On the evening of December 19, the CCTV Financial Channel program Economic Half Hour, in its special feature “Setting Sail for 2026 – Improving Soft Power,” highlighted Geely Auto Research Institute and Yonggang Group as benchmark cases for using Feishu AI to improve quality and efficiency.
Geely Auto Research Institute: The “Size Master” That Listens
Sun Mengwei is a quality engineer at Geely Auto Research Institute. His daily job involves measuring different parts of vehicles. But this time, he carried neither pen nor paper. After each measurement, he simply spoke the data aloud and moved on to the next checkpoint. In just 5 minutes and 10 seconds, Sun had completed over 100 data measurements on the vehicle.
The “magic tool” in Sun’s hand made this possible: “I just open the meeting function of this mini‑program and start measuring. It records and organises everything I say, and creates a table. It’s very convenient – I only need to speak, not write anything down.”
This mini‑program is an “intelligent size detection” tool developed by Geely at the end of September, built on Feishu AI. With simple phone operations, the software automatically records voice, announces measurements, and organises them into tables.
Just 90 seconds later, Chief Engineer Xu Lailai saw Sun’s measurement data on his computer. The system also provided an intelligent analysis, pointing out which data had room for optimisation and offering suggestions on how to monitor, spot‑check, and prevent issues.
Xu Lailai said: “This system is based on the experience of our most seasoned engineers. We feed their expertise into a knowledge base, and combined with AI’s logical capabilities, it can quickly give us accurate information.”
Because this “intelligent size detection” tool incorporates a large amount of engineering knowledge, the team affectionately calls it the “Size Master.” Xu said that this AI agent has greatly reduced the workload of quality engineers. Their department is part of an automotive R&D institution. Here, every small‑batch prototype vehicle must be meticulously inspected to ensure it meets design requirements. Quality engineers inspect 10‑20 vehicles per day. In the past, this took more than four hours daily.
Previously, they had to collect over 100 data points from more than 50 locations on each vehicle using two different tools, writing all measurements directly on the vehicle body. Then they would take a paper measurement sheet and transcribe the values by hand. After that, they would return to their desks and enter the values into a computer. Xu said: “The whole process was tedious. Manual transcription was prone to errors – wrong cells, missed entries, incorrect values – requiring repeated on‑site checks. Overall, it was very slow.”
Dimension matching validation is a critical step in new vehicle R&D. The old method relied heavily on experienced engineers, and analysing data was both time‑consuming and labour‑intensive. If recording errors occurred and were not caught in time, they would delay new vehicle development.
Therefore, Xu and his team had long wanted to explore new inspection methods. They considered purchasing specialised equipment or using camera‑based AI recognition, but both proved unsuitable. A dedicated camera system costs 300,000 yuan, making it impossible to purchase many units – only one person could use one at a time.
Just as Xu was grappling with the problem, they stumbled upon a source of inspiration from Feishu, their enterprise collaboration and management platform.
In daily office work, Feishu is used for meeting minutes. Its multidimensional tables can also analyse data and provide AI summaries. During casual conversations, they suddenly realised that in addition to office work, frontline size detection might also benefit from these capabilities.
However, a factory floor is noisy, with lots of background noise, and frontline engineers have different accents. How to ensure that voice recordings are transcribed into accurate data became the challenge.
Li Liyin and Chen Xinkang are members of Feishu’s customer success team, specialising in serving automotive industry clients. Since April this year, they have frequently visited Geely.
In July, after learning that the customer was experimenting with AI in size detection, Li Liyin intensified her support. She said: “Feishu started as an office software, but its application scenarios have now extended from the office to the production line.”
The noisy workshop environment required cleaning the audio data – removing extraneous parts and purifying the relevant content. Feishu staff and Geely staff worked together to design a chain of thought and continuously optimise the prompts for AI inference. After two months and 47 rounds of iterative adjustments, the intelligent size detection AI tool was finally put into use, transforming the traditional “hand‑written measurement values + manual data analysis” model into a new path of “voice‑generated measurement values + intelligent decision analysis.”
Compared to purchasing specialised equipment, the AI tool saved the company millions of yuan in investment. This success has given Geely Auto Research Institute greater confidence. In the future, they plan to collaborate with more productive service enterprises to apply artificial intelligence across all scenarios.
Song Wei, Deputy Director of the Pilot Production Center at Geely Auto Research Institute, said: “We will also draw on external scenarios – good algorithms, good smart manufacturing scenarios – to free people from heavy work and focus more of their energy on innovation.”
Li Liyin is further improving the AI’s ability to clean audio data, enabling it to better recognise foreign‑language recordings. Meanwhile, 200 kilometres away in Zhangjiagang, her colleague Zhou Jingwei is providing AI tool platforms to a steel enterprise – Yonggang Group.
Yonggang Group: The Blast Furnace Expert that Cuts Shift Handover from 30 Minutes to 5
In May this year, Yonggang Group used the Feishu aily agent to develop a “Blast Furnace Expert Assistant.” This new AI employee has reduced the shift handover time for blast furnace operators from half an hour to just 5 minutes.
Xiang Nan, a blast furnace operator at the No.1 Branch of Yonggang Group’s Ironmaking Plant, explained: “It takes 6 hours from charging the furnace to tapping hot metal. So if there is any fluctuation in furnace conditions during our shift, it must be caused by a ‘root problem’ left from the previous shift. Therefore, shift handover must be thorough – it used to take nearly 30 minutes.”
As a blast furnace operator, Xiang Nan is responsible for operating the furnace. In an interview with CCTV, he pointed out: “This new AI colleague collects massive amounts of real‑time data from the blast furnace, extracts the key data and anomaly data, and enables us to conduct efficient shift handovers.”
Blast furnace ironmaking accounts for 70% of steel production costs and energy consumption. Under the traditional working model, although the manufacturing execution system could automatically collect anomaly data from the furnace, the specific causes still had to be analysed by workers. Moreover, with a 6‑hour lag in furnace response, decisions on charging, temperature control, etc., relied entirely on experience – like opening a huge “blind box,” with decision errors exceeding 20%.
Zhang Dong, Plant Manager of the No.1 Branch of Yonggang Group’s Ironmaking Plant, said: “The operator’s actions are totally based on experience, which inevitably leads to fluctuations in furnace conditions. Last year alone, the losses exceeded 4 million yuan – more than 10,000 yuan per day on average. That truly pained us.”
More than 4 million yuan evaporating in the furnace due to operational errors – that is real, hard cash lost. At the beginning of this year, Zhou Bin, head of Yonggang’s digitalisation team, tried using large AI models to solve the pain points of blast furnace operation. They quickly deployed a local large model but found it impractical. The standalone AI they deployed was unable to receive external data. Furthermore, the staff themselves were not specialised AI developers, and actual implementation still required data development.
Urgent needs but insufficient implementation capability – this is a common challenge for many traditional enterprises undergoing digital and intelligent transformation. Just as Zhou Bin was at a loss, Zhou Jingwei, who focuses on serving the metallurgical and steel industries, provided him with a mature Feishu AI tool.
Using this Feishu AI tool to develop the blast furnace expert assistant, they could call upon mature communication tools and workflows to connect in real time with the steel plant’s manufacturing execution system. Zhou Bin and his team took just three days to develop the first version of the blast furnace expert assistant.
Zhou Bin said: “It’s like before, to shear a sheep, we had to make our own scissors. Now, we’ve been given a pair of scissors that are both fast and excellent – we can use them directly to shear the sheep.”
Now, after several iterations, the Feishu AI Blast Furnace Intelligent Analysis Expert system not only collects real‑time furnace data but also uses structured knowledge to infer trends and provide operators with subsequent operation suggestions. Operators no longer work like “blind men feeling an elephant” – their decisions are now evidence‑based, moving away from experience dependence and toward scientific decision‑making.
Moreover, the Feishu AI Blast Furnace Intelligent Analysis Expert system uses real‑time data to generate a daily furnace report and push it to the plant manager. In the past, manually collating the data would take 8 hours, and the manager would only receive a weekly report.
Zhang Dong said: “We can now detect problems in time and solve them promptly, reducing furnace condition fluctuations and avoiding small issues turning into big ones.” Since Feishu’s Blast Furnace Expert Assistant went live in May, furnace stability has improved by approximately 15%, with all furnaces now in a healthy state. By reducing unscheduled maintenance, improving the furnace utilisation coefficient, and lowering labour costs, the system saves the company an estimated 2.56 million yuan in annual expenses.
Zhou Jingwei, Feishu Enterprise Efficiency Consultant, believes the Blast Furnace Intelligent Analysis Expert system has achieved such good results thanks to the industry’s relatively high level of digitalisation: “Today’s steel enterprises have a fairly high overall digitalisation level. Production data has already been structured, so AI can better read this structured data and make trend predictions. These predictions can then guide further improvements in production efficiency.”
Industry Context
In the first three quarters of 2025, the value‑added of China’s service industry grew by 5.4% year‑on‑year. Among them, the information transmission, software and IT services, and leasing and business services sectors grew by 11.2% and 9.2% respectively. These rapidly growing modern service industries are deeply empowering a wide range of sectors through professional solutions and efficient digital tools.
The cases of Geely Auto Research Institute and Yonggang Group both originated from the “2025 Feishu AI Efficiency Pioneer National Competition” organised by Feishu. This competition is China’s first enterprise‑level practical competition focused on “AI + real business impact.” This year, 130 enterprises and more than 900 frontline “efficiency pioneers” participated, bringing 230 real‑world business scenario cases.