How the assistant sounds, and the numbers behind each call
How the assistant sounds
Every call carries a few numbers about how it *sounded*, not only what was said. You do not have to do anything to get them; they are read from the call itself after it ends.
On the Calls page
Under a call's summary you may see a Sound line, for example: *p90 turn latency 1.2s · 6 turns · 1 interruption*.
- Turn latency is how long the assistant took to start answering after the caller stopped talking, measured across the call (the p90 is the slow end). Under about two seconds a call feels like a conversation; over three it feels broken.
- Turns is how many things the caller said. Fewer turns to reach a booking is better: every extra question is another chance to mishear.
- Interruptions counts the times the caller cut the assistant off. A lot of them usually means the assistant was talking too long or answering the wrong question.
- Possible false barge-in appears when the assistant stopped mid-sentence and resumed without the caller having said anything, which usually means background noise made it pause.
A call that ended within a few seconds with nothing captured shows Tagged: spam_short; those are not counted against your plan.
On the Analytics page
Median turns to book is the typical number of things a lead had to say on the calls that ended with an appointment. Watch it over time; if it climbs, listen to a few recent booked calls for repeated questions.
What the assistant now does differently
- Reads contact details back. A phone number it captures is read back digit by digit, an email spelled out, and it waits for a yes before moving on. Times and the service address were already confirmed this way.
- Notices frustration. If it has had to ask for the same thing twice, or the caller says something like "I already told you" or "you're not listening", it stops the interview and offers to get you on the line, or to have you call back, without being asked. The transcript is marked so you can find those calls.
- Sounds a little more natural. Light acknowledgements ("mm-hm"), a steadier voice, background-noise filtering, and it is primed with your business name and the cities in your service area so it hears them correctly. These are platform settings; you do not need to tune them.