Business Objective
Our client is a leading financial institution, offering banking, wealth management, and brokerage services to its customers.
The client has deployed a chatbot on Google Cloud™. The intelligent assistant enables their customers to perform trading and banking activities like getting a quote, getting market updates, placing a trade, transferring money, investment FAQs, etc.
The objective was to build a scalable engine within the chatbot to detect customer’s intent and offer appropriate solutions, thereby increasing chatbot accuracy.
Challenges
- Difficult to tag and identify good quality chat instances that can be used as bot training data
- Each user transcript needs to be manually reviewed to check for classification accuracy
- The off-the-shelf platform does not provide any user adoption metrics or agent performance metrics. The data required to track these metrics are scattered across different sources
Solution Methodology
- Created models to identify the set of users who have a higher chance of using the intelligent assistant
- Used NLP and ML algorithms on the vectorized user transcripts to create an initial prediction of intent along with their confidence score and entropy (uncertainty)
- Created sampling schemes based on entropy to create an entropy sample to manually review and for classification of true negatives and false positives
- Iteratively enhanced the models on Google Cloud™ Dialogflow™ API with enriched data to create a self-learning feedback loop
Business Impact
- Increased user adoption and engagement with chatbot by increasing first contact resolution
- Better routing of engaged clients for targeted upsell opportunity
