Connecting
Every answered call is judged in the first second, person or recording, and the disposition goes straight back to your dialer. Runs inside your network. Never charges you per call.
The FTC caps abandoned calls at 3.0% of answered calls; Ofcom sets the same limit for silent calls. Every live answer dropped in error counts against it.
Asterisk's detector times the call. Silence length, greeting length and word count, each measured against a fixed threshold. Never against the audio.
So it cannot separate a slow hello from a recorded one.
Recordings are released before they reach a headset. Capacity rises without adding a seat.
Nothing plays to the called party while the call is judged.
The five second figure is the default waiting time in standard Asterisk detection. Both panels are drawings of how a verdict lands, not measurements: the number of wrong verdicts shown is illustrative. Accuracy and speed are design targets, not measured results.
An agent who answers a recording loses roughly ten seconds before they can dispose of it and move on. On a BPO floor that repeats all day, and it is time you have already bought.
A worked model, not a measurement. Substitute your own volumes; the arithmetic is seats x calls x pass-through x handling time, against a floor of seats x hours. Ten seconds is our own estimate of dispose-and-move-on time.
Hosted detection makes a vendor a processor of your customers' voices. Under GDPR, or a client data agreement, that is where most evaluations end.
No egress to review. No sub-processor to declare.
Run it beside what you already have. Compare dispositions on live traffic for a week. Reverting is one field.