What is a handoff?

A handoff is the moment when a piece of work passes from one party to another: from an AI agent to a person, from a person back to an agent, or from one agent to another. A good handoff carries the full context, so whoever receives the case can act without starting again or asking the customer to repeat themselves.

Why does a handoff matter once agents act?

Customers judge agents most harshly at the handoff. An agent that handles easy questions well, then passes a difficult case to a person who knows nothing about it, has made the experience worse. A poor handoff also wastes staff time and loses facts the agent gathered.

Handoffs are also where accountability moves. When an agent hands a case to a person, the person now owns the decision. If the record does not show what was handed over and when, nobody can later say who decided what.

How does a handoff work in practice?

A useful handoff includes:

• the reason: why the agent is passing the case on (outside its role, above its depth, customer asked for a person, uncertainty, a bound it hit);
• the state: what has been done, what is pending, any deadlines;
• the facts: what the agent found, with the source of each;
• the draft, if there is one, so the person can approve or change it rather than write from scratch;
• the customer's words, so the person can see the request as it was made.

Triggers should be explicit. An agent should hand off when the case is outside its bounds, when it is unsure, when the customer asks for a person, and when signs of vulnerability or complaint appear.

What does an example look like?

A UK insurer's agent handles first notification of loss for home claims. A customer reporting water damage mentions they are elderly and living alone, and the kitchen ceiling has come down. The agent records the claim, recognises signs of vulnerability and urgency, and hands off to a claims handler with the claim details, photographs, the customer's own words and a note of the vulnerability indicators. The handler calls back within the hour without asking the customer to repeat anything. This kind of claims work is described in AI agents in UK insurance operations.

How is a handoff different from an escalation?

An escalation is one kind of handoff: passing a case upward because a decision is beyond the agent's authority or a problem needs someone more senior. Handoffs also go sideways (to a specialist team), back to the agent (a person resolves a blocker and returns the routine remainder), and between agents (an intake agent passes to a servicing agent). Escalation is about authority; handoff is about moving work with its context, in any direction.

Between agents, the receiving agent should get only what its task needs, and the record should show which agent did what.

Where do handoffs matter most?

In wealth management, a client's request to change an investment strategy might start with an agent gathering details and end with an adviser, who must see the full picture to give suitable advice. That pattern is set out in AI agents in UK wealth management operations.

Which terms sit close to it?

Escalation, held queue, exception queue and human-in-the-loop all describe parts of the hand-over between agents and people. For how handoffs are designed before go-live, see our guide to deploying AI agents.

Where Pryme fits

In Pryme, a held action is a handoff with its context attached: the drafted action, the facts with their provenance, and the bound that caused the hold. The person who approves or refuses it is named in the filed verdict, so the record shows exactly where responsibility passed.