Processing a non-standard order that mixed photos, PDF files and handwritten annotations used to eat up a full 90 minutes at a Wuhan wear-resistant alloy manufacturer. Today it takes less than 1 minute. Changjiang Daily reported on November 13 that Wuhan Fanzhou Zhongyue Alloy Materials has built an AI "order reader" on a custom-trained industrial large language model, hitting a 100 percent accuracy rate on key information extraction and lifting overall order-processing efficiency 90-fold. The project made it onto China's 2025 list of model AI application scenarios for small and medium-sized enterprises.

Fanzhou Alloy was founded in 2002 in Wuhan's Economic and Technological Development Zone, and its specialty is high-strength wear-resistant copper alloys that end up in automobiles, construction machinery, farm equipment and wind turbines. It holds national-level "little giant" status. The problem it faced was concrete rather than abstract, because customer orders arrived in messy, non-standard formats, and one misread detail at the source could throw the whole production chain off track.
Information director Hu Bo recalled that two years ago everything was done by hand. Staff had to look up, cross-check, enter and archive every line item, and the average came to 90 minutes per order. More than a hundred such orders arrived each month. Overtime for double-checking had simply become routine.
The new flow works differently. Once an order is uploaded, the AI extracts specifications, quantities and delivery deadlines within seconds, then matches them against more than 10,000 material codes and pushes the order to multiple departments for review with one click. Processing time has dropped to under a minute. That frees staff from data entry so they can focus on order risk assessment and customer needs.
After the order pipeline came engineering drawings. The team trained the model on more than 10,000 historical drawings over a year, running six rounds of tuning before the system reached stable recognition accuracy above 95 percent. It now parses parameters automatically and recommends processes, cutting tools and inspection plans. Engineers still make the final call.
The gains show up in the business. General manager Bo Yifeng said overall revenue is expected to grow 25 percent this year, and exports are up 20 to 30 percent to markets including Germany. The company is also spending more than 70 million yuan on new machining equipment, while its new plant in Caidian is over halfway built. Next on the roadmap are joints for humanoid robots, where small-batch orders have already landed, plus domestic substitution for high-speed rail bearings. For manufacturers, digital upgrading is moving beyond robot arms replacing hands, toward large models taking over the paperwork.