KPT Souffleuse: AI-Powered Customer Service Boost
KPT
In the fast-paced world of insurance, maintaining high-quality customer service while scaling operations can be a significant challenge. This was the exact situation faced by KPT, a rapidly growing insurance company, as they sought to maintain their excellent service standards amidst substantial growth. The question was clear: Could AI capabilities support customer agents (KUBEs) in answering client requests more quickly and with the required KPT quality?
The answer, as we discovered, was a resounding yes. But the journey to this solution was far from the "magic black box" often associated with AI tools in the public perception. Our collaboration with KPT resulted in "Souffleuse," an AI tool that not only accelerates customer service but also brings several unexpected benefits.
Scaling Service: The Insurance Industry's Everest
KPT's growth meant their customer agents were handling several dozen client requests daily, with teams managing hundreds of inquiries per region. The company needed a way to ensure service levels remained high despite this increased volume. They turned to us with a clear objective: leverage AI to support their customer agents in maintaining speed and quality in their responses.
Beyond the Magic Black Box: Our Journey to Souffleuse
We embarked on a journey of development, testing, and iteration. Our process involved several key stages:
Pre-alpha Version: We started with publicly available information, primarily from the KPT website. This phase demonstrated our ability to provide structured access to this knowledge.
Alpha Version: This stage saw the most significant improvements in quality and intelligence. We involved subject matter experts in the development process, validating everything from everyday cases to rare exceptions. We also invested substantial time in refining the underlying information architecture, our vector database, and file storage and handling.
Beta Version: Once satisfied with the quality, we released the first beta version to a broader customer agent user base. This version included several advanced features:
Certainty indication based on identified insurance products and case understanding
Step-by-step analysis revealing the thinking process
Automatic email generation with customization options
Feedback loops for continuous improvement
Clarification questions for complex cases
Smart search functionality
Comprehensive product identification across various insurance offerings
Introducing Souffleuse: Named after the theater role that whispers forgotten lines to actors, Souffleuse interprets customer requests, identifies the right insurance product, checks coverage across matching sources, and generates ready-to-use answers.
It draws from an extensive knowledge base of 1,610 sources, including website content, internal documents, and third-party information.
The tool's AI data structure ensures high consistency of answers and a strict reasoning path to the right solution. This became increasingly complex as we integrated more sources and types of information, but it proved essential in enhancing the tool's intelligence without compromising output quality.
The Souffleuse Effect
Souffleuse has delivered multiple benefits:
Efficiency Unleashed: Customer agents no longer wait for answers from other departments or search through extensive documentation. The tool handles these tasks, allowing agents to focus on personalizing and verifying responses.
Onboarding at Warp Speed: New customer agents can become operational much more quickly, providing financial benefits to KPT.
Evolution in Action: The tool's feedback mechanisms allow for ongoing refinement, ensuring it evolves with KPT's needs.
Enthusiasm Spreads: There's strong internal demand to develop the tool further and make it available to all customer agents.
The Pulse of Progress
We're focusing on three primary metrics to guide Souffleuse's continuous development:
Engagement: The number of requests customer agents generate
Quality: The ratio of improvement suggestions to the number of requests
Efficiency: The number of email responses generated by each customer agent
Lessons from the AI Frontier
Our journey with Souffleuse has reinforced two critical lessons:
The AI information architecture of the data lake is crucial for success.
Early involvement of subject matter experts significantly enhances the tool's effectiveness.
As we continue to refine and expand Souffleuse, we're excited about its potential to revolutionize customer service in the insurance industry. By combining AI capabilities with human expertise, we've created a tool that not only meets KPT's immediate needs but also paves the way for future innovations in customer service.
