TRUSTBANK Uses AI Agents to Personalize Furusato Nozei Gifts
TRUSTBANK partnered with Recursive to build Choice AI—a conversational AI agent system powered by OpenAI models—to significantly reduce decision friction in Furusato Nozei gift selection through personalized, interactive recommendations.
Project Context and Objective
Japan’s Furusato Nozei (Hometown Tax) program allows taxpayers to donate to local governments and receive regional goods as thank-you gifts. However, the vast, geographically fragmented, and non-standardized gift catalogues often cause decision fatigue. As a financial institution offering this service, TRUSTBANK sought to improve gift discovery efficiency and user satisfaction.
Technical Implementation
- TRUSTBANK collaborated with AI engineering firm Recursive to develop Choice AI, a conversational AI agent system;
- The core inference layer leverages OpenAI models (version unspecified, e.g., GPT-4 or GPT-3.5-turbo) for natural language understanding and generation;
- Built on an AI agent architecture, Choice AI integrates user preferences, donation budget, regional affinity, and real-time gift inventory to generate explainable, interactive recommendation sequences;
- Recommendations unfold via multi-turn dialogue—supporting clarification, filtering, and comparison—not static ranked lists.
Practical Impact
This case represents a production-grade B2C financial application of LLM-powered agents, demonstrating their viability in tasks governed by complex regulatory constraints (e.g., Japanese local tax law, gift compliance, logistics timelines) and high subjectivity (e.g., emotional value of gifts, regional identity).