Truly “Advanced” Planning and Scheduling

One of our running hypotheses at ProDex is that production planning and scheduling is the most central operational challenge in the factory.

If a factory were a football team, the production planner would be the coach or playcaller, and the plant manager would be the quarterback. Enabling proper decision-making at this level is necessary to win. Given a docket of orders (the scouting report on opponents) and labor/capital resources (skilled players), one must efficiently create an operational plan that is clear, consistent, and enables maximum performance.

This process — planning and scheduling — requires our quarterback (the plant manager) not only to execute with ease, but also to incorporate live data from the field (the factory floor) and synthesize that data with extreme efficiency. Our coach (the production planner) does this at a macro-level from the sideline. This is an extremely difficult task.

Like football teams draw plays, factories draw plans on whiteboards. We've seen everything from Post-It Notes and Expo markers to Excel. Clunky but custom-built solutions that were not even the state-of-the-art in the 90s.

In the Age of AI, this is unacceptable. Even football is light years ahead in its use of digital technology for analytics and orchestration.

The status quo

Most factories have an ERP/MRP system. That system has two functions: it is a record keeper and a calculator. It is good at the former, bad at the latter.

It is a good record keeper because it was the first software in the factory to store info. It is a “system of record.” Bills of materials, due dates, orders, and lead times can all be found in this one system. Naturally, this system that has the data has attempted to do stuff with that data. That stuff is the production scheduling calculation.

Most ERPs have some function that allows its users to take its orders and due dates; use the static cycle times, BOMs, and lead time assumptions; and output a production plan. This production plan is often static (i.e. lead times are fixed and processes/demands are deterministic) and capacity is assumed to be infinite. This is not how a factory operates, of course, so the ERP schedule is not the final plan for the factory.

Every one of our customers to date has reverted to some elaborate set of Excel spreadsheets with manipulable cells that allow experienced operators to reevaluate the production schedule as assumptions change. Of course, this is not optimal. These tools are clunky, prone to human error, and force users to wave their hands where calculation is necessary.

This is where “Advanced Planning and Scheduling (APS)” enters the picture. In theory, APS exists to correct the most obvious failures of ERP-based planning by introducing finite capacity, shared resource contention, sequencing constraints, and changeovers. In practice, APS systems are off-the-shelf job shop solvers wrapped in complex UI and integration layers.

For companies that successfully deploy APS, it is almost always the result of a multi-year (and multi-million $) implementation effort involving consultants, custom integrations, hand-tuned constraint models, and dedicated in-house operations research expertise.

The solver may be “advanced,” but the surrounding process is fragile. Cycle times are still assumed to be fixed. Downtime is modeled, if at all, as a static buffer. Human variability, supply disruptions, and execution noise remain largely outside the model.

As a result, APS often produces a theoretically optimal schedule that breaks almost immediately upon contact with the real factory. When disruption occurs, the system must be manually re-parameterized and re-solved. This reinforces the same behavior we see with ERP: operators export the APS output back into spreadsheets, override it by hand, and rely on tribal knowledge to keep production moving.

In other words, today's APS tools attempt to optimize a snapshot of a factory that does not exist. They assume determinism in a world that is inherently stochastic, and they require a level of data cleanliness and modeling precision that most manufacturers simply do not have.

The “dream state”

The dream state is not a perfect static schedule. It is a system that continuously plans, schedules, and adapts as the factory changes. Instead of fighting spreadsheets and frozen assumptions, planners and plant managers work with a living operational model of the factory that ingests ERP data, live shop-floor signals, labor availability, material constraints, and machine health in near real time. Plans update as conditions change. Risk is quantified. Tradeoffs are explicit. The computer should be a perfect calculator; a crystal ball for operators.

AI is the layer that makes this possible at scale. Rather than asking humans to manually rebuild models or retune solvers, AI agents translate messy real-world data into structured planning inputs, detect when assumptions are breaking, and propose corrective actions across planning, scheduling, and execution. The planner no longer spends their time reconciling data and forcing feasibility but rather making high-level decisions worthy of human intelligence.

In this world, simulation is no longer a one-off engineering exercise. It becomes an always-on validation layer for every major planning decision. Schedules are not trusted because a solver says they are optimal; they are trusted because they survive when compared to the real-world. Execution continuously feeds back into planning, and planning continuously reshapes execution.

This is what we mean by AI-native operations: not a chatbot on top of ERP, not a brittle APS optimizer, but a closed-loop decision system that learns from the factory every day and gets better over time. The factory becomes a system that can reason about itself.

No consultants, no waiting for and then learning a new software for months, just us working with you to systematize your planning process. We're as excited to make factories the most beautiful machines in the world as you are.