How Suning’s OCR “Fire Eye” Robot Revolutionized Financial Invoice Processing

Suning’s OCR “Fire Eye” robot team spent nearly two years turning a zero‑start project into an industry‑leading solution for financial invoice verification.

Initially, external vendors could only handle a few invoice types, while Suning needed to process over 300 varieties. Internal OCR efforts focused on e‑commerce, but financial scenarios presented tougher challenges such as dirty, folded, or overlapping paper.

In 2017, product director Guo Jin‑zhi and AI R&D head Xu Li‑qun launched a pilot using OCR to automate the “sales list” (a detailed sheet attached to invoices). The goal was to automatically extract invoice codes and match them to invoices.

Because the sales‑list format varied widely, the team had to build custom models for each invoice type. After extensive data labeling—tens of thousands of samples across 300+ formats—the robot achieved about 80% single‑page recognition accuracy, dramatically reducing manual effort.

Subsequent efforts expanded to more complex tax‑invoice and train‑ticket verification, reaching 96‑98% accuracy for VAT invoices and over 90% for train tickets. By early 2018 the team introduced invoice classification to handle the massive variety, paving the way for higher‑level process automation.

Today, the OCR Fire Eye robot is being deployed across additional document types such as return orders and price files, with the aim of fully replacing manual verification and freeing finance staff for higher‑value work.