The customer is a remote software provider offering a suite of various services for its customers, in a SaaS model. As part of its ongoing automation processes, the company had launched a customer service chatbot. Chatbots are some of the most frequently found integrations of AI (particularly natural language processing), with Customer Service being one of the key adoption domains. Obvious benefits of using chatbots are: the ability to provide customer service 24/7 and lowering costs. A great customer service however, similar to that provided by a human agent, is still difficult to achieve. Our customer wanted to refine its customer service chatbot, by providing better, more relevant help articles to customers queries. In order to achieve this, the bot had to correctly determine what the user intent is, and select the most relevant article based on what it thought the user intent was.
The existing user funnel was as follows:
1) A user types in a query to the bot
2) The bot determines what is the use intent is
3) The bot sends the user an article based on what it thinks the intent is.
To ensure the highest level of relevance, the goal is to match intents to queries based on the degree of similarity between the solution to the intent and the solution to the query. That is, how relevant is the article that matches the intent to the person entering the query. It's not how similar the statements are, but how relevant is the article attached to the intent.
The customer reached out to super.AI to help with intent clustering, which would allow them to improve the chatbot article recommendation algorithm.
Based on the queries provided by the customer, we clustered each article and provided an exclusive choice output Not helpful, Somewhat Helpful, Very Helpful
Thanks to the intent clustering we provided, the customer was able to increase the relevance of its chatbot, therefore improving its overall customer service and reducing dependency from human agents.