Managing Conversations
Every interaction with your customers becomes a conversation in Hay. Learn how to view, manage, and take control of customer conversations.
What is a Conversation?#
A conversation is a complete interaction between a customer and your Hay agent. It includes:
- All messages exchanged
- Customer information
- Which agent handled it
- How it was resolved
- Any feedback or ratings
Viewing Conversations#
Conversations List#
Go to Conversations in the left sidebar to see all your interactions.
What you'll see:
- Conversation - Title and truncated ID
- Channel - Where the conversation came from (shown when multiple channels are active)
- Status - Current state (Active, Resolved, Escalated, Closed)
- Assigned To - Human agent, if the conversation has been taken over
- Duration - How long the conversation has been active
- Satisfaction - Customer satisfaction rating
- Updated - When the last activity happened
Searching and Filtering#
Find specific conversations quickly:
Search by:
- Conversation title
- Conversation ID
Note: Search currently matches conversation title and ID only. Searching by customer name/email or message content is not yet implemented.
Filter by:
- Status - All Status, Active, Resolved, Escalated, Closed
- Timeframe - Today, This Week, This Month, All Time (preset options)
Note: The timeframe filter is not yet functional — selecting a timeframe currently has no effect. Channel and agent filters are planned but not yet available.
Conversation Statuses#
Understanding what each status means:
| Status | What It Means | What Happens |
|---|---|---|
| Open | AI is actively handling the conversation | Agent responds automatically |
| Processing | AI is thinking and preparing a response | Wait a few seconds |
| Pending Human (shown in the UI as "Needs Attention") | Customer needs human help | Appears in your Queue |
| Human Took Over (shown in the UI as "Manual Control") | A team member is now handling it | AI stops responding |
| Resolved | Issue was solved successfully | Conversation is complete |
| Closed | Conversation ended | No further action needed |
Opening a Conversation#
Click any conversation to see the full details:
Conversation Thread#
The main area shows all messages in chronological order:
- Customer messages appear on one side
- Agent responses appear on the other
- System messages (like "Agent joined") in the middle
- Timestamps on each message
Message Features#
For each AI response, you can see:
- 👍 👎 Feedback - Rate the quality of the response
Conversation Info Panel#
On the right side, view:
- Customer details - Name, email, history
- Conversation metadata - Channel, start time, duration
- Agent info - Which agent handled this
Taking Over a Conversation#
Sometimes you need to step in and handle a conversation yourself.
When to Take Over#
- Customer specifically requests a human
- Issue is too complex for AI
- Situation requires empathy or judgment
- Agent isn't providing good answers
How to Take Over#
- Open the conversation
- Click Take Over button at the top
- Start typing your messages
- The AI will stop responding automatically
What happens when you take over:
- Status changes to "Human Took Over" (shown in the UI as "Manual Control")
- AI stops generating responses
- You can chat directly with the customer
- All your messages are logged
Exiting Takeover#
When you're done:
- Click Release Conversation to hand the conversation off. A dialog appears with two options:
- Return to AI - The agent resumes handling messages automatically
- Return to Queue - The conversation is placed back in the queue for another team member to pick up
- Click Close Conversation (a separate button) to mark the conversation as closed
Supervision Mode#
Want to review AI responses before they reach customers? Use Test Mode (also called Message Approval).
What is Test Mode?#
- AI responses are held in a queue before sending
- A team member reviews and approves each message
- Messages can be edited before approval or blocked entirely
Note: A "Supervise" toggle also appears on conversation pages, but it is not yet functional — it only changes the button label without affecting behavior. For actual review-before-send capability, use Test Mode.
How to Enable Test Mode#
- Go to Agents → select your agent
- Under Message Approval, choose Require Approval
- Alternatively, set it organization-wide in Settings → General under Test Mode Default
When enabled, bot messages appear with a "Pending" status and must be approved by a team member before being delivered to the customer.
Perfect for:
- Onboarding new agents
- High-value customers
- Sensitive topics
- Quality assurance
Test Conversations (Playground)#
Practice and test your agent without affecting real customers.
Starting a Test#
- Go to Conversations
- Click Conversation Playground
- Chat with your agent like a customer would
What's Different in Test Mode#
- ⚠️ Marked clearly as "Test"
- Messages send automatically (no waiting)
- Doesn't count in your analytics
- Perfect for trying different questions
Use playground to:
- Test new playbooks
- Try edge cases
- Train team members
- Show demos to stakeholders
Conversation Actions#
Export Conversation#
Save a complete record:
- Click Export button
- Choose format (PDF or CSV)
- Download includes all messages, timestamps, and metadata
Perfect for:
- Sharing with team members
- Record keeping
- Training materials
- Compliance documentation
Managing Customer Messages#
Messages You'll See#
Customer Types:
- 💬 Regular message
- 📎 File or image attachment
- ⭐ Feedback or rating
- 🔄 Follow-up question
Agent Types:
- 🤖 AI response
- 👤 Human agent message
- ℹ️ System notification
- ⚠️ Error or escalation notice
Handling Attachments#
When customers send files:
- Click the attachment to view
- Download if needed
- AI can read text in images and PDFs
- Use context to provide better answers
Best Practices#
Reviewing Conversations#
Make it a habit to review conversations to improve your agent:
Look for:
- ✅ Great responses (what worked well?)
- ❌ Poor responses (what needs improvement?)
- 🔁 Repeated questions (add to your docs!)
- 😟 Frustrated customers (why did it escalate?)
Responding to Feedback#
When customers rate messages with 👍 or 👎:
- Read those conversations carefully
- Update documents if information was wrong
- Adjust playbooks if flow was confusing
- Add more examples for unclear topics
Following Up#
Even after AI resolves an issue:
- Check resolved conversations periodically
- Reach out to customers who seemed unsatisfied
- Turn common questions into documentation
- Celebrate wins when AI nails it!
Common Questions#
Can I delete conversations?#
Conversations are archived rather than deleted for compliance and learning purposes. Contact your administrator if you need to remove sensitive data.
How long are conversations stored?#
Based on your organization's settings. Typically conversations are kept for analysis and improvement.
Can customers reopen closed conversations?#
Yes! If a customer messages again on the same channel, Hay can either continue the previous conversation or start a new one.
What if two agents respond to the same conversation?#
Hay prevents this with a "locking" system. Only one agent (AI or human) can respond at a time.