AI Basic Settings : Human takeover, Smart reply, & Troubleshooting

The settings below are related to the Artificial Intelligence app.

• From Settings > Apps, click Artificial Intelligence and activate the app..

The human takeover happens in these cases:

• When a human agent replies to a user’s message.
• When the Settings > Artificial Intelligence > Human takeover > Active option is checked. It is triggered automatically when the chatbot fails to understand a user’s message or it can be requested manually by the user explicitly asking to contact a human agent.

• When the human takeover is activated the chatbot is automatically disabled for as long as the last human agent that replied to the conversation is online. If no agents have responded to the conversation yet, the chatbot will remain active. After that, the chatbot is activated again but no default fallback messages will be sent within 10 days of human takeover if the chatbot doesn’t know the answer to the user’s question in the same conversation. You can force a Dialogflow message to be always sent by adding to the Intent the custom Payload value “force-message”: true.
• The chatbot is fully activated again if the conversation is archived or deleted.
• If you’re using Slack, no messages will be sent if the chatbot knows the answer. Once the human takeover is active all conversation’s messages are sent to Slack.
• When a human agent reply to a user the human takeover is activated automatically.
• The human takeover feature is compatible with OpenAI and will continue to function even if the Dialogflow chatbot is deactivated.
• To fully disable the chatbot on human takeover, check the option Human takeover > Disable chatbot.
• If the human takeover is already active, it will not be activated again, and the human takeover message will not be sent. The fallback message will be sent instead.
• The Settings > Artificial Intelligence > Human takeover > Fallback message feature sends an automatic fallback message when a user sends a new message while human takeover is active. It’s useful for letting users know that their messages have been received and that an agent will reply shortly. The fallback message is sent at most once every 30 minutes and only if the user’s new message is sent more than 30 minutes after the last agent reply.
• Human takeover works seamlessly across all messaging apps, like WhatsApp, Messenger, Telegram, and others. Users don’t need to click a confirmation button, any text reply confirming the request will activate the human takeover automatically.

When the option Human takeover is active the following happens:

• If enabled, the request is sent only if the message sent by the user is longer than 3 chars and contains at least two words. This optimization prevents mistaken requests.
• The conversations to which the chatbot was able to answer correctly, are marked as read and moved at the bottom of the Inbox list.

When the user confirms the human takeover, or if it is automatic, the following happens:

• The conversation is marked as unread and moved on top of the Inbox.
• Any future user messages in the same conversation sent within 10 days of the human takeover will trigger agent notifications.
• If agents email notifications are active, an email notification is sent to the agent assigned to the conversation, or, if the user’s conversation is assigned to a department, to the agents assigned to that department, otherwise it is sent to all offline agents.
• If Push notifications are active, a push notification is sent to the agent assigned to the conversation, or, if the user’s conversation is assigned to a department, to the agents assigned to that department, otherwise it is sent to all offline agents.
• If text message notifications are active, a text message is sent to the agent assigned to the conversation, or, if the user’s conversation is assigned to a department, to the agents assigned to that department, otherwise it is sent to all offline agents.
• No human takeover requests within 10 days of the human takeover will be sent in the same conversation.
• The follow-up message and offline message are activated.
• If queue settings are active, the queue or routing is activated.

• You can manually trigger the human takeover by creating a new Dialogflow Intent with the following Custom Payload response: { “human-takeover”: true }.
• The manual takeover automatically notify agents via email and leave the conversation marked as unread.
• To manually send a human take over request use the chips rich message with ID sb-human-takeover. Ex. [chips id=”sb-human-takeover” options=”Human support,Cancel” message=””].

• The human takeover can be requested manually by the user explicitly asking to contact a human agent. This setting is active by default.

You can enable the human takeover for specific user messages via Q&A by adding the human takeover action in set data and actions.

To enable human takeover on an OpenAI Assistant add the following function in the OpenAI functions area:

                                {
                                    "name": "sb-human-takeover",
                                    "description": "I want to contact a human support agent or team member. I want human support.",
                                    "parameters": {
                                        "type": "object",
                                        "properties": {},
                                        "required": []
                                    }
                                }
                            

Smart Reply suggests quick responses in real-time during a conversation. Once active, you will see the suggested replies in the conversation area, if any.

• The Smart Reply feature initially checks for suggestions from the Dialogflow chatbot, provided it is active. If there is at least one suggestion, it will promptly display the results. However, if Dialogflow is inactive, or there are Dialogflow suggestions, the OpenAI suggestions will be returned instead.
• If the language detection feature is active, the smart replies will use language detection as well.
• If the multilingual via translation feature is enabled, the smart replies will utilize multilingual translation as well.
• To read the complete text of a smart reply, hover the mouse pointer over it for a duration of 3 seconds.
• To restore the previous message, press the keyboard shortcuts CTRL + Z.

CHAT.devad.io provides powerful tools to assist you in the process of creating the chatbot.

• Activate Artificial Intelligence > Google > Dialogflow chatbot. If you have general or specific inquiries that require a definitive response, or if you want to use rich messages like buttons, the Dialogflow chatbot is the best option and it can work alongside the OpenAI chatbot. The Dialogflow and OpenAI chatbots can work together simultaneously. More details here.
• Keep improviding the chatbot from the chatbot training panel.
• Activate Artificial Intelligence > Smart Reply. More details here.
• Activate the following OpenAI settings: Chatbot, Spelling Correction, Dialogflow spelling correction, Rewrite Message Button. More details here.
• Train your OpenAI chatbot with your own content, more details here.
• Set the value of Google > Dialogflow Intent detection confidence to 0.81.
• If your website receives traffic from multiple countries, consider activating the following Google options: multilingual via translation, automatic translation, language detection.
• Change the chatbot name from Settings > Users > Bot name.

The most frequent reasons for OpenAI or Google not functioning properly are listed below. For more details about the issue, open the browser developer tools and then the console tab, send a message through the CHAT.devad.io chat, and an error should appear in the console. On Chrome you can open the console from Settings > More tools > Developer tools > Console.

• There is a human takeover.
• You can hire us to make the synchronization for you, details at the hire us page.

• Click the Settings > Artificial Intelligence > OpenAI > Troubleshoot problems button and check for any error. If there is an error, it will be related to your OpenAI account, follow the error message instructions to fix the issue.
• Make sure to check the option Settings > Artificial Intelligence > OpenAI > Chatbot.
• You reached the quote limit or you are encountering billing problems with your OpenAI account. Check it at https://platform.openai.com/account/usage.
• If Dialogflow is enabled, OpenAI may not work correctly because your Dialogflow agent does not have the Fallback Intent. Please verify its presence on the Intents page, and if it’s missing, you can create it again by following the instructions here. To quickly check if this is the issue, you can disable Dialogflow and send a message consisting of 2-3 words. Then, check if the browser console has errors.
• If the chatbot is not replying at all and you have some Google setting active, try to disable the Google settings and check if the chatbot is working.
• Make sure to read the general information.
• If the training is not working as intended, for instance, if the chatbot is not responding to questions relevant to the training data, try to delete all training data from Chatbot > Training > Informaion > Delete all training data and train the chatbot again.
• If the training is not working as intended, for instance, if the chatbot replies are cut off or partially incorrect, try asking the same question a few more times. The responses should improve over time. OpenAI is needs to learn from the user interactions.
• If your training data is from a website, and the website is multilingual, make sure the lang attribute of the <html> tag contains the correct language.
• Delete all of your OpenAI settings, leave only the OpenAI key and try again.
• Try to use the OpenAI assistant instead.
• If you are having issue with ai answers, you can use our OpenAI key by setting Settings > Artificial Intelligence > OpenAI > Sync mode to Automatic. If this resolves your problems, it means that the issue originates from your OpenAI account. You need to purchase credits to use this option.
  

Here are the top 10 common issues your end users might face with the OpenAI:

API Key Invalid or Expired
Users may encounter errors if their OpenAI API key is invalid, missing, or expired.

Rate Limiting or Quota Exceeded
Requests may be throttled or rejected if users exceed their allotted usage limits or request rate.

Latency or Slow Response Times
Users might experience delays in receiving responses from the OpenAI API due to network or server load.

Unexpected or Incorrect AI Outputs
The AI may generate inaccurate, irrelevant, or misleading content, sometimes called hallucinations. We highly Recommand you using default settings, if it failed use our Ai credit or try OpenAI Assistant API for better results

Authentication Failures
Failures in authenticating API requests due to improper key usage or configuration issues.

Usage Costs and Billing Concerns
Users may face unexpected charges or misunderstandings about billing tied to their consumption of the OpenAI service.

Integration or Compatibility Issues
Challenges integrating OpenAI API into user applications, including SDK version mismatches or unsupported platforms.

Security and Data Privacy Questions
Concerns regarding how user data and queries are handled, stored, or protected by OpenAI.

Limited Feature Access or Model Availability
Access restrictions to certain models or features based on plan limits or regional availability.

• Click the Settings > Artificial Intelligence > Google > Troubleshoot problems button and check for any error. If there is an error, it will be related to your Google account, follow the error message instructions to fix the issue.
• Make sure to check the option Settings > Artificial Intelligence > Google > Dialogflow chatbot.
• If your synchronization was not successful we suggest reviewing our documentation and repeating the synchronization steps to correct any errors. If needed, we provide integration services. Details at the hire us page.
• You selected the wrong agent location.
• Make sure are not using a mega agent.
• Make sure to read the general information.