Putuo District launches the city's first "Smart Window Assistant," cutting preliminary review of work injury determination materials from 20 minutes to 1 minute

Every month, Li Shijun, an HR specialist at Shanghai Shangjia Logistics Co., Ltd., has to make several trips to the Putuo District Government Service Center to handle work injury declarations for employees in a batch. In the past, bringing a stack of materials to the window, where staff manually checked them one by one and looked for errors and omissions, took at least an hour. "It's much faster now. What's saved is not only the time companies spend running around, but also the time for injured workers to enter the recognition process promptly," Li Shijun said. After the materials are scanned, the system can indicate within a minute what is missing and where the errors are, and they can be corrected on the spot. The change stems from the "Smart Window Assistant" that Putuo District launched ahead of the rest of the city. Before that, the "first gate" of receiving and pre-reviewing work injury recognition applications relied entirely on manual comparison—application materials often numbered more than ten types, involving accidents that occurred within the workplace, while out on work duties, or during the commute, and window staff had to hand the materials to the back office to verify each document's license information, filling standards, and logical consistency one by one. Once multiple windows were receiving applications at the same time, complex materials tended to pile up, and the waiting time for applicants grew accordingly. The "Smart Window Assistant" targets precisely the speedup of this upfront step. After materials are scanned and automatically categorized, the system uses large-model semantic recognition to automatically capture key information such as ID cards, business licenses, medical records, and labor contracts, generate a standard application form with one click, complete the formal review within minutes, and provide instant feedback on missing items. Three months after launch, window staff at the Putuo District Government Service Center are already proficient in using it: after scanning paper materials, a plug-in on the side of the screen automatically captures information such as ID cards, business licenses, medical records, and labor contracts, and pre-fills the work injury recognition application form with one click; multiple attachments are sorted into categories such as the application form, proof of labor relations, and medical records; pre-review results then pop up, with missing and erroneous items reported on the spot, and staff can receive and process applications without switching interfaces. Window staff member Shen Jun said: "The system helps determine whether an application can be accepted and whether missing items can be supplemented on the spot, so people don't have to make repeated trips to deliver materials." As of the end of October, Putuo has focused on 10 high-frequency business types in the work injury field, and 5 of them are already online. Since the pilot began, 1,176 cases have been accepted, of which 901 were handled through the Smart Window Assistant, accounting for 76.6%; material pre-review has been reduced from more than 20 minutes to within 1 minute. A single set of about 30 pages of materials is categorized in an average of 30 seconds and recognized in 40 seconds. Material recognition accuracy has risen from 69% at the start of the pilot to over 90%, field pre-filling coverage has reached 70%, and a total of 401 application forms have been automatically generated. The system has detected more than 1,000 problems in total, and 95% were corrected on the spot; the one-time pass rate for materials has risen from 67% to 89%, and the return rate for incomplete materials has dropped from 10% to 3%. Taking new work injury recognition applications as an example, the average daily number of cases handled per window under the "only one visit" model has increased from 3 to 8; during peak hours, the average waiting time for the public has been cut from 15 minutes to 5 minutes. Putuo District adheres to "AI upfront screening + human fallback responsibility," retaining a manual acceptance channel for complex situations that are difficult for the model to recognize; window staff have entered more than 900 real cases in total, the system has been iterated 20 times, a biweekly review mechanism has been established, and 50 optimizations have been completed. The accumulated data on material missing frequency and common filling errors has formed a problem list of 24 high-frequency missing materials and 32 high-frequency incorrectly filled fields, and the knowledge Q&A database contains 408 entries. This accumulated data not only enables advance reminders to applicants, but also forces the reengineering of the approval process and provides a more objective basis for formulating policy standards.