Conversations Overview
A conversation in oHallo is a thread of messages between a customer and your team on a single topic. Whether the customer reaches out by email, chat, or WhatsApp, the exchange is tracked as one conversation from start to finish.
Conversation statuses
Section titled “Conversation statuses”Every conversation has a status that tells you who is handling it and where it stands:
| Status | What it means |
|---|---|
| Open | The AI is actively handling this conversation. It will read incoming messages, look up relevant information, and draft replies automatically. |
| In progress | A human agent has taken over. The AI has paused, and the agent is composing replies manually. |
| Resolved | The conversation is complete. Either the AI resolved it or a human agent marked it as done. |
How the AI handles conversations
Section titled “How the AI handles conversations”When a customer sends a message, oHallo processes it through a series of steps — without any manual intervention:
- Intent classification — A planning agent reads the message and determines what the customer needs.
- Information gathering — Specialist agents query your business systems (via MCP tools) to look up orders, check inventory, retrieve account details, or perform other tasks relevant to the request.
- Reply drafting — A message agent writes a response using your knowledge base articles, policy rules, and the information gathered in the previous step.
- Validation — A validation agent checks the draft reply for accuracy and compliance with your policies. If the reply does not pass, the message agent revises it.
- Delivery — The approved reply is sent back to the customer through the same channel they used.
This entire sequence happens automatically for each inbound message.
When conversations need human attention
Section titled “When conversations need human attention”The AI handles most conversations end to end, but certain situations require a human to step in:
- Validation failures — If the AI cannot produce a reply that passes validation after multiple attempts, the conversation is flagged for human review.
- Policy triggers — Some policies explicitly require human approval. For example, a policy might state that quotes above a certain value need a manager to sign off.
- Missing information — If the AI cannot find the information it needs in your knowledge base or business systems, it may flag the conversation rather than guessing.
- Sensitive topics — Conversations involving complaints, legal matters, or other sensitive subjects can be configured to escalate automatically.
When a conversation is flagged, it appears in your inbox with an attention indicator. You can review it, take over, compose a reply, and either continue managing it or release it back to the AI.
The conversation lifecycle
Section titled “The conversation lifecycle”A typical conversation follows this path:
- Inbound message — A customer sends a message through email, chat, or WhatsApp.
- AI processing — oHallo classifies the intent, gathers information, drafts a reply, and validates it.
- Reply or escalation — If the reply passes validation, it is sent to the customer. If not, the conversation is flagged for human attention.
- Continued exchange — The customer may reply again, and the cycle repeats for each new message.
- Resolution — The conversation is marked as resolved, either automatically by the AI or manually by a human agent.
After resolution, oHallo analyses the conversation through its learning loop. If gaps in your knowledge base or policies are identified, the system proposes new entries for your review — so the AI handles similar questions better next time.