A good handover summary lets the next person act within a minute without asking the customer anything again. The agent writes it before the transfer, in a fixed shape: who the customer is and how they were verified, what they want, the facts with sources, what the agent did and what stopped it, and anything promised. It reads the key points back to the customer first.
What is a handover summary, and who owns it?
A handover summary is the note that travels with a case when it moves from the agent to a person. It is short, structured and written for someone who has thirty seconds before they speak to the customer. It is not a transcript and not a case history; those exist separately.
The support operations lead owns the summary format. The receiving teams decide what they need in it, because they are its readers. The field-by-field layout is in the handover note template from an AI agent to a person. Why this moment matters so much is argued in the handover to a person is where support agents are judged. The wider support position is in the AI support agent that knows when to say no.
How does the agent write the summary, step by step?
- Agent, depth 1: gathers the case. The conversation, the records it read, the actions it took or attempted, and every verdict filed during the contact.
- Agent, depth 1: sorts what it knows into three kinds. Facts from your systems, with their source. Things the customer said, marked as the customer's account. The agent's own inferences, marked as such. These never blend.
- Agent, depth 2: writes the summary in the fixed shape. Verification status and method. The request in one sentence. Key facts. What the agent did. What it could not do and which bound stopped it. Promises made to the customer, with any time. Open questions.
- Agent, depth 1: reads the key points back to the customer. "So a colleague will know that you were charged twice for the March order and want the second charge returned. Is that right?" It corrects the summary from the answer.
- Agent, depth 3: files the summary on the case before the transfer, so the person sees it on arrival.
- Platform: files the handover verdict. It links the summary, the reason for handover and the bound that decided.
- Person, receiving colleague: reads, acts and corrects. If anything in the summary is wrong, they correct it on the case, and the correction is kept.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Gather | Reads the contact and its verdicts | 1 | Nothing |
| Sort | Separates facts, claims, inferences | 1 | Nothing |
| Write | Fills the fixed shape | 2 | Changing the format |
| Read back | Confirms key points with the customer | 1 | Nothing |
| File | Adds the summary to the case | 3 | Nothing |
| Act | Nothing | None | Everything after handover |
What makes a summary specific to this case?
The section most summaries miss is "what stopped me". A note that says "customer wants a refund" leaves the colleague to rediscover why the agent did not give one. A note that says "refund of 240 requested; above the customer's limit of 100; held for approval" tells them exactly what decision is theirs. Because every verdict names the bound that decided, the agent can state it precisely rather than vaguely.
The second is promises. Any time the agent told the customer something would happen, such as "a colleague will call you within the hour", it goes at the top. Broken promises made by the agent become complaints against the person who inherits them.
The third is the read-back. Customers correct summaries more often than you would expect: the wrong order, the wrong amount, a second issue the agent did not register. One question before the transfer prevents the most common handover failure.
How does this differ from routine case notes?
Case notes record what happened in every contact, for the file. A handover summary is written for one reader, about one decision, at one moment. They share the discipline of separating facts from claims, and the method for notes after every contact is set out in writing case notes after every contact. The same shape also works outside support; compliance teams use it when an agent passes an alert to an analyst, as in letting an AI agent summarise alerts in your transaction monitoring system.
What should you measure?
- Questions the colleague had to ask the customer again, sampled from transcripts. The main test.
- Corrections to summaries by receiving staff, by section.
- Read-back corrections by customers, which show where the agent mishears.
- Time from handover to first action by the colleague.
- Promises listed against promises kept, checked a day later.
Where does it usually go wrong?
Summaries grow. Every reader asks for one more field until the note is a page long and nobody reads it. Keep the fixed shape and a word limit.
Inferences pass as facts. "Customer is likely to complain" sits next to "payment failed on the 3rd" as if both were known. Label inferences.
The summary is written after the transfer. The colleague picks up the case and starts with nothing. File first, then transfer.
Sensitive details are copied in. A health condition mentioned in passing ends up in a note many people can read. Include only what the next person needs to act, and follow your rules for special-category data.
Where Pryme fits
In Pryme, every verdict filed during a contact names the bound that decided it, so the handover summary can say exactly what stopped the agent rather than guessing. Facts carry their provenance class, which lets the summary keep sealed records apart from what the customer declared.
Questions people ask
Should the customer see the summary?
They hear the key points in the read-back. Showing them the full internal note is rarely useful, but they are entitled to their personal data on request, so write nothing you would not want them to read.
What happens to the summary after the case closes?
It stays on the case as part of the record, alongside the verdicts. Once filed, the case can be reconstructed later; the term is explained in what a filed case means for AI agents.
Can the receiving person ask the agent questions about the case?
Yes, privately and at depth 1. The agent answers from the case record and its verdicts, not from memory of the conversation.