AI Reply from Document Flow Block
Last updated
Last updated
The "AI Reply from Document" feature enables you to create automated responses for your chat flows. By uploading a document, linking to a static URL, or using a dynamic site, the AI can extract relevant information and provide accurate answers to user inquiries. This is ideal for businesses or organizations looking to provide quick, consistent responses to common questions.
In the chat flow, you can design a flexible dialogue experience, send a variety of messages and buttons, set the tags to order information, assign conversation, and so on. You can also embed the AI knowledge into part of the chat flow, And AI will answer customers' questions in specific scenarios.
The following are the six steps of the knowledge base:
The AI reply from document will first ask questions and provide guidance. The purpose of this question is to remind customers to ask questions about a certain theme and let the customer ask you the questions from the file that you uploaded.
Choose the published knowledge, and AI will answer the customer's question about the document in the knowledge.
The drop-down menu can select the Chinese/English -Disclaimer or N/A. The statement will be displayed in the same message as the AI answer, but the position will be at the end of the message.
You can give specific guidance to the AI Chatbot’s response. For better performance, you need to provide the prompt in both English and Chinese.
Here is an example for setting system prompt in English and Chinese:
You are a helpful Customer Service Assistant. If anyone asks who you are, just explain that you are a helpful assistant who is here to answer inquiries. Use the provided context information to answer the question in detail and in a polite and professional tone.
您是一位非常幫得手的客戶服務助理。請利用所提供的情境資訊，詳盡、禮貌且專業地用繁體中文回答問題。如果有人問起你是誰，您可以禮貌地解釋你是一位專門在此回答查詢的助理。
Now, we support three end conversation conditions: only end when users click on any buttons, end after answering once, and timeout. You can only choose one of them to close your AI Reply from document conversation. Please note that if the AI Dialog is not closed, you may not trigger other flows.
When AI fails to respond or there is a mistake, you can set its follow-up process, such as sending text messages and buttons to let the customer leave.
We have added a new "Required" toggle for exiting the AI flow by triggering keywords.
If the toggle is inactive, when the client's input matches another flow's keywords, the conversation with AI will end, and the client will be directed to that flow. Please note that only specific keyword flows can be triggered; it will not trigger an "ANY KEYWORD" flow.
For example, when your client is in the AI customer service conversation, they ask about the delivery question first, and your AI CS assistant provides answers. Then the client inputs "check order," and "check order" is the keyword you set for another order checking flow, so the client's input triggers another flow, and they will be directed to the order checking flow immediately.
If the toggle is active, it means the AI flow is required. Therefore, even if the client's input matches keywords for other flows, they will remain in the AI flow.
The above is a process of asking pricing questions. The trigger condition of the process is the customer inputs the keyword "Pricing."Then, you will enter the AI Reply from Document.
The AI Reply from Document will first ask the question and guide the question. The purpose of this question is to let the customer ask questions about a certain theme. For example, "Feel Free to Ask Anything Pricing of Chatalog :)" helps prevent customers from asking questions and not being deviated to get reasonable answers.
Then, select "Published Knowledge," the knowledge should be the same theme as the guidance problem. After that, choose a Disclaimer; you should choose the same language as the file. Then, you can set your system prompts in English and Chinese. And set the end of the conversation condition. Because after the customer gets the answer, you need to set the logic of leaving this block. You can choose "Only end when users click on any buttons" to leave. If you set "Only end when users click on any buttons," you need to set a text message with a button, prompting the customer to click the button after the query is completed. Finally, set the options for exceptions.
