December 2023 Summaries

- As AI technology advances, customer experience (CX) leaders will be forced to innovate and adopt new strategies to stay competitive. With the power of Large Language Models (LLMs), CX teams can create sophisticated AI agents that resolve 97% of tickets while maintaining high customer satisfaction scores. In 2024, CX leaders will shift from ticket-takers to strategic collaborators, bringing actionable customer insights to their conversations with other departments. They will become an invaluable customer voice and advocate, providing valuable data-backed arguments for product decisions. To succeed, CX teams must relentlessly prioritize, focusing on one problem to solve at a time, rather than trying to tackle everything at once. By embracing this new approach, CX leaders can demonstrate the power of prioritization and become respected leaders in their organizations.

The Voiceflow team created an AI-driven support system called Tico using their extensible conversational AI platform, which transformed their customer support by deploying a consolidated knowledge base, implementing advanced query handling, integrating seamlessly with existing systems, and providing a consistently on-brand user experience. Since its implementation, Tico has drastically reduced ticket volume, resolved 90.9% of support tickets, cut human support rates by 13.6%, enhanced response time and efficiency, optimized live agent productivity, and improved customer satisfaction levels to 93.8%. The AI agent's transformative impact on support and user engagement led to significant cost savings, with a 48% reduction in direct support costs and a positive ROI within three months. The Voiceflow team plans to expand Tico's capabilities, including handling nuanced interactions, personalized account assistance, and transactions and commerce experiences, cementing its position as a leading standard for AI agents driving efficiency and innovation.

Natural language understanding has evolved significantly since the development of GPT-3, with transformer language models like large language models (LLMs) gaining prominence. However, natural language understanding units (NLUs) still offer advantages in terms of control and performance, particularly when it comes to fine-tuning for uncommon terminology, reducing inference costs and latency, and maintaining data sovereignty. On the other hand, LLMs excel at emulating NLU behavior with greater accuracy, enabling free-form behavior like open-ended entities and producing accurate results without retraining. Ultimately, the choice between NLUs and LLMs depends on specific use cases and requirements, highlighting the importance of conducting due diligence to determine the most suitable technology for each application.

Roam's team used Intercom's AI chat tool before Voiceflow, but it fell short due to lack of customizability, pricing model, and workflow mismatch. They then replaced it with Voiceflow, using a template for an AI agent and a trained knowledge base to launch their digital chat agent onto their homepage in a matter of days. The team iterated the agent through rapid testing and transcript reviews, leveraging analytics dashboards to inform product decisions and gauge where they needed to double down. Since deploying Voiceflow, Roam has saved invaluable time and effort educating prospects on their services, with a drastic decrease in inbound calls, an increase in sales calls booked asynchronously, accurate and comprehensive answers powered by AI, efficient launch and easy maintenance, and insight into customer needs.

Winston and his team used Voiceflow to automate their help center for better customer support by creating a knowledge base-powered agent that integrated with their existing interface. They curated a knowledge base consisting of website information, deployed the Knowledge Base Query API for external access, and connected it to their custom interface using Voiceflow's API step and CloudFlare. As a result, they have seen significant decreases in customer support ticket volume by 70%, faster response times, and more efficient allocation of live agent time, ultimately improving user experience and reducing the burden on their support team.

The text discusses the five levels of AI maturity, from Awareness to High-Complexity Agent, with Stage 2 being the Demo stage where companies have built a demo generative AI application but haven't deployed it. The main reasons why most companies stall out at this stage include lack of good data, inability to deploy or test with confidence, and unnecessary bureaucracy. To overcome these hurdles, companies can use tools like knowledge base, customizable widgets, and platforms that provide regression testing frameworks to improve their data and deployment processes. While deploying a Stage 3 model is a huge achievement, there's still room for growth, and companies may want to focus on deflection metrics such as serving customers without speaking to live agents rather than constantly trying to keep up with the latest AI advancements.