The handoff system manages how calls move between agents and from AI to human. It follows a strict three-level escalation model with a core principle: the agent always asks for consent before transferring a call.
| Level | Handler | When Used |
|---|---|---|
| L1 -- General AI | Your primary agent (e.g., Receptionist or CS) | First point of contact for all calls |
| L2 -- Specialist AI | A secondary agent with specialized knowledge | When the caller's issue requires expertise the L1 does not have |
| L3 -- Human | A human operator at your configured phone number | When the AI cannot resolve the issue or the caller requests a person |
Every transfer follows this protocol:
This consent requirement is built into the agent's tool configuration. The transfer tools include explicit instructions that the agent must ask before proceeding.
Tip: You can customize the transfer announcement message in your routing configuration. Make it specific to your business: "Let me connect you with our billing specialist" sounds better than a generic "transferring you now."
Zentrix Agent Studio supports two strategies for L1-to-L2 handoff over phone calls, each suited to different customer tiers:
Strategy 1: Persona Switch (Standard -- Included)
The same AI agent switches its behavior mid-call. Instead of a real transfer, the agent "becomes" the specialist by activating a different set of instructions and knowledge:
How it works technically: The agent's system prompt includes instructions for multiple "modes." When it detects a specialist topic, it announces the handoff, pauses briefly, and then responds as the specialist persona using the specialist's knowledge base content.
Caller: "My API keeps returning a 500 error."
Agent (as Receptionist): "I see, that sounds like a technical issue. Let me connect you
with our technical specialist. One moment please."
[Brief pause]
Agent (now as Tech Support): "Hi, this is Ravi from the technical team. I understand
you're seeing a 500 error with the API. Can you tell me which endpoint you're calling?"Strategy 2: Dedicated L2 Agent (Enterprise)
A separate AI agent with its own phone number receives the transferred call. This uses native telephony call transfer:
Best for enterprise customers who want true agent separation and independent scaling.
| Feature | Persona Switch | Dedicated L2 |
|---|---|---|
| Transfer experience | Seamless, no tone | Brief transfer tone |
| Context preservation | Automatic (same call) | Via escalation log tool |
| Extra phone number | No | Yes |
| Extra cost | None | Number + telephony |
| Agent independence | Shared agent instance | Separate agent instances |
| Best for | Standard / Bundle | Enterprise |
When a call is transferred (either strategy), the receiving persona or agent gets context about the conversation so far:
Persona Switch: Context is inherently preserved because it is the same underlying AI conversation. The specialist mode has access to everything discussed in the L1 phase.
Dedicated L2: Before transferring, the L1 agent calls a log_escalation_reason tool that sends a summary to your server. This summary includes:
The L2 agent's greeting then references this context: "I understand you were transferred because of an API issue. Let me take a look at that for you."
During a transfer (Dedicated L2 strategy), the caller experiences a brief hold:
The web widget shows a visual hold screen with a countdown during this period.
When AI cannot resolve the issue -- or when the caller explicitly asks to speak to a person -- the agent escalates to L3:
transfer_to_support tooltransfer_call to route the call to the human phone number configured in your agent settings (escalationPhone)The escalationPhone is set per agent instance during setup. It can be a direct line, a call center number, or a mobile number.
The routing configuration includes a max_transfers_per_call setting (default: 3). After reaching this limit, any further transfer attempts automatically route to L3 human escalation instead of another AI agent. This prevents scenarios where agents pass a caller back and forth indefinitely.
All transfers are logged in the call_transfers table with:
Review transfer data in your analytics dashboard to identify patterns -- for example, if 40% of calls are transferred from Receptionist to CS, consider adjusting your Receptionist's knowledge base to handle more support queries directly.
A Product by BRTNeura Technology LLP
Last updated: 2026-03-05