An agent that answers and forgets is an answering machine with better manners. The half that pays for itself is what happens once the caller hangs up.
Where the call lands
/conversations — the same inbox as every other channel. A phone call is not a separate system to check, which matters because a separate system to check is a system nobody checks.
Each call carries its Call Recording, the transcript (with Copy transcript to clipboard), the Caller ID, a Call timeline, and an outcome:
Connected · Cut short · Voicemail left · Voicemail received · Missed · No answer · Busy · Failed
Cut short is worth knowing by name: “Connected, but the caller hung up before the agent finished.” A pile of those is a message problem, not a telephony one.
You can filter the inbox by agent, by assignee, by contact and by label.
What gets pulled out
A Dialog Insight is a named set of fields extracted from every conversation — up to twenty. Each has a name, a type, optional allowed values, and Extraction guidance: “Tell the AI exactly when to set this, and when to leave it out.”
Not a fixed schema. For a dental practice the fields might be treatment asked about, is this a new patient, preferred time. For a contractor, job type, postcode, urgency. Nobody else’s schema fits your business, and you are not given one.
Attach it to the agent, not the campaign
This is the setting worth getting right once.
An insight attached to the agent runs on every conversation that agent handles, inbound and outbound. A campaign using that agent then inherits it — the composer says “Inherited from the agent. This campaign has no override of its own.”
Attach at the campaign level instead and you get “Overrides the agent default”, which is the right tool for a campaign that genuinely asks different questions, and the wrong default for everything else. Agent-level means calls from every direction come back in the same shape.
Two failure modes, both silent:
- Not attached — “A Dialog Insight only runs on conversations from an agent or campaign it is attached to.” Saving is not attaching.
- No fields — “A Dialog Insight with no fields never produces a result.”
Triage happens on its own
Conversations are prioritised and flagged by Auto-Triage, marked AI-triaged in the inbox, so a queue sorts itself rather than being read top to bottom.
A Dialog Insight field can also become a label, which is how a value you defined turns into something you can filter and trigger on. If the label is one auto-triage also applies, the editor warns you that a workflow triggered on it “will fire from two sources” — worth heeding before you wire an alert to it.
When it actually arrives
Analysis normally completes within seconds of the call ending. The guarantee is a sweep that runs every half hour.
Both are true and the difference matters. The fast path is an event, and events can be missed — a restart at the wrong moment drops one. The sweep is what makes sure nothing is lost, and it is what you should set expectations by.
If an insight is not there the second you hang up, nothing is broken. If it is still missing half an hour later, that is worth looking into.
Callback, rather than transfer
The pattern we would suggest, and the one we run: define a field that marks a call as needing a human, and let an operator ring those back.
Transferring a live caller is supported and sounds better on a feature list. It is worse in practice, because it depends on somebody being free at that moment — and when nobody is, the caller has been passed to a ringing phone rather than served.
A flagged call is a caller who got a complete conversation, plus a callback list already sorted by priority.
A call ending is a trigger
The end of a call is an event a workflow can start from, and it is the same event as a chat conversation closing. So the follow-on is not a short list of receptionist features somebody chose for you — it is whatever a workflow can do.
Take a message and email it onward. Create or update the contact. Book something. Add the caller to a list. Post to a channel your team already watches.
You can scope a workflow to one channel or one agent, so your after-hours line and your main line can behave differently.