The most common reason voice agent projects fail is not that the agent speaks badly — it is that it does not stop where it should. An agent that keeps trying on a problem it cannot solve leaves the customer both unresolved and annoyed. Had that call been handed to a human, nothing would have been lost.
Handoff is not a weakness of the agent; it is part of the design. Set up correctly, customers are glad they spoke to the agent. Set up badly, trust in the whole channel goes.
Four handoff triggers
1. The customer asks. The clearest trigger. The rule should be: on the first request the agent may try once more to help; on the second it hands off without argument. Resisting with "I can help you with that" is the fastest way to produce dissatisfaction.
2. The nature of the topic. Some topics sit outside the agent's authority and that must be defined up front: negotiating refund amounts, contract termination, complaints with legal content, payment disputes, medical advice. On those the agent informs but does not decide.
3. Emotional intensity. A customer whose anger becomes clear, who raises their voice, or who repeats the same complaint should be handed off. At that point technical accuracy stops mattering; the customer wants a counterpart, not a solution.
4. The call stalling. If the agent has answered the same question twice and the customer is still unsatisfied, a third attempt will not help. Likewise, a call that exceeds your set duration without progress should be handed off.
Three handoff formats
How the handoff happens matters as much as the decision to hand off.
Live transfer. Best when the team is available. The critical point: the agent must summarize the matter before transferring, and the representative must see that summary. Making the customer explain everything again nullifies the benefit.
A promised callback. If the team is not available, the agent gives a specific commitment: "we will call you back by five today." That promise must become a task and be tracked. An unfulfilled callback promise is worse than no promise.
Channel switch. Some topics are better handled in writing — sending documents, filling long forms, sharing images. The agent can close the call and move the matter to WhatsApp.
Which format is used should be governed by rules based on time of day, team availability and topic type. Leaving it to the agent's in-the-moment discretion produces inconsistency.
When writing thresholds
Thresholds must be concrete, not open to interpretation. A working set looks like this:
- The customer asks for a representative a second time: hand off.
- The topic is a refund amount, a cancellation penalty or a complaint: hand off.
- The call passes eight minutes without completing the transaction: hand off.
- The agent has given the same information twice and the customer is still asking the same question: hand off.
- The customer expresses clear anger: hand off.
- There is no answer in the knowledge base and the topic concerns price, delivery or contracts: hand off.
Adapt these to your sector. A technical support business needs a longer duration threshold; an order-taking business a shorter one.
Over-handing-off is also a problem
There is a failure in the other direction: an agent that hands off at every hesitation. That design drowns the team and removes the agent's reason to exist.
The symptom of over-handoff is that in most handed-off calls the representative repeats what the agent already said. Look for that when listening to recordings: is the representative doing something new, or restating the same information? If it is the latter, loosen the thresholds.
The way to find a healthy balance is to measure not the handoff rate but handoff quality.
Measurement
Four numbers are enough to evaluate a handoff design:
- Handoff rate. What percentage of calls were handed off?
- Post-handoff resolution rate. What percentage of handed-off calls were actually resolved? If it is low, the problem is on the team side, not the agent.
- Unnecessary handoff rate. Calls where the representative repeated the agent's information. Measured by sampling recordings.
- Late handoff rate. Calls handed off only after the customer became angry. This number shows thresholds firing too late, and it is the most expensive mistake.
The fourth matters most, because that is where customer loss comes from. A late handoff does more damage than no handoff: the customer has lost time and, having had to reach a human in the end, remembers the agent as a pointless obstacle.
Do not forget the team side
A handoff design does not work without a counterpart on the team. Where a handed-off call lands, who owns it, and how quickly it is answered must all be defined. A queue that receives handoffs nobody looks at is the quietest way for this design to fail.
You can see how conversations gather in a single inbox on the omnichannel page.