, knowledge gap, policy change Also consider: Workflow creating knowledge article or topic utterance from repeated corrections In practice: Weekly review queue of low-CSAT transcripts assigned to content owners Optimize for scale and operational observability.

Architect Perspective

Agents improve only with closed-loop ops — corrections in Case comments are wasted unless piped to owners.