nib Health Insurance's Nibby AI Assistant Handles 4 Million Queries and Delivers $22M in Savings with 60% Automation Rate
Australia's nib Health Insurance deployed Nibby — now powered by Anthropic's Claude 3.5 Sonnet on AWS Bedrock — to handle routine member inquiries, achieving a 60% automation rate, a 15% reduction in live phone calls, and over $22 million in cumulative savings since the 2021 launch.
Background
Private health insurers face high volumes of routine member inquiries — policy questions, claims status, provider lookups, payment schedules — that are well-suited to automation but have historically required live agent time. nib recognized an opportunity to use AI to handle this first-line volume, freeing human agents for complex cases and hardship situations. Nibby launched in 2017 as a pioneering move in the Australian health insurance market.
What Was Implemented
- Nibby: digital AI assistant embedded in nib's member portal, mobile app, and voice channel (after-hours)
- First iteration: enterprise cloud-based NLU platform
- Second iteration: Rasa open-source conversational AI framework (migrated ~2019–2020) for greater ML model customization and predictable update cycles
- Current iteration: Anthropic Claude 3.5 Sonnet on Amazon Bedrock for generative AI–enhanced understanding
- Automated member identification (international visitors, students, Australian residents, healthcare providers) for intelligent routing
- First-line support for high-volume routine inquiries; human escalation for complex cases
Results
Since the 2021 deployment milestone: - Over 4 million member queries handled cumulatively - 60% automation rate (interactions resolved without human involvement) - 15% reduction in phone calls requiring live agents - Over $22 million in cumulative savings Note: The book states "more than 4 million queries annually" — the primary sources confirm 4 million cumulatively since launch (not annually). This distinction is flagged as an unverified book claim.
Lessons
- Progressive platform evolution (cloud NLU → open-source Rasa → Claude/Bedrock) is a viable long-term strategy; each migration extended capabilities while protecting existing conversation designs
- Automated intelligent routing at the first interaction — identifying member type and inquiry category — meaningfully reduces handle time and misdirects
- In regulated industries (health insurance), generative AI adoption requires legal and compliance alignment before deploying cloud-hosted LLMs against member data
- The ROI case for health insurance chatbots is clearest when measured cumulatively against staffing headcount avoided as membership scales