The AI receptionist captures a callback request, sends an email, and reports success. Three days later, nobody has called the customer. Intake worked; follow-up did not.
Measure callback completion as a staff workflow with an owner and a final outcome. A notification is only one step along that path.
Define what a completed callback means
Decide whether you mean an attempted call, a connected conversation, a resolved enquiry, or another agreed outcome. Report them separately if several matter.
For a fictional service business, a useful task model distinguishes pending, assigned, attempted, connected, resolved, and closed-with-reason. These are proposed business states, not a requirement to use those exact labels in your software.
Preserve the details staff need to act
| Callback detail | Why it matters |
|---|---|
| Usable number | Staff can reach the intended person |
| Reason for contact | The caller does not need to start again |
| Preferred window | Staff can plan a suitable attempt |
| Assigned owner | The request does not belong to everyone and nobody |
| Outcome and timestamp | Reporting reflects actual work |
If the caller corrects their number, update the task and preserve enough traceability to explain the change. Keep caller-reported preference separate from a promised appointment with a staff member.
Track attempts without hiding unresolved work
A busy signal or voicemail may count as an attempt, but it should not automatically count as a connected callback. Decide the retry and closure policy with operations.
Use a reason when closing a request without contact: invalid number, customer withdrew, duplicate request, or another approved category. Avoid a single “done” button that makes different outcomes indistinguishable.
For a worked example, suppose 20 fictional requests are eligible for callback, 15 receive an attempt, and 9 reach the customer. Attempt coverage is 15/20, or 75%. Connected-callback coverage is 9/20, or 45%. These illustrate definitions, not results from a client pilot.
Give the queue a daily operating view
Show unassigned requests, due work, previous attempts, and missing details. Choose review frequency and response promises based on staffing. Do not let the AI promise a deadline the team has never agreed to meet.
If an email fails but the task is safely stored, the queue should still expose the work. If the task save fails, an email alone may be insufficient evidence that follow-up is owned.
The service-request ownership article explains why the office workflow belongs in the integration scope. The booking-cost guide shows how outcome definitions affect cost reporting.
Callback reporting questions
Is sending an alert a completed callback?
No. It establishes notification delivery at most.
Should every request remain open forever?
Use an agreed retry and closure policy, preserving the final reason.
What should a pilot report show?
Show eligible requests, assignment, attempts, connections, and unresolved work. Use those definitions when reviewing intake performance.