Treat handover as a service workflow, not a fallback button.
Customers normally ask for a person because the issue is important, unusual or already frustrating. The transition therefore needs defined ownership, response expectations and a clear customer message.
A handover workflow should decide who can receive the conversation, whether that person is available and what happens when immediate support is not possible.
Define when the AI should escalate.
An explicit request for a human should be handled consistently. Organisations may also configure escalation for sensitive complaint types, repeated failed answers, low-confidence responses or actions that require staff authority.
These rules should be understandable to service owners and tested against realistic conversations. Escalating everything removes the benefit of automation, while escalating too little creates a poor customer experience.
- The customer directly asks to speak to a person
- The request needs an authorised decision
- The conversation reaches a configured risk or sensitivity boundary
- The automated route cannot confidently progress the request
Control agent capacity and the waiting queue.
Availability is more than an online status. Each agent should have a manageable number of active conversations, and new requests should enter a visible queue when that limit is reached.
The customer should be told whether the transfer has started, whether they are waiting and what alternative follow-up is available outside operating hours. This avoids promising an immediate human response when no agent can accept the conversation.
Transfer enough context for the agent to continue.
The agent needs the conversation history, the customer’s stated need and any enquiry or complaint reference already created. Relevant workflow fields should be visible without requiring the customer to repeat them.
Access must still follow the organisation’s permissions. A complete transcript should not become a reason to expose information to an agent who is not authorised to handle that customer or request type.
Measure the handover as part of the whole journey.
Useful measures include queue length, acceptance time, conversations per agent, unresolved transfers and the reasons customers asked for a person. These signals help teams improve both the AI route and workforce planning.
Rosebay’s handover model connects configured Emily conversations to agent availability, capacity-aware queues and the existing customer context. Organisations remain responsible for their service rules, staffing and response commitments.