This week's short tutorial is a quick tour of the inbox, the place where you talk to your customers.
Set your status to Available, and calls, texts, web chats and emails start coming to you in one list, longest waiting first. You get up to three conversations at a time, the rest wait in the queue, and messages pause while you're on a call.
Video: https://youtu.be/pw0g8fljzck
How does your team juggle channels today?
Update: Comcent is now omnichannel. Calls, SMS, web chat and email in one inbox, with an AI bot, your team, or both answering. Same pricing: from $20/month, no per-user fees. 1-minute walkthrough: https://youtu.be/7hLxtknOndg. Code PRODUCTHUNT still gives $25 free credit until Nov 30.
we build voice agents at Dial so "the promises made" line jumped out at me. the thing I'd want from that feature isn't just the list, it's confidence that a promise got flagged as a promise in the first place. a committed-sounding line like "I'll have that shipped by Friday" is easy to catch, but a lot of real promises come out softer, "I'll look into it" or "should be fine" said in a tone that the customer absolutely heard as a commitment even though the words are hedged. does the extraction lean on phrasing patterns or is it trying to catch the implied commitment too, since the gap between those two is where the actual broken promises hide
@galdayan Thanks Gal, great question. We don't use phrase patterns. An LLM reads the full transcript and pulls out commitments from context, so a hedged line like "I'll look into it" can still be flagged when the conversation treats it as a promise. Your point on soft promises is pointing us in the right direction. We're adding these cases to our LLM evals and will keep growing that set as testing goes on, so the softer commitments get captured reliably.
@pavanputhra good to hear it's going into evals rather than staying a one-off fix. one thing I'd watch for there is cultural/industry variance in how hedging sounds - "should be fine" from one rep might be filler, from another it's basically a promise, and that's before you get into non-native speakers whose hedge words don't map cleanly to confidence level. curious if you're building separate baselines per team or per agent, or treating the whole transcript corpus as one distribution
@galdayan agreed, that variance is real. Right now it's one corpus, no per agent baselines yet. Per agent feels right to me since hedging is personal, but it needs enough calls per agent before it means anything.
@galdayan This comment thread feels like I am having a conversation with a bot. Are you a bot designed to end the comment with questions so that thread goes infinity? 😂😉