Feedback automation is most useful when it connects the original message to a clear next action. A label alone does not tell a team where to respond, who owns the case, or whether the notification was delivered.
My published BMRC Hospital project account describes an n8n, OpenAI, Supabase, and Gmail workflow for multilingual feedback routing. The design suggestions here are a general planning pattern, not a claim that every detail is part of that implementation.
Give each item a stable identity
Assign or preserve an identifier as early as possible. If the source sends the same event again, the workflow can check whether it has already been processed rather than creating duplicate records or notifications. Store the identifier with processing status and relevant timestamps.
Store the decision and make notification verifiable
Keep the original message or a protected reference, the generated category, the applied routing rule, and the final destination together. Where practical, record which model or workflow version produced the suggestion. Avoid storing unnecessary copies of sensitive content; access and retention should follow the organization's policy.
Creating a ticket and notifying its owner are different actions. Capture whether the notification service accepted the request, and define what happens if it fails. The case should remain visible in the system of record even if email is temporarily unavailable. A retry should not create another ticket or repeated alerts without an explicit reason.
Plan exception paths
Test missing message text, unsupported language, ambiguous category, duplicate event, database error, and email failure. For each case, specify whether to retry, pause for a person, or mark the item for follow-up. Alerts should identify the workflow and record without exposing more customer information than necessary.
Stable identifiers, separate processing states, and explicit recovery rules help a team trace the path from source message to final owner. See the [BMRC feedback-routing case study](/case-studies/bmrc-hospital-ai-feedback-agent/) for the documented project scope.