1. Integration
1.1. Facebook
1.1.1. Suggested Approach
1.1.1.1. OOTB
1.1.1.1.1. Connect a bot to Facebook Messenger - Bot Service
1.1.1.1.2. Limitations
1.1.1.1.3. Multi-Bot Support
1.2. WhatsApp
1.2.1. Suggested Approach
1.2.1.1. Either skip altogether or Custom via WhatsApp BussinessAPI
1.2.1.1.1. Overview - WhatsApp Business API - Documentation - Facebook for Developers
1.2.1.1.2. Not Tested
1.3. Viber
1.3.1. Suggested Approach
1.3.1.1. Open Source
1.3.1.1.1. se02035/Viber-and-the-Microsoft-Bot-Framework
1.3.1.1.2. Not Tested
1.4. Website Mobile App
1.4.1. Suggested Approach
1.4.1.1. Web Page with Bot Framework Web Chat
1.4.1.1.1. Microsoft/BotFramework-WebChat
1.4.1.2. Multi-Bot Support
1.4.1.2.1. Distinguish by BotID
1.5. Slack
1.5.1. Suggested Approach
1.5.1.1. OOTB
1.5.1.1.1. Connect a bot to Slack - Bot Service
1.5.1.1.2. Limitations
1.5.1.1.3. Multi-Bot Support
2. Multilanguage Support
2.1. LUIS
2.1.1. Language support - Azure Cognitive Services
2.1.2. Suggested Approach
2.1.2.1. Translator Text API - Auto Translate | Microsoft Azure
2.1.2.2. Translate in-Bot to English via Azure Translator Text API
2.2. QnA
2.2.1. Language support - QnA Maker - Azure Cognitive Services
2.2.2. Non-english knowledge base - QnA Maker - Azure Cognitive Services
2.2.3. Suggested Approach
2.2.3.1. Translate in-Bot to English via Azure Translator Text
2.2.3.2. Send translated Text to QnA Service. If an answer with sufficient confidence score is returned check language. If language is not English query SQL for translated answer based on custom Answer ID. If English return Answer from QnAMaker.
2.2.3.3. Keep custom AnswerID for QnA Pairs in QnAMaker
2.2.4. Alternative Approach
2.2.4.1. Check if QnA supports all required languages
2.2.4.2. Check viability of lesser supported languages
2.2.4.3. QnA Service for each language if all required languages are supported. €60-€70 per Service per month
2.3. Web App Bot
2.3.1. Suggested Approach
2.3.1.1. Handle Translation in-Bot
3. Chat Flow
3.1. Greeting
3.1.1. Greet the User by Name
3.1.2. Suggested Approach
3.1.2.1. Intercept System Event. Greet on Member Added.
3.2. Register Claim
3.2.1. Collect names (Text), insurance number (Number) , car plate number (Number), location (Text or GPS Attachment), time and date (DateTime), damages (Text or Picture Attachment) and finally asks the user to confirm (Confirm) or edit (Multiple choice between the data).
3.2.2. Suggested Approach
3.2.2.1. All User messages are translated and sent to LUIS for evaluation. If no "Register Claim" intent is detected the message fallbacks to QnA. If "Register Claim" intent is detected ask the user if he wants to register claim and if yes go into Register Claim dialog
3.2.2.2. Claim Information is collected via simple Question and Answer. Adaptive Cards were dropped due to integration incompatibility. Input is validated where possible.
3.2.2.3. Once all data is collected and validated send to relevant system
3.3. Continued Conversation
3.3.1. QnA
3.3.1.1. Suggested Approach
3.3.1.1.1. Message is sent to Custom KB. If no answers with sufficient confidence score are returned the message fallbacks to Chit-Chat
3.4. Human Takeover
3.4.1. Some integrated systems like Facebook support human takeover OOTB
3.4.2. For the others handle per integration
4. Agenda
4.1. FEB
4.1.1. v3 or v4
4.1.2. QnA Multilanguage Approach