Conversation Handoff Audit Checklist: From Bot to Human — What Information Must Travel and What Gets Lost

When a conversation moves from an automated response to a person on the team, it almost never fails because of the message: it fails because of the context. The customer already said what they wanted, already chose an option, already sent a voice message, and on the other side someone opens a blank chat and asks everything all over again. This checklist is meant to audit that specific step: what information has to travel, in what order to review it, and how to know the handoff worked before the customer notices.

It is not a "quality of service" checklist in general. It is specifically the moment of the handoff: when the conversation stops resolving itself and a person steps in. It can be applied to any tool that uses Meta's official WhatsApp API, because the data that travels (or does not) depends on the flow the business built, not on the software brand.

What exactly is audited in a handoff

A handoff has three layers. The first is the trigger: what caused the conversation to leave automated mode (the customer asked for a person, the options ran out, X number of messages passed without resolution, it came in outside business hours). The second is the context package: everything the person handling it needs to know so they do not have to start over. The third is the confirmation: the customer knows that someone is now handling it, and who.

Most audits stop at the first layer and check whether the bot "routed well." The real problem is in the second: the context that does not travel. A handoff that triggers perfectly but arrives empty is worse than having no handoff, because the customer feels like they talked to a wall twice.

The context package: what information absolutely has to travel

This is the heart of the audit. Context is not "the entire history": it is the minimum list of data without which the person handling it has to interrogate the customer again. If one of these is missing, the handoff feels like starting from zero.

• The request in the customer's own words: the exact phrase where they said what they wanted, not a category invented by the system.
• The options they already chose: if the bot showed them three paths and they picked one, that data has to be there.
• The data they already entered: name, order number, address, email or any field they filled in during the flow.
• The attachments: images, receipts and audio. An untranscribed audio is lost context for whoever handles it.
• The reason for the handoff: why it left automatic mode. Without this, the person doesn't know if the customer is angry, in a hurry or simply asked to talk to someone.
• The time of the customer's last message, to know if they've been waiting for two minutes or two hours.
• The tags or previous status: if it was already marked as a frequent customer, as a complaint or as a price inquiry.
• The entry channel, because the response changes if the customer wrote via WhatsApp, Instagram or Messenger.

What gets lost along the way (and how to detect it)

Context loss almost always falls into five places. You don't need a special tool to detect it: just take ten random handed-off conversations and compare what the customer said against what the person who handled it had in front of them when they replied.

Checklist to tick off, in order

1. 1Pick ten conversations that went from automatic to human in the last seven days.
2. 2For each one, write down in one sentence what the customer wanted, using only their words.
3. 3Open the conversation as the person handling it sees it and check whether that sentence can be reconstructed without reading anything else.
4. 4Mark whether the reason for the handoff is visible. If it isn't, that's a finding.
5. 5Check whether there are audio messages and whether they're transcribed or at least flagged as pending.
6. 6Verify that the data the customer already entered appears in the same place where the person replies.
7. 7Confirm that the option chosen is marked, not just the options offered.
8. 8Look at the time of the customer's last message and compare it with the time of the first human reply.
9. 9Check whether the customer received any notice that a person was going to handle them.
10. 10Confirm that whoever took the conversation was recorded.
11. 11Note how many of the ten required the person to ask again for something the customer had already said. That number is your baseline.
12. 12Repeat the measurement after every change and compare it against that baseline.

The responsibility table: who does what in the handoff

A handoff doesn't fail just because of missing data: it fails because nobody knows whose conversation it is at any given moment. This table organizes responsibilities so the handoff doesn't end up in no man's land.

What to do when the handoff fails

When you detect that information isn't traveling, the order of fixing things matters. Changing the bot's greeting doesn't help if the problem is that the reason for the handoff isn't recorded. These are the fixes by priority.

• First, the reason for the handoff: if it isn't recorded, everything else becomes guesswork.
• Second, the data the customer already entered: it's what gets asked again the most and what annoys people the most.
• Third, voice messages: if the business receives voice messages, someone has to listen to them or transcribe them before responding.
• Fourth, the timestamp of the customer's last message, so the person knows how long they've been waiting.
• Fifth, notifying the customer: this is what prevents them from feeling like they talked to a wall.
• Sixth, assigning an owner, which is what prevents the conversation from being left without one.

If after reviewing the flow the conversation still arrives empty, the problem may be in the tool's configuration or in how it connects to the WhatsApp API. In that case, it's best to escalate to support with the specific case: which conversation, what was expected to travel, and what arrived. Resolution time depends on each case and on what's failing, so don't promise deadlines to the team or the customer.

How to know if service improved after the fix

A checklist without measurement becomes an opinion. To know if the handoff improved, compare before and after with simple numbers that come from the conversation itself, not from a survey.

Take these five indicators once a week on a small sample. You don't need to measure everything: ten conversations are enough to see whether the handoff is working or whether the team is rebuilding context by hand every time.

The edge cases that break any checklist

There are situations where the handoff gets complicated even if the flow is well built. It's worth planning for them because they are the ones that lose the most context.

• The customer writes outside business hours: if there's an auto-reply for hours, the context has to still be there when someone reads it the next morning.
• The customer sends several messages in a row: if the handoff triggers on the first one, the following ones can be left out of the package.
• The customer changes topic mid-flow: the handoff reason has to reflect the last request, not the first one.
• The customer writes through two channels: if they started on Instagram and continued on WhatsApp, the context has to be able to be joined, or at least it has to be clear that they are the same person.
• The customer sends a long audio: without transcription, the person handling it loses the complete request.
• The conversation is handed off twice: if it returns to automatic mode and goes out again, the context must not be duplicated or lost in between.

Common mistakes when auditing this step

• Auditing the trigger and not the context: it's the most frequent mistake and the one that lets real losses through.
• Confusing a complete history with useful context: twenty messages without the central request are of no use.
• Assuming the customer will repeat what they already said: they won't, they'll leave.
• Measuring once and not measuring again after each change.
• Fixing the bot's greeting before the data that travels.
• Not reviewing audios, which on WhatsApp are half the context in many industries.
• Leaving the conversation without an assigned owner and trusting that someone will see it.

The goal of the checklist is not for the handover to be perfect, but for the person handling the conversation to be able to start responding without questioning the customer all over again. If you achieve that, the rest falls into place on its own: less back and forth, less frustration, and a history that helps improve the automated flow next time around.