Generative AI Member's Bot with LLM Prompting for the Melbourne Cricket Club (MCC)

Project: Melbourne Cricket Club (MCC) Generative AI 'Autopilot' Member Services Chatbot

Client: Melbourne Cricket Club (MCC)

The Melbourne Cricket Club (MCC) is one of the oldest and largest sports clubs in Australia, renowned for its rich history and prestigious membership. The club's members and audience include avid sports enthusiasts, primarily focused on cricket and AFL.

Problem:

The existing NLU bot was too static and couldn't provide real-time information on sporting games and events.

Solution:

I developed a new Generative AI chatbot that utilised the existing knowledge base of the previous bot while leveraging a large language model (LLM) to deliver more personalised and contextual answers. This included providing information about upcoming AFL matches and specific details on current events.

What I Did:

Implemented Custom Code: Added custom code to give the bot context around the current date and time. This enabled the bot to understand and respond accurately to queries such as "Who's playing this Saturday?"

Extensive Prompt Engineering: Fine-tuned the prompts to give the bot a friendly and engaging personality. This allowed it to discuss sporting events and football/cricket trivia while ensuring it did not say anything controversial or inappropriate.

Enhanced Conversational Ability: The prompt engineering also allowed the bot to blend answers within the knowledge base and answer multiple intents accurately and conversationally. For example, it could respond to a query like, "I'm coming to the game on Saturday, what's the free code for the Longroom and where's good to eat?"

Custom Fallback Mechanism: Designed a custom fallback that allowed MCC members to easily speak to a human agent if needed.

Results:

Customer Satisfaction (CSAT) increased by 40%

Containment increased by 40%

True Containment: 68.33%

83.33% of conversations resulted in the bot providing a correct business answer

95.00% of conversations did not see users asking for a human agent

This project not only enhanced the bot's ability to provide real-time, accurate information but also improved user engagement and satisfaction significantly.