Promoting Regional Development with AI-Powered Gift Suggestions

Client Background

Furusato Nozei, or “hometown tax donation,” is a Japanese system introduced 17 years ago that allows residents to donate to local governments of their choice in exchange for income or residence tax deductions. Donors can select how their contributions are used and receive thank-you gifts such as local specialties, food, or crafts. Designed to revitalize local economies and bridge the tax gap between urban and rural areas, the system also encourages greater awareness of how taxes are used while helping communities promote their unique regional appeal.

Furusato Choice, operated by TRUSTBANK, is one of Japan’s largest platforms for the Furusato Nozei system. Partnering with over 90% of municipalities nationwide, it offers more than 760,000 return gifts (as of October 2024), ranging from local specialties and daily necessities to experience-based options such as activity tickets and accommodation vouchers.

Driven by its vision of “creating autonomous and sustainable regions,” TRUSTBANK supports the independent growth of local governments and their long-term vitality. Its business goes beyond facilitating donations: it also fosters deeper connections between donors and communities, helping expand regional engagement. In addition to the Furusato Nozei platform, TRUSTBANK provides digital transformation (DX) tools for municipal operations and promotes local currencies to encourage economic circulation within communities.

Challenge

Despite its position as a leading platform in the Furusato Nozei system, Furusato Choice’s smartphone app faced several challenges that limited its potential:

• Lack of app-specific convenience and engagement: The app largely replicated the website’s content and lacked a distinct user experience, leaving donors unconvinced of its added value.
• Balancing gift selection with awareness of regional issues: While users enjoyed browsing return gifts, the app did not effectively encourage learning about local communities or the issues they face.
• Limited visibility for lesser-known regions: Donations tended to concentrate on well-known municipalities or popular gifts, reducing exposure for lesser-known areas and limiting donors’ opportunities to discover new regions.

Recursive’s Solution

Recursive first connected with TRUSTBANK when the company began exploring a redesign of its mobile app. Recursive’s mission to help build a sustainable society through advanced technologies aligned well with TRUSTBANK’s goal of driving social revitalization through regional development, making it a natural fit for collaboration.

The first challenge TRUSTBANK and Recursive addressed was the concentration of orders on specific municipalities and thank-you gifts, despite the wide range of available products. To solve this, they began developing Choice AI, an in-app chatbot that offers personalized gift recommendations. The goal was to leverage the app’s interactive features to introduce users to a broader variety of regions and products — enhancing both the value and satisfaction of the Furusato Nozei experience.

Choice AI interface integrated on Furusato Nozei mobile app

Choice AI was built using a multi-agent architecture — a more advanced form of AI agent capable of handling complex tasks. Unlike conventional systems that rely on static patterns or predefined responses, multi-agent systems feature multiple autonomous agents that interact and collaboratively or competitively solve complex problems. In Choice AI, this enables the AI to analyze conversation flow and dynamically adjust its responses.

For example, when a user starts with a greeting, the AI responds in kind and gradually steers the conversation toward relevant gift recommendations. Even if a user’s preferences are unclear at first, the AI progressively refines its understanding through targeted questions, ultimately arriving at personalized suggestions.

In addition to internal data — such as user demographics (age, location) and past usage history — external data such as seasonal trends is also integrated. Leveraging advanced algorithms, the system analyzes users’ latent needs and interests from multiple perspectives. Drawing on this detailed user profile, Choice AI’s proprietary recommendation engine selects the most suitable items from a vast database of nearly one million options.

Choice AI initiates conversation, asks clarifying questions, and suggests the most relevant gifts

Data security and privacy were also a top priority. To protect personal information, Recursive implemented measures such as masking sensitive data before storing it in the database using LLM-based techniques. The system features multi-layered defenses: encrypted communication channels and databases, strict access controls, vulnerability management systems, and detailed audit logging. Additionally, advanced moderation is in place to detect and filter out legally sensitive or irrelevant content in chat interactions.

Performance Evaluation Using LLM-Based "Simulated Users"

Before deployment, Recursive rigorously tested Choice AI using Large Language Model (LLM)-based simulated users designed to reflect realistic donor behavior, including characteristics such as age, gender, and hometown. Choice AI was evaluated across four difficulty levels:

• Very Easy: The user knows exactly what they want and provides relevant information without being asked.
• Easy–Medium: The user knows exactly what they want but only shares relevant information when specifically asked.
• Medium: The user is unsure of what they want and only provides relevant information when prompted, but can envision using the gift for a specific event.
• Hard: The user does not know what they want and only provides relevant information when prompted, but can be persuaded to choose a gift if it is well presented.

Choice AI combines three technologies to deliver a highly accurate system capable of handling complex and challenging tasks.

Choice AI achieved an impressive accuracy rate of 78%, significantly outperforming the existing recommendation system, which primarily relied on popular weekly or monthly items. These results demonstrate Choice AI’s ability to deliver smarter, more engaging, and highly personalized recommendations — greatly enhancing the donor experience.

Voice of Customer

Mr. Kenichi Kawamura, CEO of TRUSTBANK

“The collaboration with Recursive, using their cutting-edge AI technology, will be a significant driving force in our pursuit of achieving independent and sustainable communities. We will aim to build sustainable regional societies by first focusing on enhancing the donor experience on Furusato Choice, and also expanding our collaboration to support municipalities and other relevant sectors.”

Mr. Yuki Tateyama, General Manager, Product Division, Choice Business Unit, TRUSTBANK

“By combining Recursive’s cutting-edge technology and constantly evolving expertise with TRUSTBANK’s connections to over 1,700 municipalities, we see this as more than just a technology implementation — it’s an opportunity to support new approaches to regional management and to create new business models. As Japan faces the challenge of a declining population, we are eager to take on bold initiatives toward building sustainable regional communities.”

Mr. Issei Hirano, General Manager, Platform Promotion Department, Product Division, Choice Business Unit, TRUSTBANK

“Choice AI has just launched, but we are excited about its potential. As we analyze the results, we aim to evolve the experience beyond simply selecting thank-you gifts — for example, by introducing features like subscription menus or recipe suggestions. We want to create a more emotional and engaging search experience that helps donors discover the deeper appeal of each region through its products.”