Chatbot CSAT after human handoff tells you whether escalation repaired the customer experience or added another delay. Measure the completed interaction, not the moment the AI agent transfers the conversation.
Define the handoff cohort before you measure CSAT
Start with support and CX leads who already use an AI support agent with live-agent escalation. Define exactly which conversations belong in the report.
Count a conversation as an AI-to-human handoff when the chatbot transfers the customer to a human or invites the customer into a human interaction. Keep these conversations separate from chats resolved entirely by the AI agent. They represent different support experiences and should not share one blended CSAT score.
For every handoff, record:
- The handoff reason.
- Whether the customer requested a human or the agent triggered the handoff.
- The time between the handoff request and the human response.
- Whether the human received the full transcript, a summary, or neither.
- Whether the issue was resolved in the same interaction.
- Whether the customer contacted support again about the same issue.
Use the conversation transcript and handoff event as the unit of analysis. A survey response is only one part of that record. The transcript can show whether a low score came from a wrong AI answer, a cold transfer, a long wait, or an unresolved human interaction.
Send a survey that measures the completed support interaction
Send the CSAT survey after the human agent closes or completes the interaction. A survey sent when the transfer begins measures the customer’s reaction to waiting, not the full support experience.
Use one direct question tied to the entire interaction:
How satisfied were you with the support you received today?
Add one optional open-text question:
What could have gone better?
A short survey keeps the response focused on support. Qualtrics’ CSAT guidance covers the standard satisfaction-score approach, while SurveyMonkey’s customer satisfaction survey guide provides examples of short survey questions.
Store the response with the handoff flag, handoff reason, transfer delay, resolution status, first-contact resolution result, and context passed to the human. Report response count and survey response rate beside the score. Do not treat every unanswered survey as a dissatisfied customer.
Build the weekly post-handoff scorecard
Report post-handoff CSAT as its own segment. Keep non-escalated CSAT beside it as a reference point, not as part of the same average.
| Metric | Definition | Why it belongs in the report |
|---|---|---|
| Post-handoff CSAT | Satisfaction among customers whose conversation reached a human after AI involvement | Shows how customers experienced the completed escalated interaction |
| Non-escalated CSAT | Satisfaction among conversations resolved entirely by the AI agent | Provides a separate reference for AI-only support |
| Escalation rate | Eligible chatbot conversations that reached a human, divided by eligible chatbot conversations | Shows how often the AI agent needs human help |
| First-contact resolution | Conversations resolved without another support contact | Shows whether the interaction finished the issue |
| Transfer delay | Time between the handoff request and the human response | Shows whether waiting is part of the problem |
| Repeat contact | Customers who contacted support again about the same issue | Shows whether the first interaction held up |
| Response count and response rate | Number of CSAT responses and the share of invited customers who responded | Gives the score context |
Define the denominator before publishing the report. Eligible chatbot conversations might exclude test chats, internal conversations, or chats that ended before the agent could answer. Use the same rule each week.
First-contact resolution also needs a clear boundary. Count a conversation as resolved when the issue is handled without another support contact. If the human sends the customer to a separate channel and the customer must start over, do not count it as a clean resolution.
Break the score down by handoff reason
Create reason tags that match your support operation. Billing, account-specific requests, product questions, bug reports, and explicit requests for a human are practical starting points.
For each reason, compare post-handoff CSAT, transfer delay, first-contact resolution, repeat contact, and survey response rate. A low billing score with a long transfer delay points to a different problem than a low product-question score after a fast transfer. Review the transcript and the context passed to the human before deciding what to change.
A high escalation rate with strong post-handoff CSAT may mean the AI agent is routing difficult conversations correctly. A low escalation rate with poor post-handoff CSAT may mean customers reach a human too late, or the human receives an incomplete record.
Use the result to assign the fix to the right part of the operation:
- Knowledge base: The agent escalates because the answer is missing or unclear.
- Handoff rules: The agent waits too long or transfers the wrong request.
- Human workflow: The agent receives the conversation but lacks ownership or context.
- Response coverage: Customers wait too long during a predictable period.
Some conversations should reach a person. Reducing escalation rate is not the goal by itself.
Inspect the handoff experience, not only the score
Check whether the human agent received the customer’s original question, prior AI replies, relevant customer details, and actions already attempted. If the customer repeats the same explanation, the transfer created work instead of removing it.
Review conversations with long transfer delays and low CSAT. Look for repeated explanations, missing context, unclear ownership, and customers sent elsewhere after the human takes over.
Compare two groups separately:
- Customers who asked for a human.
- Customers transferred because the AI agent could not answer or needed help.
The first group may be expressing a preference. The second may be encountering a knowledge gap or a handoff rule. Their scores should not be treated as evidence of the same problem.
Track whether the human resolved the issue in the same interaction or sent the customer to another channel. That outcome belongs beside CSAT because a friendly conversation that ends without a resolution still leaves work for the customer.
AssistLoop Human Handoff is one implementation example. The AI agent answers from company content, then hands the conversation to a human when needed. Your measurement should preserve both parts of that interaction.
Run the playbook with AssistLoop
Start with the knowledge base. Train your AssistLoop agent on your own content before measuring handoff quality. AssistLoop supports uploaded PDF, DOCX, and TXT files, website crawling, pasted text, and exact Q&A pairs. Exact Q&A pairs help preserve wording for policies such as refunds.
Embed the agent in your website widget and identify conversations that move from AI support to a human. Review the shared conversation history when a human takes over. Apply the handoff reason, resolution status, transfer delay, and repeat-contact fields from your scorecard.
Use AssistLoop’s Human Handoff workflow to keep the AI and human portions in one conversation record. That gives your team the context needed to review a low score instead of relying on the survey number alone.
The wider AssistLoop features cover the rest of the support automation workflow. AssistLoop is designed around website support. If you need WhatsApp, Telegram, Instagram, or Messenger support today, those channels are listed as coming soon rather than shipped features.
Turn the report into changes your team can test
Set a regular review cadence for post-handoff CSAT, escalation rate, first-contact resolution, transfer delay, and repeat contact. Weekly review works when conversation volume supports it. If response count is small, use a longer window and label it clearly.
Choose one low-performing handoff reason for each review cycle. Test one change, such as adding a missing training source, passing more customer context to the human, changing the handoff trigger, or adjusting response coverage during a period with long delays.
Compare the same handoff reason before and after the change. Keep non-escalated CSAT as a separate reference point. Do not judge the change by overall chatbot CSAT alone, because a shift in the mix of escalated conversations can hide the result.
Escalation rate is not a quality score by itself. The useful question is whether the handoff helps the customer finish the interaction. A good report connects the trigger, the wait, the context passed to the human, the resolution, and the customer’s response.
When your team is ready to test the workflow, create your AI agent.
FAQ
How should you measure CSAT after a chatbot hands a conversation to a human?
Send the CSAT survey after the human interaction is complete, then tag the response as an AI-to-human handoff. Report it separately from conversations resolved entirely by the AI agent.
What should a chatbot-to-human handoff survey ask?
Use one short satisfaction question about the completed support interaction and one optional open-text question about what could have gone better. Store the handoff reason, transfer delay, and context passed to the human with the response.
Should escalated conversations have a separate CSAT score?
Yes. Escalated conversations include both an AI interaction and a human interaction, so they need their own segment. Comparing that result with non-escalated CSAT shows whether the handoff is helping.
Which metrics belong beside post-handoff CSAT?
Report escalation rate and first-contact resolution beside post-handoff CSAT. Add response count, survey response rate, transfer delay, and repeat contact so the score has operational context.
When should you send a CSAT survey after live chat escalation?
Send it after the human agent closes or completes the interaction, not when the transfer starts. The survey should evaluate the full support experience while the record preserves the handoff reason and transferred context.
Written by