Chatbot-to-Human Handover: Best Practices

The best AI chatbots know when to stop being chatbots.

No matter how good your AI is or how comprehensive your knowledge base is, there will always be situations that require a human touch — complex billing disputes, emotional customers, edge cases your documentation doesn't cover, or questions where getting it wrong has real consequences. The mark of a great chatbot isn't that it never hands off. It's that it hands off at exactly the right moment, with full context, to the right person.

This guide walks you through setting up seamless chatbot-to-human handover in LoopReply — from designing your escalation triggers to configuring agent routing, preserving conversation context, handling offline hours, and monitoring the system over time.

Table of Contents

• Why Handover Matters
• Step 1: Design Your Escalation Triggers
• Step 2: Configure the Human Handover Node
• Step 3: Set Up Agent Routing
• Step 4: Configure Agent Notifications
• Step 5: Handle Offline Hours
• Step 6: Test the Handover Flow
• Step 7: Monitor and Optimize
• Best Practices
• Frequently Asked Questions
• Next Steps

Why Handover Matters

Here's a stat that should matter to every business running a chatbot: 73% of customers say they'll switch to a competitor after a single bad support experience. A chatbot that can't escalate — that loops endlessly, gives wrong answers, or forces the customer to call separately — creates exactly that kind of experience.

Conversely, a chatbot that recognizes its limitations and smoothly connects the customer with a human agent builds trust. The customer thinks: "This bot knew when it couldn't help me, and it made sure someone who could was brought in immediately." That's a positive experience, even though the bot itself couldn't resolve the issue.

The goal isn't to eliminate human agents. It's to make sure they spend their time on the conversations that genuinely need them, while the AI handles the rest.

Step 1: Design Your Escalation Triggers

Before touching the workflow builder, decide when your chatbot should hand off to a human. There are four main trigger categories:

Low Confidence Responses

When the AI isn't sure about its answer, it should escalate rather than guess. In LoopReply, the AI Response node provides a confidence score. You can use a Condition node after the AI Response to check this score:

• Confidence above 80% → Deliver the AI response to the visitor.
• Confidence below 80% → Route to Human Handover.

This is the most important escalation trigger. It catches the cases where the bot would otherwise give a bad answer.

Negative Sentiment Detection

When a visitor is frustrated, angry, or upset, continuing with automated responses can make things worse. Use the Intent Router or a Condition node to detect negative sentiment:

• Phrases like "this is ridiculous", "I want to talk to a person", "you're not helping"
• Repeated questions (the visitor asking the same thing multiple times, suggesting the bot isn't helping)
• Escalating language patterns

Explicit Request for a Human

Sometimes visitors simply ask to speak with a person. Respect that immediately. Common triggers:

• "Can I talk to a human?"
• "I need to speak to someone"
• "Transfer me to support"
• "Connect me to an agent"

Use the Intent Router to catch these phrases and route directly to handover without delay.

Topic-Based Escalation

Certain topics should always go to humans regardless of AI confidence:

• Billing disputes and refunds — Financial matters need human judgment.
• Account security issues — Password resets, unauthorized access, account lockouts.
• Legal or compliance questions — The AI should never give legal advice.
• Complex technical issues — Multi-step debugging that requires back-and-forth with a specialist.

Configure these as intents in your Intent Router, and wire them directly to the Human Handover node.

Step 2: Configure the Human Handover Node

In the visual workflow builder, drag a Human Handover node onto the canvas and connect it to your escalation triggers.

Click the node to open its configuration:

Handover Message

This is the message the visitor sees when the handover happens. Get this right — it sets expectations:

Good examples:

• "I want to make sure you get the best help possible. I'm connecting you with a team member now. They'll have the full context of our conversation."
• "Let me bring in one of our support specialists for this. Someone will be with you in just a moment."

Bad examples:

• "Transferring..." (too cold, no context)
• "I can't help with that." (negative, unhelpful)
• "Please hold." (creates anxiety without information)

Context Passing

Enable full conversation context so the human agent sees everything that happened before the handover. This includes:

• The visitor's original question
• All messages exchanged with the bot
• Any variables collected (name, email, order number)
• The knowledge base chunks the AI referenced
• The AI's confidence score (so the agent knows why it was escalated)

Nothing frustrates a customer more than repeating themselves after a handover. Context preservation eliminates that.

Priority Level

Set the priority based on the escalation trigger:

• High priority for negative sentiment and explicit escalation requests
• Medium priority for low confidence responses
• Low priority for general topic-based routing

Priority determines the order in which conversations appear in your agents' queue.

Step 3: Set Up Agent Routing

When a conversation is handed over, it needs to reach the right person. LoopReply's human handover system supports several routing strategies:

Round Robin

Conversations are distributed evenly across all available agents. This is the simplest strategy and works well for general-purpose support teams.

Best for: Small teams (2-5 agents) where everyone handles all types of questions.

Skill-Based Routing

Route conversations to agents based on the topic or required expertise:

• Billing questions → Finance team
• Technical issues → Engineering support
• Sales inquiries → Sales team

To set this up:

1. In your LoopReply workspace settings, create agent groups (e.g., "Billing," "Technical," "Sales").
2. Assign team members to appropriate groups.
3. In the Human Handover node, select the target group based on the escalation trigger.

For example, if the handover was triggered by a billing intent, route to the "Billing" group. If it was triggered by low AI confidence on a technical question, route to "Technical."

Availability-Based

Route conversations only to agents who are currently online and active. If no agents are available, the system follows your offline hours configuration (covered in Step 5).

This prevents conversations from sitting in a queue with no one to pick them up.

Step 4: Configure Agent Notifications

Fast response times after handover are critical. A visitor who was just told "someone will be with you shortly" and then waits 20 minutes has a worse experience than if they'd never been promised a human at all.

Configure notifications across multiple channels to ensure agents respond quickly:

In-App Notifications

Agents using the LoopReply shared inbox see new handover conversations appear in real time, highlighted with the escalation priority. This is the primary notification channel.

Push Notifications

Enable browser push notifications so agents are alerted even if they're not actively looking at the LoopReply dashboard. Push notifications include:

• Visitor name (if collected)
• Escalation reason
• Priority level
• A direct link to the conversation

Email Notifications

Configure email alerts as a backup channel. Useful for teams that don't have LoopReply open all day. Set up email notifications in Workspace Settings → Notifications.

Custom Webhooks

For teams using Slack, Microsoft Teams, or other tools as their primary communication hub, configure a webhook that posts to a dedicated channel when a handover occurs. Use the API Call node in your workflow to send a notification to your preferred tool before the Human Handover node.

Tip: Layer your notifications. Use in-app as primary, push as secondary, and email as a safety net. The goal is zero missed handovers.

Step 5: Handle Offline Hours

Your agents aren't online 24/7, but your chatbot is. You need a plan for handovers that happen outside business hours.

Option A: Queue with Expectations

Let the handover happen but set clear expectations:

"Our team is currently offline (we're available Monday-Friday, 9am-6pm EST). I've saved your conversation and a team member will follow up as soon as they're back. If you'd like, leave your email and we'll reach out directly."

Add a Collect Input node after the offline message to capture their email for follow-up.

Option B: Extended AI Support

Instead of handing over during offline hours, keep the AI active with a modified prompt that acknowledges its limitations:

"Our support team is offline right now, but I'll do my best to help. If I can't fully resolve your question, I'll make sure the team has all the context when they return."

This keeps the visitor engaged and often resolves the issue anyway. If the AI still can't help, it collects contact information for follow-up.

Option C: Callback Scheduling

Add a scheduling flow that lets the visitor book a time to speak with an agent. Use Collect Input nodes to gather their preferred time and contact method, then use a Send Email node to notify the scheduling team.

To implement offline hours logic, use a Condition node that checks the current time against your business hours. Route to the appropriate path based on whether agents are currently available.

Step 6: Test the Handover Flow

Testing handover requires two perspectives — the visitor's and the agent's. Here's how to test both:

Test as a Visitor

1. Open the test chat in the workflow builder (or test on your live site).
2. Trigger each escalation path:
   • Ask a question the bot can't answer (low confidence trigger).
   • Express frustration: "This isn't helping at all" (negative sentiment trigger).
   • Ask directly: "Can I talk to a human?" (explicit request trigger).
   • Ask about billing or account security (topic-based trigger).
3. Verify:
   • The handover message appears correctly.
   • The transition feels smooth, not jarring.
   • You're not asked to repeat information.

Test as an Agent

1. Open the LoopReply shared inbox in another browser or device.
2. Verify:
   • The handover conversation appears in your queue with the correct priority.
   • Notifications arrive promptly (in-app, push, email).
   • The full conversation context is visible — you can see everything the bot discussed.
   • You can respond immediately and the visitor receives your message in real time.
   • After the agent responds, the conversation continues as a live chat.

Test Offline Hours

1. Set your business hours to exclude the current time.
2. Trigger a handover and verify the offline flow activates.
3. Check that the visitor receives the correct offline message and email collection works.

Step 7: Monitor and Optimize

Once handover is live, track these metrics in your LoopReply analytics dashboard:

Key Metrics

• Handover rate — What percentage of conversations get escalated? If it's above 30%, your AI or knowledge base may need improvement. Below 5% might mean you're not escalating enough (visitors might be getting bad AI answers instead).
• Time to first agent response — How quickly do agents respond after handover? Target under 2 minutes during business hours.
• Resolution rate post-handover — Are agents actually resolving the issues? Low resolution rates might indicate routing problems.
• Visitor satisfaction post-handover — Track CSAT scores specifically for conversations that included a handover.

Optimization Cycle

1. Review escalation reasons weekly. Look at why conversations are being handed over. If the same topic keeps appearing, add better content to your knowledge base to let the AI handle it.
2. Adjust confidence thresholds. If too many conversations are escalating unnecessarily, raise the threshold. If visitors are getting bad AI answers, lower it.
3. Refine routing rules. If certain agent groups are overwhelmed while others are idle, rebalance your routing configuration.
4. Update the offline flow. Based on feedback, adjust your offline message, add more options, or refine the callback scheduling process.

Best Practices

When to Escalate

• Always escalate when the customer explicitly asks for a human. Don't make them ask twice.
• Always escalate billing disputes, security concerns, and legal questions.
• Escalate quickly when sentiment turns negative. The longer a frustrated customer talks to a bot, the angrier they get.
• Don't escalate too early for routine questions. Let the AI try first — you'll be surprised how often it resolves the issue.

What Context to Pass

• Full conversation transcript (always)
• Collected variables (name, email, order number, etc.)
• Knowledge base chunks the AI referenced
• AI confidence scores
• The escalation trigger reason (so the agent knows why the bot escalated)
• Visitor's device and browser info (for technical support)

How to Handle the Transition

• Acknowledge the transition. Don't silently switch from bot to human — the visitor should know a real person is now helping.
• Have the agent introduce themselves. A quick "Hi Sarah, I'm David from the support team. I can see you've been asking about..." builds rapport instantly.
• Don't make the visitor repeat themselves. This is the number one complaint about handover experiences. The full context should already be in front of the agent.
• Keep it warm. The handover message should feel helpful, not like the bot is giving up. Frame it as bringing in expertise, not admitting failure.

How to Handle No Available Agents

• Never leave the visitor hanging with no response and no explanation.
• Provide a clear estimated wait time if agents are busy.
• Offer alternatives: email follow-up, callback scheduling, self-service resources.
• If wait times exceed 5 minutes, send periodic updates: "You're next in the queue. Estimated wait: 2 minutes."

Frequently Asked Questions

What happens if no agents are online when a handover triggers?

The system follows your offline hours configuration. You can show a custom message explaining when agents will be available, collect the visitor's email for follow-up, or keep the AI active with an adjusted prompt. The conversation is preserved in the queue and agents see it when they come online.

Can I route different types of issues to different teams?

Yes. Using skill-based routing, you can create agent groups (Billing, Technical, Sales, etc.) and configure the Human Handover node to target specific groups based on the escalation trigger. Conversations about billing go to the billing team, technical issues go to engineering support, and so on.

Does the visitor know they're talking to a bot vs. a human?

By default, yes. The handover message clearly indicates a human is joining the conversation. Transparency is important — visitors who discover they've been talking to a bot without knowing tend to lose trust. We recommend being upfront about the bot/human distinction. For more on this topic, see our AI chatbot vs. live chat comparison.

How fast should agents respond after handover?

We recommend under 2 minutes during business hours. LoopReply's real-time notifications (via Pusher/WebSocket) ensure agents are alerted immediately. If response times consistently exceed 5 minutes, consider adding more agents, adjusting your routing, or reducing unnecessary escalations by improving the AI's knowledge base.

Can the bot re-engage after the human resolves the issue?

Yes. When the human agent closes the conversation or marks it as resolved, the bot can resume handling the conversation if the visitor sends another message. This is configured in your workflow — add a branch after the handover that loops back to the beginning of your flow.

How do I measure whether my handover setup is working?

Track four metrics: handover rate (target 10-20%), time to first agent response (target under 2 minutes), post-handover resolution rate (target above 90%), and post-handover CSAT score (target above 4.0/5). These are available in the LoopReply analytics dashboard. If any metric is off-target, the optimization section above explains how to adjust.

Next Steps

You now have a complete handover system. Here's how to build on it:

1. Improve your knowledge base — The fewer unnecessary escalations, the better. Train your chatbot on more data to reduce handover volume. Follow our guide to training your chatbot on custom data.
2. Build your complete workflow — If you haven't built a full conversation flow yet, see our no-code chatbot building tutorial.
3. Add lead capture — Use your chatbot to qualify leads before they ever talk to sales. See our lead qualification chatbot tutorial.
4. Install on your website — Get everything live with our chatbot installation guide.
5. Set a review cadence — Review handover conversations weekly. Identify patterns, update your knowledge base, and refine escalation thresholds based on real data.

The best handover experiences are invisible — the customer barely notices the transition because it's so smooth. That's the goal.