Transforming Agent and Customer Experience with Generative AI in Health Insurance

nib · Insurance · 2025 · source Agentic: yes · promotional bias: med · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Health insurance customer and agent support

nib needed to improve customer-service scalability, contact-center efficiency, agent onboarding, and claims/document processing across a large Australian health-insurance business.

Constraints:

• nib covered approximately 2 million people across multiple brands.
• The company needed automation across chat, voice, call summaries, agent knowledge, and documents.
• Agent after-call work created productivity drag.
• Cloud migration was a prerequisite for later AI deployment.
• Responsible AI governance and continuous improvement mattered in a regulated industry.

Architecture

nib built on a decade-long AWS cloud migration to deploy multiple AI systems: Nibby on Amazon Lex for chat and voice, LLM-based call summarization, an internal knowledge GPT for agents, and intelligent document processing for claims. Amazon Connect and cloud-native infrastructure provided the contact-center foundation.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  nibby_chatbot["Nibby chatbot"] --> amazon_lex["Amazon Lex"]
  nibby_chatbot --> amazon_connect_contact_center["Amazon Connect contact center"]
  amazon_connect_contact_center --> call_summarization_service["Call summarization service"]
  call_summarization_service --> human_agents["Human agents"]
  agent_knowledge_gpt["Agent knowledge GPT"] --> human_agents
  claims_document_processing["Claims document processing"] --> human_agents
  human_agents --> nibby_chatbot

Components

Data flow

Customer interactions enter chat or voice channels. Nibby handles many requests directly; unresolved cases route to agents. Calls are summarized by LLMs to reduce after-call work, agents use an internal GPT knowledge application for answers, and claims documents flow through intelligent document processing.

Why this architecture

nib used a portfolio of AI systems rather than one assistant because the customer-service bottlenecks spanned chat deflection, voice, agent notes, knowledge access, and claims documents.

Models used: Amazon Lex · Large language models · Amazon Nova Sonic

Scale

• nib covers approximately 2 million people.
  “nib, an Australian health insurance provider covering approximately 2 million people”summary, stated
• Nibby handled more than four million customer interactions.
  “the system has handled over four million customer interactions across both channels.”summary, stated
• Nibby achieved about 60% chat deflection.
  “In the chat channel specifically, Nibby achieves approximately 60% deflection”summary, stated
• Voice deflection was about 10%.
  “The voice channel shows lower but still meaningful deflection at around 10%”summary, stated
• Nibby generated approximately $22 million in savings.
  “Nibby has generated approximately $22 million in savings”summary, stated
• Call summaries reduced after-call work by about 50%.
  “this had improved to approximately 50% reduction.”summary, stated

Alternatives considered (roads not taken)

Keep physical data centers

Rejected because: nib completed a cloud migration to enable AI prototyping and deployment.

Use chatbot automation only in chat

Rejected because: Nibby evolved across both chat and voice channels.

Patterns

tool-use-orchestration (customer service automation)

nib coordinated chatbot, voice, summarization, knowledge assistant, and claims-processing workflows across support operations.

human-in-the-loop-gate (agent support)

AI handles deflection and summarization while human agents remain in the workflow for escalations and supported tasks.

latency-optimization (contact center productivity)

Call summarization directly reduced after-call work time for agents.

Tradeoffs

Build on a cloud foundation before scaling AI.

Chose AWS cloud-native infrastructure over Continuing physical data-center operations.

Rationale: The cloud foundation enabled faster prototyping and AI deployment.

Use multiple AI initiatives rather than one system.

Chose Nibby, call summaries, knowledge GPT, and document processing over A single monolithic AI assistant.

Rationale: Different operational bottlenecks needed different AI capabilities.

Gotchas (what bit them)

Cloud migration may take longer than planned but still unlock AI speed later.

Root cause: nib's five-year data-center exit target took ten years.

Resolution: The completed cloud-native foundation supported rapid AI prototyping and deployment.

Voice deflection can lag chat deflection.

Root cause: Voice automation is harder than text chat in the reported implementation.

Resolution: Treat voice as an improvement area and explore newer speech-to-speech models.

Evaluation

nib measured AI systems through interaction volume, deflection, savings, and after-call-work reduction.

Metrics: 2 million covered people. · 4 million customer interactions handled by Nibby. · 60% chat deflection. · 10% voice deflection. · $22 million savings. · 50% after-call work reduction.

Outcomes

• nib deployed Nibby across chat and voice.
• LLM call summarization reduced after-call work.
• Agent knowledge and claims document processing broadened the AI transformation beyond customer chat.

The case reports 2 million covered people, 4 million Nibby interactions, and $22 million in savings.