Production Planning Simulation | Detailed Scheduling Guide

When someone says “we want to simulate our production plan,” what they actually have in mind varies enormously. Some picture a digital twin that recreates the whole factory in a virtual space. Some picture a spreadsheet with formulas that run a few numbers. And some have already decided the whole idea is an academic exercise that will never survive contact with the shop floor. This article narrows production planning simulation down to one specific thing — validating the short-term plan, the level that turns directly into today’s work instructions.

The subject here is not the return on investment of AI, and not which scheduler product wins a comparison. It is the concept itself — what a simulation feature actually is, and where in short-term (detailed) scheduling it earns its keep. For how AI is categorised and how to think about its ROI, see What AI Production Planning Really Means. For the RFP and PoC stages of choosing a product, see Production Scheduler Comparison 2026. What follows is the step before both of those — understanding what the capability does.

The three layers of production planning, and where simulation fits

A production plan is not a single sheet

The reason “production plan” causes people to talk past each other in meetings is that planning has layers, and each participant is thinking of a different one. Production planning is generally split into three layers — the long-term plan, the medium-term plan, and the short-term plan. Each has its own planning horizon, its own time bucket (the unit the plan is sliced into), and its own owner.

The long-term plan covers one to five years. The time bucket is quarters or months. It belongs to executives and corporate planning. What gets decided here is the frame that only pays off over years — capital investment, headcount planning, how production is allocated between sites.

The medium-term plan covers three months to a year. The time bucket is days. It belongs to production control. This is the layer where the MPS (master production schedule) and MRP (material requirements plan) are built — what to make, when, in what quantity, and when the materials have to be ordered to support that.

The short-term plan covers one week to a few months. The time bucket is hours and minutes. It belongs to production control together with the line leaders on the floor. Only here does the plan become an actual instruction — which machine, which operator, which item, starting at what time. When people on the shop floor say “the plan,” this is almost always what they mean.

The lower the layer, the more often it gets rewritten

Lay the three layers side by side and the pattern is clear. The lower the layer, the finer the time slice and the more often it changes. A long-term plan can be revisited a few times a year. A medium-term plan is typically refreshed weekly. A short-term plan gets rebuilt several times a day in plenty of factories, and nobody finds that unusual.

And the more often a plan gets rewritten, the less time there is to think before rewriting it. Thirty minutes before the morning meeting to resequence the day. A material delay discovered after lunch that forces the afternoon assignments to change. Every one of these decisions is made without checking what the new sequence will actually do, because there is no practical way to check.

That is exactly where production planning simulation belongs — the layer that changes constantly, allows the least time to decide, and yet feeds straight into delivery performance. In other words, the short-term plan and the boundary where it meets the medium-term plan above it.

What-if checking as the everyday use

In practice, the most frequent use of simulation is what-if checking. “If we squeeze this rush order into this evening, what else slips?” “If we group similar part numbers to cut changeovers, how much does inventory grow?” These are questions you answer before anything is released to the floor rather than after.

The important thing is that simulation does not produce “the correct plan.” It produces what will happen if you choose a given plan. The decision to insert the rush order is still a human one. Simulation’s job is limited to showing the consequences of that decision in advance. If that boundary is left vague during a rollout, the project ends in the familiar complaint that “the system was supposed to build the plan for us.”

Why plans and reality drift apart

Plenty of factories run plans that never quite happen, and the cause is structural rather than a shortcoming of the planner. The problems built into manual scheduling come down to three — dependence on individuals, slow response to change, and uneven load.

Dependence on individuals

In most factories the short-term plan is built in a spreadsheet plus the planner’s head. Which part numbers run back to back to keep changeovers short. Which machine loses yield on humid days. Which operator is genuinely comfortable with which fixture. None of that appears in a cell of the planning sheet. It lives in one person’s memory.

The consequence is that plan quality changes depending on whether that person is in the building. A single annual holiday is enough to degrade it, and a resignation or a repatriation far more so. This is not a question of competence; it is a question of decision criteria never having been written down.

At Japanese-owned plants in Thailand, this dependence acquires an extra layer. The person holding the reasoning is often a Japanese manager, while the person entering the daily data is Thai staff. What gets handed over in that arrangement is the routine — “this is how we do it” — while the reasoning behind it stays unspoken. Once only the routine survives, nobody can correct it when the underlying assumptions change.

Slow response to change

A plan starts aging the moment it is finalised. Material does not arrive. A machine stops. A customer changes the quantity. An operator calls in sick. These are daily events, not exceptions.

The problem is the lag between the event and the plan reflecting it. When one machine goes down for half a day, which part numbers route through it, which downstream processes now have open capacity, and which customer orders are about to miss their dates? Chasing that by hand turns into a chain of phone calls and spreadsheet edits, and simply establishing how far the impact reaches can take hours.

Meanwhile the floor keeps running on the old plan. By the time the rebuild is finished, something else has happened. The rebuild never quite catches up, and the planning sheet quietly gets demoted to “reference material.” What simulation compresses is precisely this tracing time. If you can lay out two or three resequencing options and see which orders slip under each within minutes, the speed of the decision itself changes.

Once the planning sheet becomes reference material, the floor starts deciding sequence on its own judgement. At that point measuring plan-versus-actual loses its meaning too, so the next plan is no more accurate than the last. Drift breeds more drift.

Uneven load

The third problem is not noticing that load is concentrated on particular machines or particular people. A planning sheet shows what gets made and when. It does not show how much load that puts on whom at that moment.

The result is a state that hardens over time — one machine permanently running on assumed overtime while the machine next to it sits idle, or work piling onto the only operators with a specific certification, so the whole plan collapses when one of them takes leave. These peaks and troughs are frequently invisible even to the person who wrote the plan.

What makes uneven load awkward is that it is rarely recognised as a problem. As long as due dates are met, habitual overtime gets treated as simply how things are. It surfaces only when order volume rises or when a key person leaves — and by then the operation has been built around the imbalance. Being able to calculate load as a stacked profile lets you check how much additional volume the imbalance can absorb before it becomes a crisis rather than after.

How much capacity is quietly disappearing

When these inefficiencies stack up, a meaningful share of a factory’s capacity is never used at all. According to one analysis, manufacturers may lose 10 to 30 percent of production capacity to avoidable inefficiencies such as downtime, poor scheduling, and unbalanced workforce allocation.

That is a wide range, and it must vary considerably by industry and production model. It is not a figure to treat as settled. But add up the changeovers caused by poor sequencing, the machines left waiting, and the lopsided overtime, and the underlying feeling — that the plant is not using all the capacity it already owns — is one most production control managers recognise. Before adding equipment, confirm what percentage of existing capacity is actually being used. Simulation is, among other things, a tool for making that check.

The six elements of production simulation and what each reveals

Production simulation models the real production floor and tests the effect of a plan in advance across multiple scenarios. Modelling here does not mean recreating the factory faithfully. It means extracting only the elements a decision depends on and putting them into a form that can be calculated.

There are six principal elements. Each corresponds to a specific category of shop-floor failure you can catch before it happens.

Line balancing

Equalising work content across processes. The pace of a whole line is set by its slowest process. However fast the others run, the line backs up there.

Model it, and you can calculate which process is governing the line, and how the overall pace changes if that work is split or redistributed. The failure this catches in advance is the classic one — adding people and seeing no change in output. Adding headcount to a process that is not the bottleneck does not move the end of the line. You can verify that before anybody is reassigned.

Capacity planning

Forecasting maximum output from equipment capability. You stack up how many pieces each machine can process per unit of time and derive how much can be made in the target period.

The failure this catches is an impossible delivery date promised when the order is taken. The date sales committed to already exceeded the ceiling of the equipment. Nobody notices at the moment of the promise, and it comes to light just before production. With the capacity ceiling held as a model, that check happens before the answer goes out.

Resource constraints

Building limits such as tooling counts and operator skills into the plan. A machine may be free, but if the fixture that part number needs is mounted on another machine, it cannot run. An operator may be on shift, but without certification for that process they cannot be assigned.

The failure this catches is the same-day stall where “the machine is open but we can’t run it.” A plan that works on paper and not on the floor. That mismatch only becomes calculable once tooling and certifications are in the model. The less cross-training a factory has, the more this element pays back.

Buffers between processes

Making work-in-process (WIP) volume and dwell time visible. How much intermediate stock accumulates between two processes, and how long it sits there.

The failure this catches is running out of storage space and stretching lead time. A plan that is achievable purely on output numbers can still generate more WIP than the intermediate storage will hold. In real life the aisles fill up with material set down temporarily, and handling takes extra effort. Model the buffers and this is visible at the planning stage.

Changeover time

Modelling the setup time that comes with switching products. Running part number B after A takes a different setup than running C after A. This sequence-dependent time goes into the calculation.

The failure this catches is capacity lost to how the sequence was assembled. Plans built purely from quantity tend to produce sequences packed with changeovers. Add changeover time to the calculation and the numbers show that the same quantity yields different running hours depending only on the order. In high-mix, low-volume plants this is frequently the element with the largest payoff.

Variability

Reproducing equipment breakdowns and fluctuating task times as probability distributions. Where the previous five elements assume fixed values, this one takes the movement of those values as its subject.

The failure this catches is a plan that only holds together at theoretical values. A plan built on everything running to standard time is never once achieved in practice. Put failure rates and task-time distributions into the calculation and the same plan produces a range between the good case and the bad one. Seeing that range is what lets you decide where to build in slack.

You do not need all six before the method is usable. Two or three is usually enough to start. Decide first which row your own recurring failures belong to, and put that element into the model. That is the realistic way to get simulation running in day-to-day operations.

Making risk visible before live operation, and quantifying improvement by comparing scenarios against KPIs — those two capabilities are the core value of the method.

The relationship to digital twins and APS, and where the technology is going in 2026

The APS market is growing at roughly ten percent a year

Most simulation features reach factories as part of a production scheduler product in the category known as APS (Advanced Planning and Scheduling). That market keeps expanding.

The APS market is forecast at $1.32 billion in 2025 and $1.47 billion in 2026, a year-on-year CAGR of 11.2 percent, growing to $2.11 billion by 2030 at a 2026-2030 CAGR of 9.5 percent. The growth is being driven by the integration of AI and machine learning, the spread of cloud-based APS, demand for digital twin simulation, the expansion of global production networks, and rising demand for real-time decision-making.

Of those five drivers, the two that will matter soonest to a Japanese-owned plant in Thailand are the shift to cloud and the expansion of global production networks. The first lowers the entry cost. The second raises the need for planning that spans the Japanese head office, the Thai plant, and sites in neighbouring countries. A setup where Japanese demand forecasts and Thai short-term plans live in separate files breaks down the moment volume has to be reallocated between sites.

Dynamic scenario analysis with digital twins

The other current is the digital twin. Dynamic scenario analysis using digital twin simulation virtually reproduces equipment performance, operator capability, and material flow, letting you identify bottlenecks and optimise the schedule before real operations are affected. In one industrial manufacturer’s case, redesigning production flow and scheduling through digital twin simulation was reported to cut monthly costs by 5 to 7 percent.

What makes that figure meaningful is that no equipment was replaced. The flow and the sequence changed. If the cost structure moves with the same machines and the same people, the difference was created by how the plan was built.

Simulation does not mean deploying a digital twin

The caution here is how much ground the term “digital twin” covers. Picture a full 3D reproduction of the plant synchronised in real time with sensor data from live machines, and both the investment and the lead time look enormous. In practice, aiming straight at that configuration usually stalls in the data preparation phase.

Modelling the six elements described above, by contrast, requires neither a 3D reproduction nor real-time synchronisation. Standard times per part number, equipment capability, tooling counts, and a changeover time matrix. With roughly that much information, simulation works as a planning-side capability.

So the order runs the other way. Start using simulation as a way to validate plans, let that practice settle, and then increase how tightly it synchronises with actual production data. A digital twin is one possible destination, not the entrance. Reverse that order and the usual outcome is an impressive system that nobody uses for planning.

Where automated optimisation and agent-style AI fit into this is outside the scope of this article. Treat the six elements above as the foundation you need either way, whether or not AI is in the picture.

How to start using simulation at a plant in Thailand

What is specific to contract manufacturing in Thailand

Japanese-owned plants in Thailand, especially those with a high share of contract manufacturing, face structurally more sources of planning disruption.

First, high mix and low volume. Dozens of part numbers run on the same line and changeovers happen several times a day. Changeover time is the element most likely to pay off here, but the number of part number combinations that have to go into the model is also large, so preparation takes more effort.

Second, the frequency of customer-driven changes. Quantity changes and pulled-in due dates from the Japanese parent company or local customers arrive after the weekly plan was supposedly fixed. This is an environment where slow response to change bites directly.

Third, long holidays and staff turnover. Capacity swings significantly around Songkran and other extended holidays. In plants where operator turnover is faster than in Japan, the skill matrix is often in use long after it stopped being accurate. If resource constraints are going into the model, establishing the current state of skills comes first.

Fourth, material procurement lead times. Where a high proportion of parts are imported, customs delays undermine the assumptions the plan rests on. That uncertainty cannot be absorbed on the short-term planning side alone; it has to be linked to the medium-term material plan.

Do not model every process at once

The thing most worth avoiding in this environment is trying to model the entire process flow from the start. Part number master, process master, standard times, changeover matrix, skill matrix. Insist on having all of that in place before beginning and the people doing the work will burn out before the data is ready.

The workable approach is to start narrow, confirm the output is usable for decisions, and then expand.

Start with the bottleneck process. If you know which process sets the pace of the line, modelling that one alone is enough to calculate the effect of a sequence change. If you do not know, the first task is measuring actual performance process by process.

Next, add the processes with the most complaints and late deliveries. List the delayed orders from the past six months and count where they stalled. Take the top two or three processes by count and put them into the model alongside the bottleneck. At this point coverage is still partial, but most of the causes of lateness should be inside it.

Third, add elements one at a time. Changeover time first, then resource constraints, then variability. Add several at once and you lose the ability to tell which element changed the result. With each addition, reconcile against actuals to confirm the model still holds.

Fourth, put the output in front of a decision. Who looks at the simulation result, when, and what do they decide? Without that agreed, the calculation becomes a report nobody reads. The point at which the weekly production meeting can hold a conversation like “we compared option A and option B and chose B” is the point at which the rollout is genuinely complete.

Whether the floor accepts it is a separate question

Being able to build a model technically and having the floor use the result are two different things. Hand a calculated sequence to people who have been deciding sequence by experience for years, and the first response is that the calculation does not understand the shop floor. Most of the time that objection is correct. Constraints exist on the floor that are not in the model.

What determines whether the tool sticks is whether that feedback is patiently folded back into the model. “You can’t run this part number after that one.” “This machine is only available in the morning.” Each of those becomes a new line in the resource constraints or the changeover matrix. The more objections surface, the closer the model gets to reality. Treat the objections as resistance instead, and the model stays detached from reality — and eventually nobody uses it at all.

In Thailand, which language that dialogue happens in is a practical question of its own. Putting constraints into words is the line leader’s job, so the interview has to happen in whichever language that person explains most easily. Working through translation drops the fine nuances of a condition. The dropped conditions never make it into the model, and they reappear later in the form of “we ran it exactly as calculated and the line stopped.”

Separating product selection, execution systems, and AI

Once the scope is clear, the next question is the tool. Comparison criteria for APS products, what belongs in the RFP, and what a PoC should verify are covered in Production Scheduler Comparison 2026. The depth of simulation functionality varies widely between products, so deciding which of the six elements you actually need before starting a comparison will get you to a decision faster.

The other boundary worth drawing is between planning and execution. Simulation covers building and validating the plan. Recording whether the floor actually followed it belongs to the execution layer — the MES (manufacturing execution system). Perfecting the planning side alone will not improve the next plan if actuals never come back. Perfecting data collection alone leaves the slow response to change intact if planning is still manual. Which one to tackle first depends on the plant, but once selecting an execution system is on the agenda, Four Evaluation Axes for MES Comparison is a useful reference.

How far to go with automated optimisation and AI is an investment decision of a different character from the other two. The way benefits are measured and the questions asked in the approval process both change, so reading What AI Production Planning Really Means before entering that discussion makes the sequence of the conversation easier to organise.

Frequently asked questions

What is production planning simulation

It is a method that models the real production floor and tests the effect of a plan in advance across multiple scenarios. Equipment capability, tooling and skill constraints, changeover time, and work-in-process between operations go into the calculation, and you see what will happen when that plan is actually released. The goal is not to have a machine decide the optimal plan, but to give people the evidence they need when choosing between plans.

What is the difference between the short-term plan and the medium-term plan

The horizon, the time bucket, and the owner. The medium-term plan covers three months to a year in daily buckets, and production control uses it to build the MPS (master production schedule) and MRP (material requirements plan). The short-term plan covers one week to a few months in hours and minutes, and production control together with line leaders takes it down to instructions — which machine, who, what, starting when. It helps to think of the medium-term layer as deciding what to make, when, and how much, and the short-term layer as deciding how exactly to run it.

Will simulation eliminate late deliveries

No. What simulation reduces is lateness caused by how the plan was built — changeovers that piled up because of poor sequencing, delivery dates promised beyond capacity, a fixture shortage nobody noticed until the day. It cannot prevent sudden customer changes, customs delays on imported parts, or an unexpected breakdown. What does change is how those events propagate, because you can establish which orders slip and by how much in a short time rather than several hours.

Can production simulation be done in a spreadsheet

Within limits, yes. In a plant with few processes, a limited range of part numbers, and changeover times that do not depend on sequence, a spreadsheet can represent capacity stacking and load profiles perfectly well. It gets hard when sequence-dependent changeover time, contention over tooling, and probability-distribution variability all have to be handled at once. The number of combinations rises sharply and both computation and maintenance pass the practical ceiling of a spreadsheet. There is also the fact that the formulas themselves become dependent on whoever wrote them. Starting in a spreadsheet and moving the elements that hit the ceiling into a dedicated tool is a realistic path.

How does APS differ from production simulation

APS is the name of a product category; production simulation is the name of a method. Many APS products include simulation features, but the two are not the same thing. At the centre of APS is automatic generation of a schedule that satisfies a set of constraints. Simulation verifies what happens when a schedule is run, whether that schedule was generated automatically or built by a person. Improving the accuracy of automatic generation and letting people evaluate the generated result both require a way to verify. When comparing products, check the strength of automatic generation and the expressiveness of the simulation separately.

What data do we need before we can start

The minimum is three things — standard times by part number for the target processes, machine counts and capabilities, and approximate changeover times. With those three you can begin stacking capacity and comparing sequences. Tooling counts and skill matrices come in at the stage where resource constraints are handled. Variability distributions are more realistic to add once a reasonable volume of actual data has accumulated. Rather than assembling everything before starting, add only the data that corresponds to the element you are working on.

Summary

Production planning simulation covers a lot of ground, and different people imagine different things when they hear it. This article narrowed it to one part — validating the short-term plan.

Production planning has three layers, long-term, medium-term, and short-term, and the lower the layer, the more often it is rewritten. The structural reasons plans and reality drift apart come down to three — dependence on individuals, slow response to change, and uneven load. According to one analysis, avoidable inefficiencies may cost manufacturers 10 to 30 percent of production capacity, which is exactly the room worth checking before buying more equipment.

The content of a simulation is six elements — line balancing, capacity planning, resource constraints, buffers between processes, changeover time, and variability. Each corresponds to a particular kind of failure, so identifying which failures actually recur in your plant also identifies which element to start with.

The APS market is forecast to grow from $1.32 billion in 2025 to $1.47 billion in 2026 and $2.11 billion by 2030, with demand for digital twins listed among the drivers. But a digital twin is a destination, not an entrance. Start with simulation as a planning-side capability and expand from the bottleneck process and the processes that generate delays. In the high-mix, low-volume, high-change environment typical of Thailand, that order is the one that sticks.

Which process to model first, and how far you can validate with the data already on hand — this is the stage where plants in Thailand most often get stuck. At TOMAS TECH we start by listening to how your plan is built today and what data you have, then work through which element will show results soonest. You are welcome to get in touch while your plans are still at the exploratory stage through our contact page.

References

• Asprova, “Production Planning — Long-Term, Medium-Term and Short-Term Plans”
• Asprova, “Production Simulation”
• Research and Markets, “Advanced Planning and Scheduling Software Market Report”
• Simio, “Why Production Scheduling Software Will Look Different in 2026”
• Jodoo, “Production Capacity Planning in Manufacturing — Strategies to Meet Demand Without Overspending”