Anshin Life wanted to use customer feedback to improve operations and service quality, but manually reading and classifying up to 18,000 unstructured feedback entries each year was slow, inconsistent, and difficult to scale with traditional machine-learning approaches.
Reported impact
Anshin Life significantly reduced the time needed to review and classify feedback, standardized judgments across reviewers, and uncovered previously hidden customer trends that it can now use to improve operations and future service decisions.
Anshin Life co-created a feedback-classification and analysis workflow with IBM Client Engineering using IBM watsonx.ai as the core reasoning layer. Customer comments from calls, online channels, and surveys are ingested into a generative AI process that first classifies them against roughly 40 insurance-industry codes and then extracts company-defined keywords and clusters so teams can analyze patterns hidden inside unstructured text. The design combines watsonx.ai natural-language processing with retrieval-augmented generation so the insurer can update information and launch new products without repeated fine-tuning, which reduces maintenance effort while preserving classification quality. Business and IT teams worked together in weekly agile cycles, using an “application owner system” to move from proof of concept to production quickly. This makes the solution more than a tagging tool: it is now part of Anshin Life’s voice-of-customer operating model, helping the insurer surface trends, improve hypotheses, and expand generative AI into other departments.
Customer feedback from phone, online, and survey channels is fed into watsonx.ai, classified into industry codes, enriched with extracted keywords and clusters, and then returned to business teams as structured insight for operations, service quality, and policy planning.
“Processing time for reading and categorizing each customer feedback entry was reduced by approximately 75%, from 7-8 minutes to 2-3 minutes.”
“System prompts and RAG eliminated the need for fine tuning when launching new products and updating information.”
“The company receives up to 18,000 customer feedback entries annually”
“customer feedback from phone calls and other sources was captured by generative AI and classified into approximately 40 industry-wide codes.”
“the combination of watsonx.ai’s natural language processing technology and Retrieval Augmented Generation (RAG) technology achieved both high accuracy and flexibility.”
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
IBM Sample Aug 11, 2026 Inter Inter needed to evolve a large, API-heavy conversational banking environment into a more scalable AI operating model without breaking the integrations, governance controls, and continuity behind critical customer… 45M customers served IBM Cloud · IBM watsonx Orchestrate IBM Sample Aug 10, 2026 Farmers State Bank Farmers State Bank needed to connect fragmented website, phone, and branch interactions because customers expected digital-first convenience while the bank still needed to preserve the personal relationships central to… IBM watsonx Orchestrate · Incede.ai · Penny IBM Sample Jul 28, 2026 LEXXARI by CCG Software Solutions Claims Connection Group needed to help carriers, agents, policyholders, and repair partners interpret dense insurance policy language quickly and consistently because manual review delayed claims decisions, drove no-pay… 40% reduction in no-pay claims IBM watsonx · IBM watsonx Orchestrate 