Support bot with human handoff

Upload your documentsCopy Anchor Link

1. In the main menu, open Knowledge Base and select Upload Document.
2. Enter a Document Name and a short Description, choose the File, and select Upload Document.
3. Repeat for each document.

Keep each document focused on one topic. See Knowledge base for size limits and how documents are indexed.

Create the chatbot projectCopy Anchor Link

Go to Projects, select New project, set Channel to Chatbot and create the project. The canvas opens with a Trigger node and a sample Intent node called Simple Greeting.

Open the Trigger node and, on the Platforms tab, select the channels the bot answers on, such as WhatsApp or Web. See Trigger for each channel's connection details.

Add a catch-all intent for questionsCopy Anchor Link

Every message is matched against your Intent nodes. Messages that match nothing go to the intent you mark as the catch-all, which is where the AI Agent should sit.

1. Add an Intent node and connect the Trigger to it.
2. On Basic Settings, set Intent Name to GeneralQuestion.
3. On Utterances, add the single utterance Fallback. An intent whose only sample utterance is Fallback receives every message that no other intent matches.
4. Save.

You can delete Simple Greeting, or keep it if you want a fixed reply to greetings.

Add the AI AgentCopy Anchor Link

1. Add an AI Agent node and connect GeneralQuestion to it. Leave its output unconnected: in a chatbot, the answer of an AI Agent with nothing after it is sent to the customer as the reply.
2. Select Model, choose a provider and model, pick your credential in API Key, and select Save. A low Temperature, such as 0.3, keeps answers consistent.
3. Select Customer support specialist to fill Instructions and Prompt with a starting template, then edit Instructions to name your business and what the bot may and may not help with.
4. Make sure Prompt includes the customer's message, for example Customer message: {{intent.inputTranscript}}.
5. Turn on Knowledge base, select Browse knowledge base, and tick the documents you uploaded. They appear under Selected files.
6. Save.

On each message, the agent searches only the attached files and answers from what it finds. See AI Agent for every setting.

Add an intent for "talk to a person"Copy Anchor Link

1. Add another Intent node, connect the Trigger to it, and set Intent Name to TalkToAgent.
2. On Utterances, add the ways customers ask for a person, for example speak to a human, talk to an agent, I want a real person and customer service.
3. Save.

Configure the Live Chat nodeCopy Anchor Link

1. Add a Live Chat node and connect TalkToAgent to it.
2. On the Agent Setup tab, under Data Stores, choose an Agent Group Data Store and a Human Agent Data Store. If you have none yet, select Create HITL Agent Groups System Data Store and Create HITL Human Agents System Data Store. These are shared by every Live Chat node in your organization.
3. Select Add Agent Group and create a group, for example Support.
4. Select Add Human Agent for each person who will answer chats. The name field lists your workspace members; pick one and put them in the group.
5. Under Agent Groups, choose Select Agent Group, and pick an Assignment Strategy: Round Robin, Availability Based or Skill Based.
6. On the Messages tab, write the Transfer Message the customer sees while they wait, for example Connecting you to our support team. Someone will reply here shortly. Leave Include conversation history on so the agent sees what the bot already said.
7. On the Fallback tab, choose a Fallback Action for when nobody is available, such as Add to Queue, and write the Fallback Message.
8. Save.

For how assignment, queueing and each fallback action behave, see Live chat handoff and Live Chat.

Set your hoursCopy Anchor Link

Open the project's Settings tab and select Hours.

1. Set Time zone. The AI Agent also uses it to know today's date.
2. Turn on Tell people when nobody is available, set Opening hours, and write What to say outside these hours. Add Closed dates (optional) for public holidays.

Outside these hours, a customer who asks for a person is told when the team is back instead of being promised a quick reply. The conversation is still queued. See Business hours.

Test and publishCopy Anchor Link

1. Open the simulator and ask a question your documents answer, then one they do not cover.
2. Type I want to speak to a person and check that you get the Transfer Message.
3. Publish the project. Settings under Hours and changes to the flow reach live conversations only after you publish.