Connect and a call centre
A call centre gives you people: judgement, empathy, and the ability to do whatever the situation needs. Connect gives you a line that always answers, says the same true things every time, and writes down what happened. Three limits are absolute on the live carrier today — no call recording, no completed transfer to a human, and a reply floor measured at 3.3 seconds — and none of them is negotiable by configuration.
What a call centre does that Connect does not#
- Complete a transfer to a person
- On the live carrier this is foundation, not available. An escalation phrase queues a transfer and the supervisor panel can act on a live call, but a completed warm transfer depends on a provider capability that is not enabled.
- Record the call
- Also foundation. The capability is asked of the provider rather than assumed, and it is not enabled on the live carrier. If you need recordings for quality or compliance, that requirement is not met here.
- Handle what nobody wrote down
- A person improvises correctly from context. Connect refuses to invent a price, an SLA or a warranty term that Knowledge does not support — which is the right behaviour, and it is still a refusal rather than an answer.
- De-escalate a genuinely angry customer
- Skill, tone and the authority to make an exception. Connect has none of the third.
- Workforce management
- Forecasting, rostering, adherence, quality scoring and coaching — an entire discipline that has no equivalent here.
What a line that never sleeps is worth#
- Every call answered. No queue, no hold music, no abandoned call at 21:00 — subject to worker capacity, which is visible rather than silent.
- The same answer every time. Grounded in the workspace's Knowledge, with the same refusal rule on every call.
- A record without anyone typing one. Transcript, summary, outcome and hang-up attribution are produced for every call.
- Continuity across channels. The caller is resolved to a Person, so the call knows about the email thread and the WhatsApp message.
- Commitments that execute. A call-back promised on a call becomes a follow-up that the phone drain rings — two per tick — under the same gates as any outbound call.
- Languages without staffing them. Language handling is per call rather than per agent rota.
Latency, honestly#
This is where a comparison usually gets fudged. The realtime path's floor is the model's first token plus its end-of-turn detection: 3.3 seconds median on the best measured call, not the one to two seconds that gets quoted for conversational AI in general. Roughly 60% of the response time is model-bound, which means no amount of tuning on this side removes it.
Against a human that is slow. Against a hold queue at eight in the evening it is not. Which comparison applies to you is the actual question, and it depends on what your callers do now when nobody picks up. See voice latency for the full measurement and what has already been taken out of it.
Capacity, and what happens at the edge#
| Situation | What happens | Where you see it |
|---|---|---|
| Workers at capacity | The line reports it rather than answering into silence | Line health, escalated into Needs You |
| A call runs long | There is a hard ceiling of 30 minutes per call | The call record's outcome |
| The model session fails | The call is attributed rather than logged as a success | Hang-up attribution on the call record |
| Budget exhausted | A call the budget cannot cover is stopped rather than started | Voice budget and usage |
| A caller needs a person | An escalation is queued and a supervisor can act; the completed transfer is not available | Supervisor panel, Needs You |
Every one of those drains from Needs You by itself as its cause clears, which is the difference between an operational queue and a to-do list.
The realistic arrangement#
Connect takes first contact on the line and finishes what is finishable — hours, status, availability, a booking, a re-order, a message taken properly with a reason attached. What it cannot finish becomes a follow-up or a case with the whole conversation attached, and a person picks it up knowing what was already said. That is a smaller team answering better calls, not an empty room.
If your requirement is recorded calls, warm transfer or a contractually measured answer time, keep the call centre and read what is not available before planning around it.
Questions#
Can a caller ask to speak to a human?
They can ask, and the escalation is queued and visible to a supervisor on the live call. What is not available on the live carrier is the completed transfer itself — the provider capability it depends on is not enabled. Plan the fallback as a call-back rather than a hand-off.
Are calls recorded?
No. Call recording is foundation: the capability is asked of the provider rather than assumed and is not enabled on the live carrier. What you do get for every call is a transcript, a summary, an outcome and hang-up attribution.
How many calls can it take at once?
That is bounded by worker capacity rather than by a headcount, and the documentation does not publish a number, because it depends on how the workers are provisioned. What matters operationally is that a line at capacity says so through line health instead of answering into silence.