An AI call centre is not one clever voice agent. It is a small phone operation: several voice agents with separate jobs, routing that sends each call to the right one, people who pick up what the agents should not handle, and reporting that shows what happened to every call. The levelpartner.ai platform gives a firm of 10 to 250 people that set-up without a telephony project. This article covers how to design the whole operation, from routing to staffing to the weekly review. Configuring an individual agent is covered in the AI call center agent guide.
Design the call flows before the agents
Start with a list of why people call you, roughly ranked by volume. A typical service firm has five or six reasons: new enquiries, existing clients with a question, booking changes, suppliers, complaints and wrong numbers. Each reason becomes a flow with an owner and an outcome.
Then decide which flows an AI agent handles, which go straight to a person, and which start with an agent and hand over. New enquiries and booking changes are usually good agent work. Complaints usually are not. Write the decision down before you open AI Agents.
Before you start
- Your list of call reasons, the flow for each and the outcome that counts as done.
- Phone numbers in the platform: one main number, plus separate numbers for any lines you advertise separately.
- Staffed hours for each team that receives transfers.
- Calendars for anything an agent may book.
- A recording and disclosure policy that covers callers in every country you serve.
How to set it up
- Go to AI Agents > Voice AI and open Agent List.
- Click + Create Agent for the front-door agent.
- Select Create Custom Agent, then click Continue.
- Enter an Agent Name, then select a Voice.
- Select a Model and Language, and set the Timezone.
- Write a prompt that finds the call reason and routes it.
- Configure the Welcome Message and click Save.
- Open Actions and click + New Action.
- Select Agent Transfer as the type for a specialist agent.
- Name it, define its trigger condition and save the action.
- Repeat with Call Transfer for reasons that go straight to a team.
- Open the Deploy tab and select your main number.
- Under Call Routing, choose Answer Calls Directly.
- Under Working Hours, choose All Hours or Custom Schedule.
- Save the deployment settings.
Staffing around the agents
The agents change what your people do, not whether you need them. Someone has to own transfers in staffed hours, someone has to clear callbacks each morning and someone has to review the agents every week. In a 40-person firm that is often one person's part-time role. Name that person before launch.
Hand-overs should land with context. When the front-door agent transfers a caller, it should pass the reason and any details collected, so the person does not ask the caller to repeat everything. A caller who repeats themselves twice will not call again.
Suppose the agency receives 200 calls a week. A front-door agent would identify the reason in the first exchange. New business enquiries would go to an enquiry agent that qualifies and books a call with a director. Existing clients would transfer to their account team, or book a callback out of hours. Supplier and sales calls would get a polite email address. Reception time would then go only to the calls that need a person.
What an AI call centre replaces
It replaces the reception desk switchboard, the voicemail box nobody clears and the out-of-hours answering service that takes a message and nothing more. It does not replace your account managers or advisers. It means their phone only rings for calls that need them.
It also gives you something you rarely have: a record of why people call. After a month you will know which calls are growing, which are avoidable and which point at a problem elsewhere, such as unclear invoices generating questions.
Firms often build the specialist agents and leave routing to a menu of "press 1, press 2". Callers dislike menus and many press zero. Let the front-door agent ask one open question and route on the answer. It is faster and you learn the real reason for the call.
Weekly running time
Plan on an hour a week across the whole operation: review the dashboard, listen to calls that ended without an outcome and adjust the briefs. Once a month, revisit the routing table, because call reasons drift as your business changes. Minutes and numbers are billed as used.
When not to build one
If you receive fewer than a handful of calls a day, one agent covering missed calls is enough. A full AI call centre makes sense when the phone is a real channel with several kinds of caller. Do not build one if nobody can own the weekly review. Unreviewed agents drift, and callers feel it before you see it in reports.
Troubleshooting
Callers get bounced between agents. The front-door agent routes on keywords and misreads mixed requests. Let it ask one clarifying question when the reason is unclear, and transfer to a person after two tries.
Transfers ring out. Staffed hours in the transfer rule do not match reality. Align the hours and set a fallback callback booking.
Reports show unknown reasons. The post-call workflow is not tagging consistently. Use a fixed list of reason tags and have each agent set one.
Out-of-hours callers feel stonewalled. Make sure the out-of-hours flow gives a specific callback time, not "someone will be in touch".
Frequently asked questions
How many voice agents does an AI call centre need?
Usually two to four: a front-door router and one per high-volume flow. Adding more makes testing harder without much gain. Merge flows that share most of their answers.
Can the AI call centre handle callers in different countries?
Yes, with separate numbers for each market and briefs that reflect local opening hours and terms. Keep disclosure and consent rules for each market in the relevant agent's greeting. For wider context, read artificial intelligence call center.
Does it work alongside our existing phone system?
Many firms forward an existing number to a platform number, so callers notice nothing. If you keep a separate phone system for internal calls, route only external lines through the agents. Our guide to the AI receptionist explains a lighter first step, and conversational AI contact center covers larger set-ups.