From 2880 minutes to 5 minutes, Changan Automobile has its frontline employees create AI tools.

A single vehicle model may have hundreds or even thousands of low-voltage wiring harnesses. Parameters such as wire diameter, pin count, current load, voltage load, and temperature must be calculated item by item, which would take a skilled engineer two days, or 2880 minutes, in the past.

Now, after importing the parameter table, you can get the results in about 5 minutes.

This tool was developed by Mao Hai, an electrical system integration and testing engineer at Changan Automobile. He compiled the wiring harness selection rules into a product requirement document, and then used AI to complete the development. AI is responsible for calculation, analysis and presentation, while engineers verify the process and make judgments.

At Hebei Chang'an Company, quality engineer Guo Ziya was solving the problem in the same way. A batch of barcode scanners could not directly input Chinese characters, forcing employees to manually enter quality issues. Different people's descriptions might differ, affecting subsequent data analysis and problem categorization. Replacing equipment would require additional investment, and developing a new system would involve waiting for a development schedule.

Guo Ziya decided to create her own tool. She entered Chang'an, Hebei in 2013, majoring in materials science, and had no software development experience. Using AI, she first created a text conversion and QR code generation tool, and then added an automatic checking function based on the byte limitations of barcode scanners. The system determines whether the QR code can be recognized, points out the content that needs adjustment, and then generates a QR code that matches the existing device.

Afterward, she continued to use AI for checking vehicle-mounted documents, automating submissions, and verifying causes and corrective measures. Since she wasn't familiar with the intelligent agent workflow, she explained the actual process step-by-step to the AI, letting it explain how the parameters were connected.

A quality engineer with 13 years of experience began directly modifying the processes he used every day.

From 2880 minutes to 5 minutes, Changan has brought AI to the front line of business: those who understand the problem best are starting to develop solutions themselves.

In the past few years, discussions about the integration of automakers and AI have mainly focused on intelligent assisted driving and smart cockpits. However, Changan Automobile is pushing Qianwen Office into the company's internal operations, integrating it into daily work such as R&D, production, procurement, operations, and service.

The case of Qianwen Office being implemented at Changan Automobile demonstrates another aspect of enterprise AI implementation.

Modeling capabilities are just the starting point. Who raises the questions, whether the business rules can be clearly explained, and how the tools created by individuals enter the organization are the first things that determine the actual effect.

01 To improve efficiency, AI should first be applied to tasks with clearly defined rules.

Mao Hai chose low-voltage wiring harnesses because of their clearly defined input, output, and computational logic. Engineers need to handle parameters such as wire diameter, pin count, current load, voltage load, and temperature, which involves a large amount of calculation, but each step follows certain rules.

Mao Haixian worked with the wiring harness engineers to organize the calculation logic, writing the inputs, outputs, and rules into the product requirements document, and then handed it over to the AI development executable tool. After the tool was put into use, a result traceability function was added, listing each step of the calculation process for easy verification by engineers.

AI performs the calculations, while engineers retain the judgment and confirmation.

Mao Hai summarized the division of labor between the two sides more directly: "Humans are responsible for judgment and guidance, while AI is responsible for calculation, analysis and presentation."

This is also a type of scenario that Changan Automobile can relatively easily implement: the work happens repeatedly, the rules can be described, the data can be structured, and the results can be verified.

Xiao Shiqiang, manager of AI application development at Changan Automobile, said: "There is no need to deliberately look for scenarios. Business pain points have always existed, and technology is just a means to solve them."

Guo Ziya's work on verifying the consistency of vehicle documentation also conforms to these characteristics. When a vehicle leaves the factory, over 100 mandatory inspection parameters in the certificate of conformity and accompanying documents need to be consistent with the national regulatory website. Previously, staff had to check each certificate, document, and website information one by one. Now, after taking photos and uploading them, the tool can extract parameters, query announcement data, and mark discrepancies; staff mainly handle abnormal results. Preliminary data shows that the verification time per vehicle has been reduced from over ten minutes to one to two minutes.

The clearer the rules, the easier it is for AI to execute stably. If processes still rely on personal experience, or if different departments have different understandings of the same field, it will be difficult for the model to fill these gaps for the enterprise.

Mao Hai also encountered this problem when conducting user VOC analysis. The team had to choose between two product solutions, with user voices scattered across car forums, social media platforms, and comment sections. Engineers first determined the keywords, classification criteria, and output format, and then AI completed data collection, sentiment recognition, opinion clustering, and table organization. What used to take one to two weeks could now be completed in about an hour after the process was debugged.

AI lowers the barrier to entry for data processing and programming. Business professionals still need to answer two questions: What data is reliable, and what results can be incorporated into product decisions?

02 Frontline staff create their own tools, shortening traditional IT workflows.

Traditional digital transformation typically involves a long chain of processes. Business departments submit requirements, product managers understand and translate them, the technical team or external vendors develop the technology, and it is only after testing and acceptance that it is handed over to employees for use.

With each additional translation, a piece of business detail may be lost.

Quality management is a typical scenario. Guo Ziya is involved in problem entry, data submission, root cause analysis, and rectification tracking every day. She also knows which steps are prone to fatigue, omissions, and data deviations. With the advent of AI, she can first create a small tool and then integrate its effective functions into the existing platform. System improvements that previously required waiting can now be directly participated in by business personnel.

This change does not eliminate the work of the IT department. Business personnel can create small tools, automation scripts, and product prototypes, but once they are officially integrated into the production system, interfaces, permissions, testing, security, and subsequent maintenance still require the responsibility of a specialized team.

Business personnel gained development capabilities, and the IT team began to assume more responsibility for platform and governance.

Gong Xuan, Senior Project Manager of Digitalization at Changan Automobile's Global Procurement Platform, has taken this change even further. Before 2025, she had no IT experience and mainly worked on cost-related tasks. By 2026, she had developed an "AI First" habit when handling tasks, and would first determine which part of the work AI could accomplish after receiving the assignment.

A usage review generated by AI showed that she had built 58 skills. She had interacted more than 240 times in two weeks, and the review estimated that it saved approximately 45 hours. This figure, estimated by the tool, cannot be directly used as a financial metric, but it reflects the depth to which AI has been integrated into her work.

Gong Xuan continuously organizes emails, meeting minutes, instant messages, and online documents into different knowledge modules. Based on this, AI updates project progress and division of responsibilities, generates daily reports, to-do lists, and meeting materials, and also helps determine the nature of system problems and suggests directions for troubleshooting.

This personal work system also requires continuous maintenance. Gong Xuan reviews the existing skills every week to determine which ones need adjustment and optimization. She describes her rhythm as: "From Monday to Friday, I let the AI do the work for me, and on weekends I discuss with the AI what areas still need improvement."

In the past, employees used software to complete a task. Now, the software is still there, but AI is beginning to become the entry point for accessing knowledge, documents, and various capabilities.

One of the roles of Qianwen Office at Changan Automobile is to enable employees to utilize these capabilities within their existing work environments. Alibaba provides the basic models, products, and platform support, while Changan Automobile maintains control over its own business processes, internal data, and application scenarios. For large enterprises, general models can be purchased, but the business context accumulated over many years still needs to be organized by themselves.

03 Frontline staff are responsible for identifying problems, while change teams are responsible for amplifying capabilities.

Saving a few hours for one employee does not mean that the entire department has shortened its processes.

Shu Junliang, Vice President and Head of Product Development at Qianwen Office, gave an example. A process originally involved 3 people, each working for 2 hours. After AI intervention, each person may only need 15 minutes, but the entire process still lasts 3 hours because the communication, waiting, and handover remain unchanged.

As individual efficiency increases, companies begin to encounter more difficult problems: how to enable AI to understand the department, project, customer, and permissions of employees; how to call upon existing CRM, ERP, and internal platforms; and how to connect work between people and between people and intelligent agents.

At Changan Automobile, these horizontal tasks are undertaken by teams such as the Transformation and Efficiency Department and the AI Application Development Department.

Business departments are closer to the problems. Horizontal teams are responsible for tool introduction, development support, training and promotion, operational data, system connectivity, and security governance. Lacking either one makes it difficult to scale applications. Delegating everything to the technical department easily leads to reverting to outdated requirement scheduling; allowing employees to experiment freely results in redundant development, uncontrolled access, and lack of maintenance.

Changan Automobile did not simply rely on administrative orders to require employees to meet usage targets. According to Xiao Shiqiang, the company first had senior executives install and use the equipment, demonstrating it firsthand to motivate all employees and make everyone truly realize that AI can solve their problems.

Backend data is also used for operations. The team identifies high-frequency users and invites them to share real-world case studies weekly; if they see low usage rates in a particular department, they go into that department to understand the reasons and provide training. Daily active users, active days, token consumption, and skill count are all monitored, but these metrics are currently mainly used to determine whether employees have actually started using the service.

Personal skills are also being transformed into organizational assets. Gong Xuan shares his professional skills with colleagues. Since different people have different computer environments and knowledge bases, recipients still need to adapt them to their own work. He estimates that other employees can quickly acquire about 70% of the skills after receiving them, with the remaining portion requiring further personal judgment and refinement.

In the past, experience mainly flowed with the mentor. Now, some work methods can flow with the skill.

Replication still comes with conditions. Just because a skill works in the purchasing center doesn't mean it can be replicated in manufacturing and sales. Fields, permissions, systems, and auditing responsibilities can all change. Organizations need to maintain consistent standards while also allowing business units to continue making modifications.

Changan Automobile has thus formed a combined top-down approach. Frontline employees start with small problems, and the company then selects which applications are worth sharing, integrating, and maintaining in the long term.

04 After the token bill increases, companies need to start calculating the complete AI bill.

Changan Automobile has not yet issued any hard targets to its departments regarding "how much cost to save or how much efficiency to improve".

Xiao Shiqiang's reasoning was straightforward. Enterprise AI is still in its exploratory phase; application effects and technological capabilities change rapidly. Prematurely converting innovation entirely into operational metrics can easily lead to false results. Forcing employees to increase usage frequency if the product experience doesn't meet expectations will also increase resistance.

At this stage, Changan Automobile is more focused on whether employees have started using the system and whether it has been successfully implemented in various scenarios. The consumption of tokens and computing power has increased rapidly as a result, which the company anticipated, and management has begun to inquire about where the costs will be borne.

Xiao Shiqiang said: "In the long run, AI still needs to be evaluated in terms of input and output."

This issue will eventually be included in the cost system.

The time saved when an employee uses AI to generate a PowerPoint presentation is relatively easy to estimate. However, once a tool is integrated into quality, R&D, and procurement processes, its value proposition becomes much more complex. It may shorten delivery cycles, reduce errors, improve product quality, or make analyses that were previously too costly to perform feasible. Simply counting the number of times a tool is called is unlikely to capture these changes.

Data security is another issue that needs to be addressed.

Changan Automobile manages data according to levels such as internally controlled, general commercial confidential, and core commercial confidential. Confidential and core commercial confidential data are mainly handled by internal AI Agents, and the company is also building privatization capabilities. External services need to clearly define data boundaries through protocols, some tasks run in a local environment, and caches must be cleared as required after calling cloud models.

Accuracy also requires accountability. Even after using AI to generate PPTs, daily reports, and data analyses, Gong Xuan still checks the results, feeds back errors to the AI, and adjusts the rules. Text-based tasks can gradually stabilize through long-term accumulation; however, even if data tasks are completed by code, fields, definitions, and outliers still need to be verified .

Changan Automobile is currently at the crossroads of two phases. The first phase addressed "whether there are enough users," getting employees to solve the first real problem; the second phase aims to answer "whether it is worthwhile to continue investing," focusing on process cycles, quality improvement, and business results in terms of daily active users, tokens, and skill numbers.

For companies preparing to promote AI, Changan Automobile's experience provides a relatively clear order.

First, find tasks with clear rules, recurring occurrences, and verifiable procedures, and involve the people who understand the problem best in development; after the scenario is tested and approved, then decide how to integrate with the system, share capabilities, divide permissions, and bear the costs.

When frontline employees begin to spontaneously use AI to create tools, management needs to take over these tools: screen for replicable scenarios, integrate them into the enterprise system, clarify permissions and responsibilities, and then use business results to evaluate the investment.

The reduction from 2880 minutes to 5 minutes represents the efficiency of a single person's tool. The real challenge for Changan Automobile is transforming a single person's tool into a capability for the entire organization, providing an excellent example of how large enterprises can truly leverage AI.

This article comes from the WeChat official account " AI Native Lab " , which continuously analyzes real-world AI implementation cases and shares enterprise AI practices and methodologies.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.