Training and refining the intent recognition on Accenture Conversational AI has not been straightforward, as we needed extensive data for training the AI chatbot and faced a learning curve in managing diverse candidate queries during the implementation process. Initially, the training was moderately easy thanks to the structured intent and entity setup, and we created various training phrases such as check status and application status. Accenture Conversational AI can be improved due to the initial learning curve for training data, as it sometimes misclassified user queries, especially with varied phrasing, prompting a need to enhance intent recognition accuracy. We diversified training phrases for each intent, improved accuracy through NLU training, and minimized fallback rate for user queries, along with enhancing context handling for better continuity in conversations. On the user experience front, there should be clear, human-readable responses and a user-friendly conversational design to avoid confusion, especially with long and unclear responses that are not beneficial for user interactions.
