I Stopped Counting Draft Replies as Follow-Up

The draft looked like progress

I once reviewed a queue with several thoughtful replies saved beside customer inquiries. The messages were clear and the suggested next steps were sensible. It would have been easy to count them as handled.

Then I checked the event history. Some drafts had never been approved. Some had no send event. One had been written for a record that was later identified as a duplicate. A polished draft was evidence of preparation, not evidence that a customer had been contacted.

I separated four states

I labeled each item as draft, approved, sent, or confirmed response. Draft meant text existed. Approved meant a person accepted the message for that case. Sent meant the system recorded the outbound event. Confirmed response meant the source showed a reply or another defined customer event. Those states answered different questions.

I also recorded the owner, source event, service-area fit, next action, and reason for any hold. A redacted table was enough to show the difference without copying full customer messages. The review became less flattering and more useful.

The queue showed where the work stopped

Some items needed approval. Some were blocked by missing information. Some had been sent but had no response. A few were duplicates or outside the service area. When those cases were separated, the team could decide whether the next repair belonged in the draft workflow, the approval queue, the sending system, or the customer-facing process.

I would now ask for the source event, draft timestamp, approval record, send event, and defined response evidence before calling an inquiry handled. If one piece is missing, I would label it unknown and assign a human next question rather than upgrade the status by assumption.

AI Cleanup Doctor can help structure a redacted draft reply audit. The owner remains responsible for privacy, communication, retention, and the final disposition. The Order page describes the fixed-scope starting point.