Every number below comes from our own AI calling system: 919 outbound calls, 48 inbound, $202.43 spent. Five questions, five views. Nothing here is a projection.
Each dot is one call. x is duration on a log scale, y is what the call cost, and depth is the hour of day. The voicemail band sits tight near 25 seconds because that is how long a message takes. Real conversations live far right, and there are almost none of them.
n=778 connected outbound calls. Log time axis. This is where the decision gets made.
216 clean hangups, bucketed by seconds into the call.
Over half the calls that drop are gone before 30 seconds. It is a cliff in the first half minute, not a gradual fade, and most drops land while the agent is still mid-sentence.
222 hangup calls, transcripts classified line by line.
32% of "hangups" were never human: phone trees, carrier messages, disconnected numbers. Another 14% produced no human audio at all. Exactly one call in 222 was a person actually declining.
The same behaviour split by which agent placed the call.
326 messages left, 234 distinct numbers, 10 callbacks.
The voicemails cost $56.47, about $0.17 each, which works out to $5.65 per callback. Nearly every callback arrived within the first hour, so the message only works while they are still near the phone.
Hours with 3+ dials. Reach means a real conversation past 45 seconds.
Morning wins. 7 AM local reaches 30% and 9 AM reaches 27%, while midday and evening sit between 10 and 19%. Most of our dials currently land in the afternoon, which is the weakest part of this chart.