Choice AI: How OpenAI Helps the Japanese Pay Taxes and Get Gifts

Making choices is always stressful, especially when they directly involve your money and tax obligations. Japan has a unique and extremely popular program called Furusato Nozei. It lets citizens direct part of their taxes to the budgets of rural prefectures and receive local delicacies, artisan goods, or vacation vouchers in return. The idea is great, but in practice it has become a logistical nightmare for users. Faced with a catalog of tens of thousands of items — from Kobe marbled beef to boxes of organic vegetables from Hokkaido — the brain simply refuses to choose. TRUSTBANK decided it was time to stop torturing customers with endless filters and turned to OpenAI's technology.

Together with startup Recursive, the bank launched Choice AI, an intelligent system that acts as a personal concierge. It's not a chatbot in the usual sense that spits out canned answers. The developers built a full-fledged multi-agent system. Inside it, several specialized AI agents each handle their own part of the task. One agent analyzes the user's profile and past preferences, another filters the massive gift database in real time, and a third makes sure the recommendations fit that person's tax limits. Instead of a tedious search, the user simply has a relaxed conversation, refining what they want along the way.

Why does this matter right now? We're seeing a fundamental shift in how businesses interact with customers. Companies used to compete on who had the handier filters and the faster search. Today that's no longer enough. Choice AI shows a move toward an "intent interface" model. Users no longer need to know how the database is structured or which product categories exist. They just need to express an intent — for example, "I want something unusual for dinner for a family of four that supports farmers hit by the rains." OpenAI's models understand context well enough to connect that abstract request to specific SKUs in the system.

Recursive had good reason to bet on a multi-agent architecture. A single LLM often tends to hallucinate or oversimplify when working with large datasets. Splitting responsibilities among agents delivers much higher accuracy. While one agent gets creative and proposes options, another acts as a strict censor, checking stock availability and whether prices are current. That makes the system fit for use in banking, where any error in the numbers can cause reputational damage.

For TRUSTBANK, the project became a way to stand out from competitors that still offer users classic web storefronts. In a saturated market, the winner isn't whoever has the most products, but whoever helps customers make a decision faster and more accurately. We're watching AI gradually become that "smart layer" between people and complex bureaucratic or financial processes. Japan's tax experiment is just the first sign. Soon, agents like these will help us choose insurance, plan investments, or pick educational courses, sparing us hours of poring over the fine print.

The bottom line: The era of digging through catalogs on your own is coming to an end. Choice AI proves that multi-agent systems can turn even tedious tax payments into a personalized, enjoyable experience. Will an AI assistant become a mandatory standard for every fintech app within the next two years?