Client is a leading hypermarket retailer based out of Singapore.
Client receives a large number of customer inquiries for general information at their call centre. Answers to most of these inquiries are present in FAQ section but customers rarely visit this section. In order to save call centre bandwidth and provide customers easy medium to get the information, it is desired to have a conversation chatbot leveraging AI & NLP capabilities.
Customers can then pose their inquiries to the chatbot application which will use AI to respond with the right answer. Key supported customer inquiries:
- When is my last transaction?
- What can I redeem with my LinkPoints?
- What is my LinkPoints Balance?
- When will my LinkPoints expire?
- How can I redeem my LinkPoints?
- How do I apply for Plus! Visa card?
- How long does it take to receive my card?
Valiance proposed an Artificial Intelligence based Bot built on Microsoft’s LUIS Framework to process text-based requests with the ability to carry out completely natural and human-like text conversations. The chatbot workflow comprised of following elements:
- Data Preparation
Since we didn’t have any history of previous text-based conversation, our team had to spend time in gathering training datasets. This was done through extracting conversations through audio calls and in case of mining social media for conversations. Our objective was to get as much diverse conversation data as possible so as to train our algorithm. In cases, we had to artificially create training data using linguists with local dialect flavour. In addition, we also created domain specific dictionaries and intent verbs.
- Algorithm Training
- Training: After creating intents (i.e salutation, balance request, etc) we trained separate predictive algorithms each so as to provide a customer with right information.
- Algorithms prioritized amongst different responses and selected right one to return.
- Human Intervention:
If the bot is not confident in its prediction it will route customer inquiry to customer care representative. This blended human assisted AI allows brand to deliver right customer experience.
Application is in UAT currently.