Use of a chatbot to reduce the average response time.
Customer advisors in this bank were not able to answer their clients questions if those questions required a detailed knowledge of the internal banking rules or its legal framework. To anwser their clients’ needs, advisors relied on experts, who were consequently particularly busy. Consquently, the average response time was up two days.
Our client asked us to find a way to reduce the response time by proving the viability of a Chatbot.
We worked on proving the viability of a chatbot designed to answer complex banking questions
We exploited an unstructured text-based data source without input from business
- Manual classification of a sample from the questions database (focus on real estate loans)
- Text preprocessing : normalization and ad hoc features
- Machine learning algorithm to classify questions into intents (questions with the same answer)
- Training a chatbot on this newly structured database, thanks to Rasa, an open-source framework
- User interface for a demonstrator
- Chabot for demonstration purposes, able to identify questions about 40 different complex topics
- Chatbot trained on a set of 2500 questions