Working a ticket
An alert about a breach is a louder version of the thing already being ignored.
The ticket was missed because nobody was watching the queue. Sending a second message to the same people, about the same queue, is not an escalation — it is the same failure with more noise.
An escalation has to change who is accountable.
Setting the clock
Four fallbacks, so a ticket almost never has no service level.
Explicitly set on the ticket, then a per-customer override, then the queue's default, then any active tier at all. Deadlines are snapshotted when the ticket arrives rather than recomputed on every read — and recomputed only when the severity, the queue or the customer changes, from the original arrival time, as though it had always been that severity.
- A first response counts when a person responds —
On a real move, or on an internal comment on a new ticket. A customer's own reply deliberately does not count — which is the distinction most systems miss, and the one that makes the metric worth measuring.
- A late response is recorded as exactly that —
The breach, plus an entry saying the answer arrived after it — with a flag so the overnight watcher does not then alarm somebody who already replied, late.
- The warning fires once, and never after a breach —
Set-once markers are the mechanism, which is what makes the sweep safe to re-run.
- One bad ticket cannot silence the night —
Each is handled in its own savepoint, so a single malformed row does not abort the workspace's whole sweep.
Escalating
Two steps, and the first breach adds somebody rather than replacing them.
A breach opens a routed approval: the assignee, then the queue's owner, then anybody holding the support lead role. If the first step itself breaches, the assignee's own manager is added as a peer rather than substituted — so nobody is quietly taken off a problem they still own.
It is idempotent, so a replayed sweep never opens a second one. And the handler that closes it only clears the pointer if it still matches that instance, because a second escalation may have opened meanwhile.
If a workspace has nobody who could take it, it records that and moves on rather than failing.
›Can we configure the working window?
Not yet. The business-hours window is fixed and there is no holiday calendar. If your service levels are contractual, know that before you commit to them in writing.
›Can the warning threshold be changed?
No — three-quarters is fixed, as is the auto-close window for resolved tickets.
›Does a customer see the clock?
No. There is no customer-facing portal, so service levels are an internal measure you report on rather than something they watch.
›Are ticketing and service levels bought together?
They are separate — a workspace can run the queue without the clock.