Every call center knows the chart: two humps a day, a Monday taller than the rest of the week and a long tail of calls after the shift ends. The rota is built for the average, so the queue grows in the humps and nobody answers after hours. A customer who waited six minutes does not remember the answer afterwards, only the wait.
This article describes how to put an AI voice agent on overflow and after hours so that it works alongside the team rather than instead of it. Without promises the product does not keep.
Three places where calls are lost
- The peak. The queue passes the limit of patience and some callers hang up. Those calls rarely come back the same day.
- After hours. A recording saying "we are open from 8 AM" ends the conversation before it starts.
- Repetitive matters in the middle of the day. Order status, opening hours, a change of date. They occupy a consultant while someone with a harder problem waits in the queue.
Adding seats solves the first problem only in the hump, and only until traffic grows again. It does not solve the second at all, unless somebody pays for a night shift.
How the agent takes overflow
The simplest setup needs one rule in your switchboard and nothing else: when the queue passes a set threshold or the shift is over, the call is forwarded to the agent's number. The agent answers immediately.
Two things matter for planning:
- Inbound calls do not use outbound channels. A campaign working through a contact list at the same time does not compete with the helpline for lines.
- You pay for minutes, not for seats. An agent that answers no calls at night costs nothing at night beyond the subscription.
If your team works in Daktela, the PBX can be connected to the agent over SIP. A transferred call then returns to your queue or an internal number rather than to an outside number. We set that connection up on our side.
There is also a variant without forwarding: your PBX sends us the missed call by webhook, and the agent calls the customer back.
When the agent hands over to a consultant
The agent answers from the material it is given: an FAQ, terms, a price list, procedures. It hands the call over when the caller asks, when the question is outside that material, or on conditions you set yourself.
The handover itself works like this: the agent can announce it and rings the number you set, while the caller stays on the line. If nobody picks up within 30 seconds, the call returns to the agent, which takes a message and promises a call back.
What the agent will not do The consultant receives the call alone, without a transcript of what the customer already told the agent. The transcript and summary are in the dashboard once the call has ended. There is no live listen-in either. If your process needs those, it is better to know before you deploy.
What is left after every call
A shift lead has a recording and a transcript of every call, plus analysis: a summary, the caller's mood, the topic, action items and the outcome with its reason. With the Daktela integration a ticket is opened after the call with the summary, the outcome and a link to the recording, so the consultant who calls back in the morning does not start from nothing.
That matters more than it seems. After hours the agent will rarely close a complaint, but it can always take it, collect the details and leave a ticket that is ready to work on in the morning.
Where to start
- Pull the number of abandoned calls by hour for the last month from your switchboard. That is your overflow.
- List the five most common topics of those calls. If most are questions with a ready answer, the agent closes them itself.
- Start with one slice: after hours only, or above the queue threshold only. After two weeks you have recordings and numbers to decide about the rest.
The details, including the SLA, the data processing agreement and prices, are on the AI voice agent for call centers page, along with a form to book a call. How to work out the cost is covered separately in the cost of an AI minute against the cost of a seat.
