Write-ups of real incidents and projects on live call-center systems. Every number comes from real call data. Company, carrier and vendor names are left out on purpose.
Callers who spoke rose from 12% to 30% after AMD came out; a safety guard left the WhatsApp number in agent replies on just 1 of 1,001 calls in a day.
A self-hosted Italian voice agent behind a ViciDial press-1 campaign: what failed between demo and real traffic, and the checklist that came out of it.
62.3% of one carrier's answers arrived with no SDP; dialer answer rate halved from 31.2% to 15.6% from one day to the next while carrier-side ASR held steady.
How Homer SQL, Asterisk logs and a CDR join proved a carrier was sending 200 OK without SDP, and why a sloppy LIKE filter makes a clean carrier look 50% guilty.
26 answered inbound calls in 44 minutes, about one in five on three trunks, had no carrier RTP accepted at the PBX. Over 12 business hours, 0 of 1,330 normal calls tripped the detector.
Turning Asterisk Strict RTP learning lines into a dead-air alert, corroborated with Homer RTCP, and why pre-answer RTCP misleads when early media is in play.