Four Transformations and Five Major Reconstructions: When AI Drives Manufacturing, "Quality" Will Be Redefined
In the industrial era, the core of quality was "conformance to specifications". In the era of AI-driven manufacturing, the core of quality will be "continuously earning user trust".
For decades, the basic understanding of "quality" in the manufacturing industry has been: whether products meet standards, whether processes are under control, and whether they are qualified before leaving the factory. Centered on this judgment, enterprises have established a complete and mature methodology system — standards, processes, inspection, auditing, correction and prevention. This system has supported the large-scale development of the manufacturing industry in the industrial era.
However, after more than 20 years of serving front-line quality management for the global manufacturing industry, I have never felt so strongly as today that this set of systems is facing unprecedented challenges. Especially when software, data, algorithms and AI are continuously integrated into the manufacturing industry, the logic behind the formation, diffusion and governance of quality issues is changing. Over the past 5+ years, as the main drafter of the standard for the software quality management system of intelligent connected vehicles and the promoter of its industry implementation, I have seen a trend more clearly: the manufacturing industry is entering an era where quality is being redefined.
The reason is not that traditional quality management is wrong, but that the objects it faces have changed. Today's products are increasingly characterized by software definition, data-driven operation, continuous update and system interaction. Quality problems are no longer just fluctuations in the production process, nor just defects detected before leaving the factory, but are increasingly manifested as systemic problems spanning the whole process of design, development, manufacturing, deployment, operation and upgrading.
This means that what AI has changed is not only the capability and efficiency of quality management, but also the boundary and connotation of quality, that is, the "definition of quality" itself. Quality is no longer just a verified result at the end of the process, but is becoming a system capability that needs to be coordinated, maintained, monitored and continuously evolved. What enterprises face is no longer just "how to produce qualified products", but also "how to continuously deliver a reliable, controllable and trustworthy system in the real environment".
This article will elaborate on what core shifts have taken place in "quality" at the moment of AI-driven manufacturing; and based on these profound changes, how enterprises should reconstruct the methodology of quality management. Moreover, enterprise managers should realize that what AI brings is not only the change of work content in the quality department, but also the change of the operation mode of the enterprise. The advantage of "quality-leading enterprises" in the future lies not in finding problems faster, but in predicting risks earlier, controlling risk diffusion more effectively, and integrating standards, processes, data, AI, organizational decision-making and supply chain collaboration into a continuously operating quality system.
The Core of "Quality" Has Shifted
If the core of quality in the industrial era is "conformance to specifications", then in the era of AI-driven manufacturing, the core of quality is shifting to "continuously earning user trust". In my opinion, this shift is reflected in at least the following four dimensions.
1. From "Defect Handling" to "System Thinking"
Traditional quality management mainly focuses on discovering and correcting problems: non-conforming batches, process deviations, unqualified parts, supplier problems, etc. In essence, this is a typical "point thinking", which means breaking down problems into several separate causes, and then solving them through detection, analysis and correction.
But in AI-driven systems, problems often no longer stem from an isolated defect. For example, a leading intelligent vehicle manufacturer once encountered sensor-related abnormalities in the mass production process. The hardware, assembly and manufacturing processes all complied with the specifications. The final traceability found that the problem came from a software OTA (Over-the-Air) update that changed the model's behavior, while the training data did not cover some edge scenarios sufficiently. This example reveals that in an AI-driven system, even if a single component, a single process and a single indicator all meet the requirements, the system may still fail in the real environment. This is not a quality problem in the traditional sense, but it directly affects safety, performance and user trust, and may even lead to serious consequences in extreme cases.
It can be seen that today's quality is no longer just a defect in single-point manufacturing, nor a single functional error, but an emerging result of the combined effect of hardware, manufacturing processes, supply chains, software, data, algorithms, deployment mechanisms and real usage scenarios. What many enterprises are facing today is no longer just the problem of "whether the product has defects", but the problem of "how the whole system will behave". Therefore, quality is transforming from a management object centered on defect handling to a management object centered on system behavior.
2. From "Factory as Boundary" to "Borderless"
In the past, quality had a very clear boundary: the factory. Products are designed, manufactured, inspected and shipped. As long as they are qualified when leaving the factory, quality is usually considered to be basically guaranteed. But in the era of AI-driven manufacturing, this boundary is disappearing rapidly.
A product that has passed all inspections may still fail in use due to software updates, cloud interactions, changes in data conditions, or unexpected combinations in real usage scenarios. Quality problems no longer stop at the moment when production is completed, but continue to extend to the operation process, user scenarios and ecosystem, spanning engineering, manufacturing, software, data, operation and supply chains. A product that is "qualified when leaving the factory" no longer means it is "reliable in use". This is not only an expansion of management scope, but also a structural change of quality objects and their management methods. Quality is no longer limited by the factory, no longer limited by the product itself, and no longer limited by the traditional quality department. It has become a cross-functional, cross-system and increasingly cross-enterprise capability. For enterprises, the real challenge is no longer just "how to make qualified products", but also "how to continue to take responsibility for quality after leaving the factory".
3. From "Point-in-Time Verification" to "Full Lifecycle Management"
In traditional manufacturing, quality is usually regarded as a judgment result at a key node, and the most representative node is the moment when the product leaves the factory. As long as the product passes the inspection and meets the specifications, quality is considered to be basically guaranteed. But in AI-driven manufacturing, this logic no longer holds.
Products are no longer static. They will continue to evolve with software updates, new data input, model retraining, cloud interactions and changes in the operating environment. Therefore, quality cannot be judged and confirmed only once when the product leaves the factory. It must run through the whole process of product design, development, manufacturing, deployment, operation, iteration and upgrading, and become a capability that is continuously managed, verified and corrected.
More importantly, full-lifecycle quality not only involves the product itself, but also a complete set of operation systems corresponding to the product lifecycle: design and development processes, manufacturing execution processes, software development and iteration, deployment management, data governance, operation monitoring and closed-loop improvement mechanisms. What enterprises need to manage is no longer just how the product is manufactured, but also how it is used and updated in the real environment, and remains reliable in the continuous evolution. In the era of AI-driven manufacturing, quality is no longer a confirmation at the time of leaving the factory, but a capability that runs through the whole product lifecycle.
4. From "Measured Qualification" to "Perceived Quality"
In the era of AI-driven manufacturing, another deeper change is taking place: the inherent characteristics of quality itself are becoming less and less completely deterministic.
In traditional manufacturing, quality is basically measurable. It can be defined, verified and controlled through indicators such as size, tolerance, durability and repeatability. As long as the product meets these indicators, it is usually regarded as "qualified".
However, the operation logic of AI systems is different from that of traditional products. It is probabilistic in essence: its behavior is shaped by data, built on training with incomplete representations of the real world, and runs in an open and dynamically changing environment. Therefore, the uncertainty of AI systems is not an exception, but an inherent attribute.
This characteristic brings a key change: even if all preset indicators of a system meet the standards, it may still fail in real scenarios. A model can achieve 99.99% accuracy, but in a rare and critical scenario, the remaining 0.01% of failure is enough to cause system failure, which may lead to extremely serious consequences. In AI systems, this kind of failure will not be understood by users as a statistical error, but will be directly perceived as a collapse of trust. This also exposes the fundamental limitation of the traditional definition of quality: measurability does not equal reliability.
Users do not judge quality by accuracy or test coverage. What they really perceive is whether the system is safe, stable, smooth, predictable and trustworthy in real usage scenarios, especially at critical moments. This is why a new quality paradigm is taking shape now — we call it "Perceived Quality".
"Perceived Quality" is not defined by enterprises themselves, nor is it a simple concept of "satisfaction". It is a subjective trust judgment formed by users through continuous accumulation in long-term real experience. Whether the system fails at critical moments, whether the screen blacks out frequently, whether functions have repeated abnormalities, whether the product is more stable or more uncontrollable after OTA updates, and whether the abnormal situation is still within the scope that users can understand and expect, all these specific experiences will eventually precipitate into users' overall perceptual judgment of quality.
Therefore, today's quality goal is no longer just "eliminating all errors" — which is almost unrealistic for AI systems. A more realistic and critical goal is to keep the system controllable, understandable and always within the safety boundary under the premise of existing uncertainty. In the final analysis, the quality goal is no longer just "whether it is qualified", but also "whether it can continuously win trust under uncertain conditions".
Five Reconstructions: A New Methodology for Quality Management
The traditional quality management model is built on a world where products are relatively stable, process boundaries are clear, and failure modes are generally determined. In this case, sampling is reasonable, standards can remain stable for a long time, and quality inspection can also serve as the last effective line of defense. But this traditional world is disappearing. As products are increasingly characterized by software definition, data-driven operation, continuous update and system interaction, quality no longer depends on whether problems can be found effectively, but more and more on whether enterprises have the capabilities of early prediction and prevention, timely containment, continuous learning and dynamic correction. When the boundary and connotation of quality are changing, the methodology of quality management must also change accordingly. What we are seeing is not a series of gradual improvements, but a structural reconstruction of how quality is monitored, understood and governed. This change can be summarized as "five reconstructions and one main line".
1. From "Sampling Inspection" to "Full Data Monitoring"
In the past, sampling was effective because the quality objects were relatively stable and repeatable, and samples could represent the whole to a large extent. But today, more and more quality problems exist in software, systems and complex interactions, with strong time sequence, strong scenario dependence and strong combinatorial nature. Sampling can hardly cover all real risks. At the same time, design, development, testing, production, delivery and operation generate massive amounts of data every day, which makes it possible to conduct full-volume, continuous and dynamic monitoring of quality. Therefore, quality management no longer only relies on samples to infer the whole, but uses data to continuously monitor the system status. Accordingly, the role of the quality department should also be upgraded from "sampling inspector" to "data monitor and risk insight explorer".
2. From "Statistical Analysis" to "AI-driven Cognition"
Traditional statistical analysis is good at dealing with stable, low-dimensional and linear manufacturing problems, but today's quality risks are increasingly characterized by cross-domain coupling: software, hardware, processes, scenarios, user behaviors and even supply chain factors all jointly affect the performance of the system. It is difficult to truly understand how risks are formed and spread only by reports, mean values and empirical induction. Quality analysis is no longer just counting results, but using AI to identify abnormal patterns, find key correlations, infer potential causality in multi-source data, and continuously form a dynamic cognition of quality risks.
3. From "Human Decision-making" to "Human-Machine Collaborative Decision-making"
In the past, quality decisions were more focused on "whether to release"; but today, they increasingly involve complex trade-offs such as software OTA, supplier disposal, recall judgment, user communication and brand risk. Facing complex decision scenarios with multi-dimensional requirements, fast response needs and serious consequences, it is more and more difficult to cope with them only by personal experience. AI can help enterprises identify anomalies, rank risks, match similar cases and provide disposal suggestions. Quality will be enhanced by AI in cognition and solution proposal, while the final decision is still made and taken responsibility for by humans. Therefore, quality management is transforming from "relying on a few experienced people" to "building a replicable, traceable and sustainable learning human-machine collaborative capability".
4. From "Static Standards" to "Dynamically Evolving Standards"
Most of the traditional quality standards are relatively fixed and phased, and can remain stable for a long time after being formulated. But in the era of software-defined products, the evolution speed of risks in real scenarios is much faster than the update pace of standards. User behaviors, network environments, version iterations, scenario changes and operation data will continuously generate new quality challenges. If quality standards remain at the level of static texts, it will be difficult for them to truly constrain and guide the system quality in reality. The standards in the future should not be a one-time compliance document, but a rule system with evolution capability that can be continuously calibrated and upgraded based on data, scenarios and feedback.
5. From "Post-event Detection" to "Real-time Control System"
In the past, enterprises usually found problems in final inspection, after-sales service or customer complaints, and then carried out correction and remedy. But in the era of intelligent manufacturing and software-defined products, when problems are found, the cost is often very high. A truly mature quality capability is not to pick out problems, but to embed quality control into the whole process of requirements, development, testing, mass production, OTA and after-sales service, so as to continuously identify, intercept and contain risks in the process. Therefore, the end point of quality management is no longer "detecting more problems", but preventing problems as much as possible through real-time monitoring, dynamic early warning and rapid closed-loop, or controlling problems before they spread.
Putting these five reconstructions together, what they point to is not only optimization, but also a paradigm shift in the methodology of quality management. There is a main line running through this shift, that is, quality management is transforming from "finding problems" to "making problems as unlikely to happen in the first place as possible". This means that the role of quality management is no longer the gatekeeper at the end of the process, but is becoming a pre-positioned, whole-process, real-time operating and continuously evolving system capability.
Quality Must Become
A Fundamental Organizational-level Capability
When quality becomes systematic, borderless, runs through the whole lifecycle and takes trust as the core, it can no longer be regarded as the responsibility of a single functional department. It must rise to a fundamental organizational-level capability — a capability that integrates standards, processes, data, AI and decision-making into a unified operating system. Therefore, quality is no longer just a management issue in enterprise operation, but is becoming the underlying capability that determines whether an enterprise can operate stably, deliver continuously and win user trust in the long run.
The real challenge of this change does not lie in whether enterprises continue to attach importance to quality, but in whether they are willing to reconstruct their own operation mode. Quality must enter the core of enterprise operation and be embedded in the whole process of design, development, manufacturing, deployment, operation and upgrading. Engineering, manufacturing, software, operation, data and supply chain partners are no longer just cooperators in quality management, but components of the joint quality formation mechanism.
Therefore, today's quality management is not only a problem of process connection, but also a problem of organizational-level system collaboration. Quality is no longer "made" or "inspected" separately in a certain link, but is continuously and jointly created within the organization and between supply chains. Enterprises must focus on the common quality goal, connect design decisions, software processes, data management, deployment rules, operation monitoring and closed-loop learning into a continuously linked and dynamically calibrated integrated operation system. In this sense, quality is the result of the overall operation mode of an organization.
. . .
In the AI era, quality is no longer a point-in-time qualification. It means that in the real environment and throughout the whole product lifecycle, the system can continuously make correct, reliable and controllable responses during long-term