An artificial intelligence call center gives a firm that has never had a call centre the thing a call centre exists for: every call answered, every caller routed, every callback made. This article answers one question for a 10 to 250 person company: which call types should you hand to an AI voice agent first, and in what order? The order matters more than the technology. Get it wrong and callers meet a voice agent at the moment they most want a person.
You do not have a call centre, and that is the point
Most established service firms run phones the same way. Calls go to reception or a main line, then to whoever is free. Out of hours they go to voicemail. Callbacks depend on someone writing a note and someone else reading it.
That arrangement works until volume rises or a key person is on leave. Then calls ring out, voicemails sit for a day and new enquiries call the next firm on the list. A traditional call centre fixes this with headcount and shift cover. An AI voice agent fixes it with a system that answers every line at once, at any hour, and writes everything to the CRM.
The call inventory method
Before you configure anything, list your calls. Pull two weeks of call logs, or ask reception to tally calls by type for a week. For each call type, record four things: how many arrive, when they arrive, how long they take, and what happens if nobody answers.
Most firms find five or six call types cover almost everything. A typical list looks like this:
- New enquiries from people who have never dealt with you.
- Existing clients asking about their account, project or case.
- Booking, rescheduling and cancelling appointments.
- Calls for a named person.
- Supplier, recruitment and sales calls to you.
- Calls you owe: callbacks, chasers and follow-ups.
Suppose the firm logs 640 inbound calls in two weeks: 180 new enquiries, 210 existing clients, 120 booking changes, 90 calls for a named adviser and 40 other calls. Say 52 of the new enquiries arrive after 6pm or at weekends and reach voicemail, and 19 of those never call back. Add around 150 outbound callbacks owed each fortnight, with about two thirds done on time. An audit like this shows where an agent would earn its keep first: the 19 lost enquiries and the 50 late callbacks.
The inventory tells you where calls are lost, which is where the AI belongs first.
The order that works
Rank call types by two questions: how much revenue is lost when the call goes unanswered, and how much judgement the call needs. Put the AI first where loss is high and judgement is low.
First: after-hours and overflow for new enquiries
An unanswered new enquiry is the most expensive missed call you have. The caller has no loyalty yet and will ring someone else. The judgement needed is low: take the details, ask three qualifying questions, offer a meeting. A voice agent can do all of it and write the result straight to a new contact and opportunity.
Start with after-hours only, then add daytime overflow when the line is busy. Reception keeps the daytime calls it can answer.
Second: booking, rescheduling and cancelling
These calls are short, frequent and rule-bound. The agent checks the calendar, offers times, books and confirms. It also handles the reschedule that would otherwise become a no-show. See AI booking assistant for how booking rules should be written.
Third: outbound callbacks
This is the call type most firms forget they have. Missed-call text-backs, enquiry callbacks, appointment reminders and document chasers are all calls you owe. An AI voice agent can make them on schedule, and a workflow can pass anything complex to a person.
Last, or never: existing clients with account questions
Existing clients calling about their case usually want the person who knows it. A voice agent can triage, confirm identity and take a message for the account lead, but it should not try to resolve a complaint or a complex question. Keep this as a routing job.
Where most people get the order wrong
The common approach is to start with the main line during office hours, because that is where most calls are. It is the worst place to start. Daytime callers are the ones most likely to reach a person anyway, and the ones most annoyed to meet an AI instead. The first impression of your new set-up becomes a complaint.
Start where no person is answering today. Out-of-hours enquiries and overdue callbacks have no human service to compare against, so the agent can only improve things. Once it is reliable there, you have evidence and confidence to move it into daytime overflow.
Setting it up on the platform
- Go to AI Agents > Voice AI and open Agent List.
- Click + Create Agent, select Create Custom Agent, then Continue.
- Name the agent for new-enquiry calls and select a Voice.
- Select a Model and Language, and set the Timezone.
- Write the prompt: who you serve, three questions, when to hand over.
- Configure the Welcome Message and click Save.
- Open Actions, click + New Action and select Appointment Booking.
- Name it, define its trigger, choose your calendar and save it.
- Repeat with Trigger a Workflow so qualified calls reach the owner.
- Open the Deploy tab and select your main number.
- Under Call Routing, choose Answer Calls Directly or Use as Backup.
- Under Working Hours, set a Custom Schedule outside office hours.
- Save the deployment settings.
For the inbound side in more detail, AI phone answering covers greetings, transfers and voicemail fallbacks.
What it replaces
For most firms in this size range, it replaces three things: voicemail as the out-of-hours answer, the sticky-note callback list, and the part-time hire you were about to make for reception cover. It does not replace your receptionist or your advisers. It changes their day from answering every call to handling the calls that need them.
Imagine two staff answer the main line and around 40 enquiries a week go to voicemail after 5.30pm. A voice agent would answer those calls, qualify the caller and book valuations into negotiators' calendars. The hour the two staff would have spent clearing voicemail could go on calling back the hot leads the agent flags.
Running time each week
Expect two hours a week in the first month: listen to a sample of calls, read hand-offs and update the instructions. After that, an hour a week is usually enough. The owner of this work should be someone who understands both the service and the callers, not only whoever set up the phone system.
The trade-off
You accept that some callers will dislike speaking to an AI, and a few will hang up. You trade that against every call being answered and every caller being recorded. For out-of-hours enquiries the trade is easy, since the alternative is voicemail. For daytime calls from loyal clients it is harder, which is why the order matters.
You also accept a compliance task. Calls are recorded and transcribed, so your privacy notice and call greeting must say so, in line with the rules in each country where you operate.
Firms write a voice agent's instructions like a web page: long, complete and formal. On a call, that becomes a monologue. Write instructions for short spoken turns, one question at a time, and test them by phoning the agent yourself.
Frequently asked questions
How is this different from an IVR menu?
An IVR asks callers to press numbers and sends them to a queue. A voice agent holds a conversation, answers questions, books appointments and writes the result to the contact record. Callers do not navigate anything.
Can the agent transfer to a person mid-call?
Yes. You define when it should transfer, such as an existing client asking for their adviser, and to which number. If nobody picks up, it takes a message and a workflow alerts the right person.
Do I need a new phone number?
No. You can keep your existing number and route calls to the agent by time of day or when lines are busy. For a broader view of AI on the phone, see the AI call centre article.
What should I measure in the first month?
Measure answered calls that used to reach voicemail, meetings booked by the agent and callbacks completed on time. Those three numbers show whether the set-up is recovering revenue you were losing.