Intelligent AI Chatbot for Customer Service
Tags: AI chatbot, customer service, e-commerce, LLM, knowledge base
The Challenge: improving an existing chatbot for a better customer service experience
Our client, an e-commerce company specialising in selling glasses, was struggling with their existing customer service chatbot.
The chatbot was unable to solve mild-complexity problems, provided generic responses and often got stuck in loops; frustrating customers to the point where its satisfaction score was only 1.60/5 Stars.
This situation caused the customer service team to be overwhelmed, spending much of their time dealing with angry customers. The poor customer experience started reflecting on their reviews and Trustpilot's score, affecting their overall public image.
Our Solution: a customer service chatbot powered by AI
To address these challenges, we developed a custom AI-powered customer service chatbot capable of deeply understanding customers' issues and providing relevant answers.
This solution involves three key components:
- A Customised Knowledge Database for the AI: To ensure the accuracy and relevance of the information provided by the chatbot, we built a knowledge database by analysing all customer service interactions over the past 1.5 years, as well as all product and service information available on the website. The AI accesses this database exclusively, which allows it to deliver precise and relevant responses to customer inquiries.
- Ability to Understand the Customer: The new AI chatbot we built is designed not only to answer customer queries but also to engage in conversations by asking relevant questions. This approach helps the chatbot better understand the customer's needs and gather more information about the customer's specific issue, allowing it to provide tailored responses — and reducing the need for human intervention.
- Human-like Interaction: We trained the LLM linked to our chatbot to use the writing style of top-performing customer service agents and gave it a large set of guidelines to follow to sound natural and human-like. Alongside the significant increase in answer quality, this significantly reduced the number of requests to "talk to an agent" or "talk to a human".
The Results
The implementation of the AI-powered chatbot has led to consequent improvements, with higher customer satisfaction scores and a reduction of the load put on the customer service team:
Key Outcomes:
- 1 Fully-customised AI Chatbot capable of understanding customers, asking relevant questions, and providing real help
- 4.6/5 stars satisfaction score (compared to 1.6 stars previously)
- 53% reduction in transfer to Customer Service agents in under 6 months
