The dashboard shows how many calls connected, but nobody knows how many callers failed before reaching the AI. Monitoring only active sessions leaves upstream setup failures outside the operating picture.
For teams running telephone reception, count eligible call attempts at the relevant entry boundary and follow them through setup. The denominator determines which failures the report can reveal.
What is a call setup failure?
A setup failure prevents the intended telephone conversation from becoming available. It can arise before application dispatch, during destination selection or while the route negotiates the connection.
Keep it distinct from a connected call with no audio and from a completed conversation with an unsuccessful business action. Those cases need different investigation and recovery paths.
Define a setup funnel with explicit states
Use the events available from the actual providers. The following is a proposed monitoring model, not a claim that every platform supplies identical event names.
| Funnel stage | Question the metric answers |
|---|---|
| Eligible incoming attempt | Did the call reach the monitored entry? |
| Route selected | Was the expected business destination chosen? |
| Application session assigned | Did an appropriate worker receive it? |
| Telephone connection established | Did signalling reach the accepted state? |
| Usable media confirmed | Could the participants exchange audio? |
Some boundaries may be unobservable in your account. State those gaps rather than treating missing telemetry as zero failures.
Group failures by actionable boundary
Retain provider status, direction, route and deployment version where appropriate. Avoid placing full caller details in broadly accessible monitoring labels.
An alert should identify an owner and a next investigation. A carrier-route failure belongs with the team that can inspect that provider, while a dispatch failure may require application deployment access.
The LiveKit SIP troubleshooting reference illustrates how setup and media evidence narrow the responsible boundary. The cross-system trace guide connects those observations to application records.
Separate customer impact from test traffic
Tag authorised tests and monitor their coverage separately. A synthetic call can detect a route failure, but it does not describe the number of affected customer attempts by itself.
My voice platform contribution supplies relevant context. A new monitoring design needs the business's actual entry points, provider visibility and reception fallback responsibilities.
Review the funnel after a configuration change or incident. If the first monitored event occurs only after successful connection, upstream call failures remain outside the denominator and need a separate evidence source.
Can application session counts measure all incoming attempts?
Only if every relevant attempt reaches that boundary. Upstream failures may never create an application session.
Is connected-call silence a setup failure?
Keep media usability separate so signalling success does not hide the audio problem.
What should we define before building alerts?
Agree on monitored entry points, failure categories and response owners in the voice AI operating brief.