We have written about what Agent Teams do, and they are good at it. This post is the other side of that: a team costs more per minute, takes longer to set up, and adds things to test. Many jobs do better with one agent, doing one thing clearly.
Start with one agent if
- The call has one purpose: confirm this order, remind about this payment, book this visit.
- A single script and a single knowledge base can answer almost everything callers ask.
- Most of your calls are short and most callers answer a question or two.
- When something falls outside the job, a person should take it anyway.
A COD check is a one-agent call. So is a day-before reminder. Adding specialists here means paying more to route a call that has nowhere to go.
Consider a team if
- One call regularly needs two different jobs, such as a prospect who asks a support question mid-pitch and then wants to return to the pitch.
- Your knowledge has real walls. You don't want support internals turning up in a sales answer.
- Part of the call is sensitive, such as a payment or a record, and should sit behind an identity check.
- A single script has grown so long and full of “if the caller asks about…” that it has become hard to edit safely.
That last one is often the real signal. When every change to the script breaks something else, splitting the job can make it manageable.
What actually happens on a team call
Your customer stays on one call. A lead agent greets them and pulls in the right specialist from what they ask, without a menu. It does not flip-flop mid-sentence, the number of moves on a call is limited, and what the caller asks for comes first. A short brief carries the context at each handoff. If the wait is real, they hear a soft tune, and a one-second pause stays silent. Each specialist can have its own knowledge.
On Standard and Ultra, each teammate can have its own voice, so the caller can hear that someone new picked up. On Premium and Ultra Premium, a single voice stays for the whole call.
A worked example
Say a lender wants to remind customers about an EMI, and customers keep asking account questions during those calls. A lead agent takes the call. The caller says, “Before that, what is my closing balance?” An account specialist answers after an identity check, then the reminder resumes. With one agent, the script needs both jobs plus the check logic. With a team, each specialist has a short script and its own documents.
Now compare a lender who only wants the reminder, and passes every account question to a person. One agent with a good hand-over list does that for less money.
A team is not a desk
These get confused. A team template, such as a clinic front desk, is a ready-made team of AI agents. A desk is a group of your own staff who take the calls an AI hands over. You can have both. When the team runs out of road, a colleague is rung and joins the same call, after a private brief, on plans that include that.
Testing a team takes more work
A team has more ways to go wrong: the wrong specialist, a handoff that loops, context that does not carry over. Fortunately it can all be rehearsed.
- A team's agent test suite runs through its coordinator, so routing and handovers are tested too.
- A Simulation Lab rehearsal runs the greeter, the routing and each specialist against AI customers, where your plan includes it.
- Then make a real test call. Ask two questions in one sentence, then ask for a person.
Keep your first team small. A lead and two specialists will teach you more than a lead and six.
Cost, plainly
A team's minutes cost more than a single agent's, at a rate for each voice tier, and your plan sets how many active teams you can have. Current figures are on the pricing page. Work out the difference on your own minutes. If a team would only tidy up your script a little, the higher rate on every minute is hard to justify, so stay with one agent.
Building one
No code is needed. Start from a ready-made team, describe the team to Build with AI, or assemble it yourself. A new team is a Draft and takes test calls only until you press Go live, which switches on every inbound route, running campaign, workflow and web widget that points at it. You can always start with one agent, and move to a team once you have read enough real calls to know where a single script is struggling. Real calls will show you the seams better than planning does.