Why the handover decides: AI agents and context

A customer writes into the chat one evening: his direct debit has been taken twice. The AI agent asks for his customer number, checks the account, finds both payments and recognises that a refund needs approval.

Then it hands over. What happens next decides the whole conversation.

In the bad version, the employee opens with "How can I help you?". The customer tells the story again, gives his customer number a second time and wonders why he bothered talking to the bot at all.

In the good version, she says: "I can see the double debit from 3 February, and I'll take care of the refund." The difference is not a smarter AI. It is a better handover.

#Why the handover is the critical moment

Automation is usually judged by how many requests it closes on its own. That is a fair question, but only half of one.

Some conversations will always reach people: because a decision is needed, because the situation is unusual, or because the customer simply prefers it.

For those conversations, what counts is not what the AI agent could do but what it passes on. A handover without context turns a half-finished request into a brand-new one, and the time the customer spent with the agent is lost.

#What a good handover contains

A good handover answers the questions the person taking over asks in the first few seconds. In practice, there are six building blocks.

The reason for handover is the block most often forgotten, yet it is the most valuable sentence. "Refund needs approval" tells the employee what to do next. "Customer wants to speak to a person" tells her to listen first.

#Context has to arrive in the right place

A complete log is of little use if it sits in a separate window. Context belongs where the employee already works: in the agent desktop, right next to the conversation, with the summary on top and the history below.

Service EN· active

Queues

Voice3 waiting

Chat5 waiting

Email2 waiting

Messaging1 waiting

Conversation

AI agent, disclosedHandover from AI agent · context complete

I received the decision letter, but the amount does not match my application.

I have the case in front of me. One moment, I am checking the calculation.

Sentiment: calmVoice

Context

Case

2026-0412

History

3 · 14 d

Univerx Assist

Suggested reply

The letter deviates from the application. I will propose a correction and send you the confirmation by message.

InsertAccept

Summary

Request: deviation in the letter. Documents on file. Next step: correction, confirmation by message.

Who takes over matters just as much. The handover should pass the recognised request to call distribution, so the conversation reaches someone with the right skills.

A perfect handover into the wrong queue creates a second handover, and with it exactly the repetition you were trying to avoid.

#The channel may change, the context may not

Not every handover stays in the same channel. A voicebot hands over to a phone call, a chat turns into a callback, a message becomes an email with documents attached. The context has to survive each of these switches.

That works when every channel runs into one shared queue and one shared case. The handover is then not an export between systems, just a change of hands within the same case. More on this in One queue for every channel.

#What the customer experiences during the handover

The customer needs a handover too. The AI agent says that it is passing the conversation on, why, and what comes next: "The refund needs approval. I'm connecting you with a colleague, and she can see our conversation so far."

That last half-sentence is more than politeness. It removes the worry of starting from scratch, and it is a promise the handover then has to keep.

If there will be a wait, say so in the same message, and offer a callback. How to design that is covered in Callback instead of hold.

#Designing and reviewing handovers

Good handovers do not happen by accident. They are defined when the agent is built, for example in Agent Studio: which fields the agent collects, when it hands over and what summary it produces.

After that, real conversations are the best teacher. Three questions help with the review:

Every "no" points to a specific fix: a missing field, a summary that is too long, an imprecise match. That is more work than tracking a single automation rate, but it improves the very moment customers remember.

#Support continues after the handover

AI support does not stop once a person takes over. Univerx Assist suggests replies during the rest of the conversation, surfaces knowledge and writes a summary for the case at the end. The employee decides what to use.

The result is a chain without a break: the AI agent does what it can do reliably, hands over with context, and the employee carries on with support. To the customer, it feels like one conversation.

The guide Introducing AI agents responsibly shows how to roll out AI agents step by step. The glossary explains AI agent and AI assistance.