Which Company's Case Study?
American Express is a financial services company that operates globally, offering credit cards and travel services. The company has been an early adopter of AI technology, utilizing machine learning for fraud detection since 2010. In recent years, it has accelerated its efforts to deepen customer relationships by leveraging generative AI to support its employees.
Challenges They Wanted to Solve
While American Express prides itself on high-quality customer service, maintaining and improving it always presents challenges. Particularly for travel plan consultations and complex inquiries, employees had to search and organize large amounts of information to provide quick and accurate responses. This process was time-consuming, leading to customer wait times and increased employee burden.
How AI Was Used
The company introduced generative AI tools to support its employees. For example, a tool called 'Travel Counselor Assist' helps travel counselors make real-time hotel and restaurant suggestions and create travel itineraries while interacting with customers. This tool leverages large language models (LLMs) to instantly generate high-quality recommendations tailored to customer requests, reducing the need for employees to manually search for information. This has allowed employees to focus on more creative and personalized suggestions.
Estimated Architecture
The following is an estimated architecture based on publicly available information and common configurations. It does not definitively describe the actual internal structure.
- ▸Input Data: Customer inquiry content (voice/text), customer's past usage history, travel-related knowledge base, real-time availability information (e.g., hotel rooms, flight seats), etc.
- ▸AI Processing / Model or Search Layer: Employs a flexible architecture that allows switching between multiple large language models (LLMs). Utilizes Natural Language Processing (NLP) technology to understand the intent of inquiries, searches and integrates internal knowledge and external information sources to generate appropriate answers and suggestions.
- ▸Output to Business Systems: Real-time display of AI-generated suggestions and summarized information on CRM and reservation system screens used by employees.
- ▸Monitoring and Governance: Implements a mechanism to monitor the quality and compliance of generative AI responses. Collects employee feedback to continuously improve models.
Implementation Effects and Key Takeaways
- ▸Confirmed Effects and Changes: 88% of travel counselors reported high satisfaction with the tool. Results such as reduced customer wait times and increased travel bookings have been reported.
- ▸Key Takeaway for Readers: The point is that AI is not provided directly to customers but is first used to support employee tasks. This allows humans to verify and adjust AI responses, ensuring service quality while improving operational efficiency.
- ▸Caution: The design philosophy of avoiding 'vendor lock-in' by not relying on a single vendor's LLM is a crucial strategy in the rapidly advancing field of generative AI.
What Japanese Companies Can Learn
The American Express case study is valuable for many Japanese companies with customer service operations. Instead of immediately launching an AI chatbot for customers, adopting an approach that first introduces AI as a support tool for call center or sales representatives is effective for accumulating AI utilization knowledge while mitigating risks. Even just by training AI with internal knowledge bases and FAQs to speed up and enhance information retrieval for representatives, the quality and speed of customer responses will significantly improve.