News Detail

Guizhou Qixingguan Tax Bureau Builds 'Tax Brain' Studio, Using RPA and AI to Lift Grassroots Efficiency

The tax bureau of Qixingguan district in Bijie, Guizhou, has set up a 'Tax Brain' studio to cut repetitive manual work through process automation, data-driven decision-making and smarter risk control. It uses RPA robots, Python scripts and a RAG question-answering system to automate form matching, data aggregation and a live tax dashboard, shortening newly registered taxpayer processing from 15 days to two. An AI analyst named 'Tax Xiaoqi' generates corporate tax health reports automatically, and the bureau says grassroots risk screening workload fell 30% while response time dropped from 24 hours to under eight.

Industry News国家税务总局贵州省税务局Source
2026-09-16

The tax bureau of Qixingguan district in Bijie, Guizhou province, has built a studio called Qixing Tax Brain, cutting workload along three lines: process automation, data-driven decisions and smarter compliance risk control. Using RPA robots and Python, the team wrote a batch of small tools such as automatic administrator matching for spreadsheets, one-click aggregation and a self-updating tax visualization dashboard, plus a RAG question-answering system. Processing a newly registered taxpayer used to take 15 days; the new assistant program does it in two. For data work, RPA pulls real-time invoice-issuing data out of the Golden Tax III system and Python displays it up front, with one dashboard tracking 28 key indicators including filings, payments and arrears. For risk control, a Python assistant automates task matching, splitting and per-taxpayer aggregation, while a RAG model combined with tax-related data powers an AI analyst called Tax Xiaoqi that automatically produces corporate tax health reports. The bureau says grassroots risk screening workload dropped 30 percent, risk response time fell from 24 hours to under eight, processing efficiency rose 60 percent and error rates dropped 30 percent. Next steps include an industry supervision platform, risk assessment models and further upgrades to the RAG large model.