Schedule Credibility under Capacity-Constrained Attention: Multi-Source Evidence from High-Mix Semiconductor Equipment Manufacturing
This paper analyzes 53,860 production snapshots from high-mix semiconductor equipment manufacturing to demonstrate that schedule credibility is primarily constrained by capacity-limited attention and administrative latency rather than structural complexity, leading to a dual-timescale Lean Six Sigma framework that prioritizes simple temporal heuristics over complex rules for effective intervention.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the world of making complex machines, a promise is only as good as the moment it is made. When a factory commits to a delivery date, that date becomes a signal that ripples outward, telling suppliers when to ship parts, engineers when to schedule tests, and customers when to prepare their own facilities. If that date changes, the entire chain of events must be rearranged. But there is a critical difference between changing a date because the work is running late, and changing a date after the deadline has already passed. The first is a necessary adjustment; the second is a broken signal. When a commitment is altered only after it has expired, it loses its power to coordinate the real world. This problem is particularly acute in industries that build highly customized, one-of-a-kind equipment, such as the massive machines used to manufacture computer chips. In these environments, the sheer variety of products and the scarcity of specialized resources make keeping a schedule incredibly difficult. The question researchers have long asked is not just how often schedules change, but whether the people managing them are actually looking at the right things at the right time to prevent those changes from becoming disasters.
A researcher set out to investigate this issue inside a semiconductor equipment factory in Vietnam. They did not rely on simple surveys or theoretical models. Instead, they reconstructed a massive digital archive of 53,860 snapshots of production status, covering 562 unique machines over a period of 111 planning dates. They combined this with independent records of labor capacity and material shortages to see what was really happening on the shop floor versus what was being reported in the planning system. Their goal was to measure "schedule credibility," which is a way of asking if the dates on the calendar are still useful to the people who need to act on them. They found that the factory was operating with a significant blind spot. Nearly half of all the target date revisions—49.2 percent—were entered into the system only after the original deadline had already passed. These late corrections were far more damaging than proactive ones. When a team fixed a date before it expired, the new date was typically only two days away from the old one. But when they fixed it after the deadline had passed, the new date jumped an average of ten days into the future. This suggests that the schedule was not just shifting; it was collapsing, leaving downstream teams with no time to react.
The researcher discovered that these delays were not random accidents scattered evenly throughout the week. Instead, the logging of these changes was heavily clustered around specific days, particularly Mondays, Saturdays, and Sundays. This pattern indicates that the updates were not happening as soon as problems were discovered on the factory floor. Instead, information was likely sitting in a queue, waiting for a scheduled administrative meeting or a batch processing cycle to be entered into the system. The data showed that while the factory might have had enough total labor hours on paper to meet its goals, it was frequently overwhelmed by the timing of the work. Multiple machines often needed the same specialized testing bay or the same expert technician at the exact same time, creating a bottleneck that aggregate numbers missed. This "concurrency overload" meant that even if the factory had enough people in total, the specific resources needed at the critical moment were unavailable, causing the schedule to break.
To understand how to fix this, the researcher tested different ways for managers to decide which machines to check first. In a factory with hundreds of machines, a manager cannot look at every single one every day. They have a limited amount of attention, or "review bandwidth." The study tested whether complex computer models that weighed many factors—like the history of a specific machine, the type of parts it needed, and past shortages—were better at predicting which machines would need a date change than a simple rule. Surprisingly, the complex models did not win. When the researcher simulated a manager who could only review the top 10 percent of machines, a simple rule based on how close the machine was to its due date performed better than the complex models. This simple rule captured 79 percent of the upcoming schedule changes. The complex models, which tried to account for deep structural issues, were actually worse at catching the immediate problems that needed attention right now. The study suggests that when attention is scarce, the most effective strategy is to focus on what is about to happen, rather than trying to predict the future based on long-term history.
The researcher concluded that the solution is not to build a smarter prediction engine, but to separate the problem into two different time scales. They proposed a dual-system approach. The first is a "fast loop" that operates daily or even hourly. This loop uses the simple, time-based rule to direct the manager's limited attention to the machines that are about to miss their deadlines. It is designed to stop the immediate bleeding. The second is a "slow loop" that operates on a weekly or monthly basis. This loop looks at the deeper, structural causes of the delays, such as recurring material shortages or specific machine families that consistently fail. This loop is responsible for fixing the root causes, like changing how parts are ordered or retraining staff, but it does not try to manage the daily crisis. By separating these two tasks, the factory can use its limited attention efficiently to protect today's promises while slowly fixing the problems that threaten tomorrow's.
The study also highlighted that the data itself can be misleading. In this factory, only about half of the machines had a complete, traceable history of their target dates. For the others, the records were missing or incomplete. The researcher used statistical methods to account for this missing information, finding that the true rate of schedule revisions was likely higher than the raw numbers suggested, but still within a predictable range. They also ruled out the idea that the delays were caused by workers intentionally hiding bad news. While it is possible that some information was withheld, the data showed that the pattern of delays was consistent with administrative bottlenecks and the natural lag of troubleshooting technical problems. The most significant finding was that the factory was not failing because of a lack of data or a lack of smart algorithms. It was failing because the way information was processed and the way attention was allocated did not match the reality of the shop floor. By aligning the management process with the actual constraints of time and attention, the factory can restore the credibility of its schedule, ensuring that when a date is promised, it is a date that the entire organization can actually act upon.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.