Ask customers what they dislike most about automated service and the answer is rarely the automation itself. It is the moment they reach a person and hear, “Can I take your account number?” after already giving it twice. Every repeated question tells the customer that the first part of the call was wasted. It also costs you handle time on the most expensive minutes of the contact.
Handover is a feature, not a failure
Some AI programs treat a transfer to a person as a failure to be minimized. That mindset produces agents that hold on too long, loop through the same questions and escalate only when the customer is already frustrated. A better starting point is to accept that some conversations belong with a person, and to design the handover as carefully as the rest of the flow.
In our programs, the calls that should move to a human are defined up front. Typical reasons include:
- The customer asks for a person.
- Signs of vulnerability, such as illness, bereavement or job loss.
- A request that needs judgment, like a hardship plan or a policy exception.
- A dispute or complaint that should be handled by a specialist.
- Repeated misunderstanding, where the agent is not getting closer to resolution.
Each of these is a trigger in the agent’s flow, tested in simulation before launch, and reported on afterwards so you can see why calls moved and whether they should have.
What the person should see
A warm handover is only warm if the specialist knows what happened. When a VoiceAI call moves to a human on Bill Gosling Voice, the specialist opens it in LiveAssist with the conversation already on screen. That means:
- The transcript so far, so nothing the customer said is lost.
- Verified identity, so the customer is not asked to prove who they are again.
- The intent and the reason for transfer, so the specialist knows why they are there.
- Actions already taken or proposed, including anything held at the write gate.
- Flags such as vulnerability or a missing disclosure, so the specialist can act on them.
With that in place, the specialist can open with something like, “I have everything you told us, so let’s start with the payment pause.” The customer hears continuity instead of a restart.
The account was my father’s. He passed away last month.
I’m very sorry. I’ll connect you with a specialist who handles this with care.
Hello, I’m Daniel. There is no rush. I can explain what happens next.
Next best action
Pause contact on the account and explain the estate process.
Compliance
- Caller is not account holder
- Limit disclosure ✓
The customer should hear the handover too
Context is not only for the specialist. The customer should know what is happening and why. A short, honest line works best: the agent says it is bringing in a person who can help with this specific thing, and roughly how long it will take. If the wait will be long, the agent can offer a callback instead of holding the customer in a queue with no information.
This is also where tone matters most. A customer who has just mentioned a death in the family or the loss of a job should not hear a cheerful transfer script. We tune handover phrasing for the reason behind it, and test it with synthetic callers who play exactly those situations.
After the handover, one record
A common problem with bolted-together systems is that the AI half of a call lives in one tool and the human half in another. Quality teams score only the human part. Compliance teams cannot see the whole conversation in one place. Reporting treats one customer contact as two unrelated events.
On our platform, both halves land on one record. The AI turns and the human turns are scored together by Auto QA, audited together in the same chain and reported together, so you can judge the contact the way the customer experienced it.
The customer experienced one conversation. Score it, audit it and report it as one.
Staff the receiving end
A handover is only as good as the person who takes it. When an AI agent absorbs the routine calls, the conversations left for people are, almost by definition, the harder ones: hardship, bereavement, disputes and customers who are already upset. That changes what the specialist role needs. Fewer people may handle more complex work, and they need the training, authority and time to do it well.
This is where Bill Gosling’s delivery model matters. Clients can run the human side themselves, or use Bill Gosling specialists and supervisors who are trained for these calls and coached live in LiveAssist while they take them. Either way, plan the human capacity at the same time as the AI flow, not after launch.
Measuring handover quality
Handover deserves its own measures. The ones we look at include how often each trigger fires, how long customers wait between the AI and the person, whether any verification or question was repeated after the transfer, sentiment before and after, and whether the specialist resolved the issue on that contact. Together they show whether handovers are happening at the right moments and whether the context is actually being used.
The bottom line
The best AI agents are not the ones that never transfer. They are the ones that transfer at the right moment, with everything the person needs and in a way the customer barely notices. Get that right and customers will trust the automation, because they know a person is always one clear step away.
← All articles