Suqian Tax Bureau Builds 20 In-house AI Applications Around Data, Task and Risk Governance

Suqian Tax Bureau in Jiangsu has spent the past few years working out how to put its data to real use. First it fixed where the data comes from: the city government issued rules spelling out information sharing with market regulation, medical insurance and power utilities, replacing one-off requests with a standing list-based supply. Incoming data is then cleaned under a rule that whoever generates a record owns it and whoever uses it validates it, with more than 25,000 base records checked and corrected and every flagged item fixed. Practical applications followed. Farm-produce purchase invoices are among the hardest to verify, so the bureau uses IoT plus video analytics to link plot capacity to invoice ceilings, deriving three indicators - capacity deviation, input-output ratio and concentration of out-of-area invoices - and grading taxpayers green, yellow or red. Reducing the load on grassroots staff is the other priority: the city bureau first screens out low-value tasks with data products and AI tools, batches what can be batched, and issues standard operating guidance only for work that genuinely needs front-line handling. The bureau also builds its own tools, organising project-based expert teams to develop 20 in-house AI applications across nine fields including tax-source management, audit and taxpayer services, with the audit application covering case selection, inspection, review and enforcement. Risk control follows the same logic: responsibilities are split across city, county and sub-bureau levels, eight key business areas and 47 risk points are assigned to specific posts, and key tax sources are monitored end to end from forecasting through collection, refunds and arrears.