When implementing finite capacity planning at a factory in Thailand, do not choose a system solely because it displays free machine hours. A machine may be free while the qualified operator is on leave, a material lot is unreleased, or the machine is reserved for maintenance. The purchase objective is an executable plan and a clear explanation when an order cannot be executed. This guide connects constraint design, data preparation, the request for proposal (RFP), factory and site acceptance tests (FAT/SAT), and management decisions.
Decide the purpose of finite capacity planning before choosing software
An infinite capacity plan can place an order at its theoretical start date even when the required resources are overloaded. Planners can adjust the load later. That approach becomes harder to manage as product variants, alternative equipment, setups, shifts, materials and maintenance compete for the same time. Finite capacity planning schedules work within defined resource availability and business rules. Siemens identifies equipment, inventory, employee qualifications and availability, and maintenance schedules among the data needed for manufacturing capacity planning. SAP describes finite planning as a way to consider scarce capacity before detailed scheduling. Neither source can decide which constraints your factory must enforce.
Start the investment case with a real order that exposed a planning failure. Perhaps a due date was promised based on machine hours while the only qualified changeover technician was assigned to another line. Perhaps an urgent order displaced an already committed order without a recorded approval. Perhaps a night shift was removed but the plan continued to assume it existed. Preserve the order ID, item, operation, promise date, actual event and decision. Turn that case into a procurement and acceptance scenario.
Management must also define the objective. “Improve on-time delivery” cannot decide between two competing orders. Agree how to rank committed due dates, work in progress, overtime, changeovers and priority customers. Do not promise to maximize all measures at once. Decide first which constraints cannot be violated; then define what the optimizer should prefer among feasible alternatives. Agree who may freeze a plan, authorize an urgent change and issue a customer delivery commitment.
Set the planning horizon and level of detail
Monthly capacity outlooks, weekly production plans and next-shift dispatch sequences answer different questions. Executives need to decide whether to invest in a bottleneck; planners need to decide whether to add overtime or subcontract next week; supervisors need the next executable job. Start at the frequency at which the plant actually updates decisions. Minute-level precision on an operation managed by day imposes data maintenance without adding value. Conversely, a day-level bucket can hide a setup conflict on a machine that changes products several times per shift.
Define the model boundary. Rather than model every plant and operation at once, select one due-date-critical resource and connect its upstream materials and downstream operations. State approximations openly: perhaps upstream work is represented by a confirmed arrival, outside processing by an accepted receipt date, and a downstream line by daily capacity. Name the person who updates each approximation when circumstances change.
Four constraints that a production capacity planning system must reconcile
“Include all constraints” is not a testable requirement. For each rule, state whether it is hard, meaning the plan must not violate it, or soft, meaning a warning and an authorized exception are allowed. Too many hard constraints may prevent any plan; too many soft ones produce plans that the floor cannot execute. The key is to test the following four against the same order and operation.
| Constraint | Minimum input | Feasibility question | Data owner |
|---|---|---|---|
| Equipment and tools | Machine ID, capacity, alternatives, setup, calendar | Is a dedicated fixture already in use? | Production engineering |
| People and skills | Shift, qualification, expiry, required crew | Is a qualified operator available? | Production and HR |
| Materials | Item, quantity, availability date, lot condition | Is the stock actually allocatable? | Purchasing and warehouse |
| Maintenance | Machine, start/end, confirmation status | Does production overlap a planned stop? | Maintenance |
Describe equipment under real operating conditions
A catalog cycle time alone excludes cleaning, inspection, mold changes, start-up checks and breaks. Keep standard run and setup time by item and machine, minimum batch size, and the relevant calendar. For an alternative machine, record eligible products, fixtures, quality approval and transfer time. SAP distinguishes detailed, production-rate and rough-cut scheduling; its documentation also notes that calculating more scheduling levels affects performance. Model the level needed for a decision rather than requesting maximum detail everywhere.
Model skill availability, not only headcount
“Two people required” does not mean any two people can do the job. Define roles such as authorized setup technician, qualified inspector and safety sign-off by operation. Include qualification expiry and shifts. If individual names are unnecessary in the planning engine, use skill pools and let supervisors assign people on the floor. The model’s detail must match the factory’s ability to keep the data current. Decide whether a sick leave event is received from HR automatically or entered by a supervisor on the day. A proposal that claims “labor constraints supported” without explaining this workflow is incomplete.
Separate recorded stock from usable material
Positive inventory can still be unusable because it is on quality hold, reserved, in a different location, or outside the permitted lot conditions. Open purchase orders have changing arrival dates. Specify which of inventory, incoming supply, quality release and lot expiry the planner uses. Distinguish confirmed supply from an estimate. An order waiting for material should show “machine and crew available; material unavailable” and provide an action for purchasing. If sales uses an estimated receipt date to promise delivery, sales and purchasing must agree the commitment rule.
Put maintenance on the same resource calendar
Preventive maintenance must occupy the same machine hours as production. A confirmed stop may be a hard constraint, while a proposed stop may generate a warning. Statutory inspection and safety stops must not be removable by a production planner. When maintenance changes, display affected orders and specify who approves the new schedule. For breakdowns, consume the event from a shop-floor or maintenance system, or enter it directly, and set a deadline for replacing the stale plan.
Prepare data by assigning a source, freshness rule and exception process
When the scheduling screen exposes a data error, users need to know which system to correct. Otherwise they will make repeated local edits that never reach the source. ISA-95 provides a common model for discussing business planning, manufacturing operations and the information exchanged between them. Use it to clarify ownership of orders, work instructions, actual production, equipment state, stock and maintenance. ISA-95 is a reference model, not proof that a vendor’s integration will work.
A minimum data dictionary names each field, meaning, unit, time zone, source, refresh frequency, exception rule and history policy. A night shift crossing midnight will be scheduled incorrectly if a local Thai shift calendar and UTC event timestamps are mixed without a conversion rule. When ERP and shop-floor terminals use different item IDs, assign an owner to the mapping table and its change procedure. When a BOM revision and work-order effectivity date disagree, say which controls material allocation.
Audit a sample of recent orders on the chosen bottleneck before attempting to clean every record in the plant. For each order, compare planned and actual start and finish, setup, stoppage, material wait and skill assignment. Classify missing data as “cannot calculate,” “can use a documented approximation,” or “requires floor confirmation.” Prioritize missing fields that change delivery promises rather than merely reporting a high percentage of completed fields. Link standard-time maintenance to product and process changes.
Missing values require different responses. If setup time is absent for a few products, the planner might use a comparable item’s time provisionally and mark the affected orders. If no one has confirmed the presence of a safety-qualified operator, an average headcount must not make the job appear ready. If a material lot’s quality status is missing, positive stock should not become a firm allocation. Mark orders calculated with provisional values and require review before confirming a delivery date or releasing a work instruction. A blanket conversion of blanks to zero or “available” can make a tidy plan that the plant cannot follow.
The audit should produce a way to detect stale data, not only a list of repairs. Keep the last update time for standard times, qualifications, confirmed maintenance and supplier arrivals. Warn when values exceed their agreed review period and keep them out of automatic commitments until checked. When a value changes, identify unstarted orders that depend on the item or machine and decide whether to replan. This is how the master data stays useful after initial migration.
Do not demand “real time” for every interface. Confirmed orders, goods receipts, absences, machine failures and maintenance changes affect decisions at different speeds. Specify allowed delay from event to updated plan, fallback entry during a communication outage, replay after recovery and an ID that prevents duplicates. Decide whether approved delivery dates and shop-floor instructions return to ERP or MES, and protect unapproved alternatives from overwriting the live plan.
Put test scenarios, not a feature checklist, in the APS implementation RFP
If the RFP only says “finite capacity scheduling supported,” vendors can tick the same box while modeling different constraints. Provide one data set, scenarios and expected decisions. Specify where software proposes an option and where a human must approve it.
Begin with a concise description of plant, product family, lines, planning frequency, ERP/MES/maintenance systems and decision owners. Rank constraints and describe mandatory exceptions, interfaces, access rights, audit history, languages, training and fallback operations. At a Japanese-owned plant in Thailand, headquarters and local teams may have different authority over a committed date. Specify Thai, English and Japanese terminology where the workflow needs it. “Translatable” does not prove that shift names, reason codes, warnings and reports will be usable.
Ask every shortlisted supplier to run the same cases:
- Release an order to an idle machine while the required certified person is absent. Does the system block the start and show why?
- Move a material arrival later. Which orders move, and which committed due dates change?
- Confirm a maintenance stop and add an urgent order inside the frozen horizon. Can the order override the stop without approval?
- Send a product to an alternative machine that requires a different setup and quality approval. Does the planner check both?
- Remove a night shift and record an absence next morning. Are cross-midnight times recalculated correctly?
- Record an actual deviation after plan approval. Can users trace original plan, reason, approver and revised plan?
Require the proposal to separate standard functionality, configuration and custom development; licenses, interfaces, master-data cleanup, training, support and export of the factory’s data. Measure calculation performance against stated order, operation and resource counts, planning horizon and refresh frequency. “Fast” is not an acceptance criterion. The planner also needs understandable reason codes and a history of manual changes. A mathematically feasible answer that no one can explain will be hard to adopt.
Apply hard-constraint compliance as a gate before scoring plan quality, data burden, usability, maintenance, expansion, total cost and local support. Management should distinguish rules that protect the business from habits that can change. For other parts of the decision, see our production scheduler comparison and guide to moving production planning from Excel. This article focuses on the procurement conditions for simultaneous constraints.
Use FAT and SAT to prove the plan works at the plant
Factory acceptance testing checks, in a supplier or controlled test environment, whether the configuration and logic produce the agreed result for defined inputs. Site acceptance testing checks the live interfaces, data, rights, shift operations and human decisions at the plant. Define the contract meaning of FAT and SAT for the project. A screen that opens is not sufficient evidence of acceptance.
Connect every case to the RFP and capture input, expected result, tolerance, evidence and owner. A job must not be assigned to a machine during confirmed maintenance; this can be tested mechanically. Prioritizing two orders due at the same time requires comparison with policy. When a key order will be late, the delay must be visible, its cause identified as material, labor or machine capacity, and alternatives available for review.
| Test | What to verify | Evidence |
|---|---|---|
| Constraint breach | No double-booked machine, unqualified labor, unallocatable material or maintenance overlap | Input-to-plan reconciliation |
| Change impact | Delayed supply or breakdown identifies affected orders and due-date changes | Before/after plans and reason log |
| Interface and time | ERP/MES/maintenance values arrive once and night shifts are handled correctly | Message IDs, timestamps and replay history |
| Human decision | Supervisor can understand, approve or reject an exception | Action and approval logs |
| Recovery | Missing data after an outage is detected and replanned | Outage exercise record |
Use the Thai plant’s real calendar in SAT: breaks, night shifts, holidays, temporary shift changes, maintenance stops and quality-held materials. Check that Thai and Japanese readers understand the same warning in the same way. Observe whether the local planner can explain tomorrow’s sequence without rebuilding it in an outside spreadsheet. Compare handling time and rework against a measured local baseline; do not invent a benefit percentage afterward.
During parallel operation, record differences between old and new plans. A difference does not automatically mean the new plan is wrong. The old plan may have hidden overtime, an informal alternative machine, early purchasing or a negotiated maintenance window. Decide whether each accepted exception becomes a formal rule or stays in an approval workflow. SAT exit conditions should include no unresolved critical defect, passing priority scenarios, transferred master-data ownership and locally usable outage instructions.
Test the case where no feasible plan exists. Several hard constraints may conflict. The system must not fabricate spare capacity or silently remove a maintenance stop. It should identify the order, operation, time window and resource in conflict, then route the issue to a real decision: approved overtime, waiting for material, checking an alternative machine or negotiating a new customer date. Verify that the change reason and approver remain visible to the next planner.
Management must own decisions that the calculation exposes
A finite planning engine does not remove delivery problems when it is installed. By making finite resources explicit, it reveals shortages that planners previously handled informally. Management must define who decides on overtime, subcontracting, alternative equipment, due-date renegotiation and priority customers. These are business choices, not decisions an algorithm should make without authority.
For the investment case, do not copy generic vendor benefit rates or another factory’s result. Measure a baseline for time to answer a delivery request, plan changes, changed dates after commitment, material waiting, conflicts with maintenance, overtime and outsourcing, work in progress, and manual planning hours. Explain expected improvement through a causal chain: which constraint becomes visible and which decision becomes faster? Include master-data upkeep, interface maintenance, local training, shift changes and version updates in total cost. Pricing depends on scope and data condition; a generic market range is a weak budget basis.
Stage expansion behind gates. One possible sequence is bottleneck and adjacent-operation visibility, then simultaneous labor, material and maintenance checks, then additional lines or plants. Let the plant’s actual bottleneck determine the sequence. At each gate ask whether data stayed current, planners adopted the proposed plan, and exceptions remained traceable. Agree a stop condition so a successful demonstration does not automatically trigger a company-wide deployment.
A useful operating rhythm is simple. Daily, review deviations from yesterday’s plan, today’s hard constraint conflicts, supply and labor changes, maintenance changes and delivery commitments that need revision. Weekly, classify repeated exceptions as data gaps, capacity shortages or conflicting rules. Monthly, review investment in alternative machines, skills, sourcing and maintenance policy. Without a forum that evaluates and updates decisions, the team will return to off-screen adjustment.
Put headquarters and the Thai plant on the same decision path
At a Japanese-owned factory in Thailand, headquarters sales may promise dates while local production and purchasing know the resource limitations. An unclear definition of “confirmed date” can lock an impossible promise into the plan. Map who can commit, revise and approve exceptions across headquarters, the Thai factory, purchasing and maintenance. English may be the common office language, but verify Thai wording for shop-floor reason codes and training.
Before procurement, walk one product family from order to shipment. Inspect movement tickets, machine-stop records, maintenance plans and material allocation, not only the planning spreadsheet. Audit bottleneck master data and decide hard versus soft constraints with management. Then issue the same scenario set in the RFP to expose differences in supplier assumptions. Contract for interface ownership, conversions, exception handling, FAT/SAT evidence, training and handover. Before go-live, a local planner should be able to explain in their own words why an order cannot start today.
This order of work applies whether you buy a dedicated scheduler or extend existing ERP/MES functions. SAP’s PP/DS documentation for 2025 FPS01 describes detailed resource and component availability planning for critical products and bottlenecks. Siemens describes scheduling with multiple resources, sequence-dependent setups and inter-operation constraints. Available product capability is only one part of success; usable local data and accepted operating rules are tested in the RFP and acceptance process.
Four go-live gates that make acceptance explicit
The data gate asks whether unstarted orders, operation times, equipment calendars, required skills, material availability and confirmed maintenance all have owners and update timestamps. A low count of blanks is not enough; check whether any missing input changes a promised delivery date. A material awaiting confirmation may require the order to be excluded from automatic promises. Data that looked complete in a test copy but is no longer maintained in daily work does not pass.
The explanation gate checks three cases: an idle machine without a qualified operator, stock on quality hold, and an urgent job colliding with a maintenance stop. The display should identify all relevant constraints, not just the first hidden reason, so the planner and supervisor know whom to contact. Alternatives must retain the necessary approval for overtime, quality release on an alternative machine or customer-date changes.
The routine-work gate follows a real morning plan revision through shop-floor instruction. Planning, maintenance, purchasing, supervisors and sales each need a current view and a way to update their own data. If a receipt moves from morning to afternoon, the changed supply must reach the plan, stale instructions must not remain on the floor, and sales must review any changed promise candidate. A separate spreadsheet built to bridge every handoff means the operating loop is unfinished.
The outage-and-recovery gate shows the last valid update when a network or source system fails. Staff must have a rule for avoiding new firm promises based on unverified data. After recovery, replay orders and machine-stop events without duplication, replan affected orders, and import manual records made during the outage without losing prior plan and approval history. Assign separate sign-offs: IT for interfaces and rights, production for sequence, purchasing for supply, maintenance for stops, sales for promises and management for exception priorities. Each open issue needs an affected order, interim procedure, owner, deadline and next decision date.
Even a small pre-purchase pilot can use these gates. Select representative orders, including a routine product, changeover, material delay, maintenance stop and absence. Ask a supervisor outside the planning team to explain why a job cannot start and compare the answer with the screen. Keep each bidder’s assumptions visible: if one treats material as infinite and another strictly checks qualifications, their due-date scores cannot be compared without adjustment.
What data should be prepared first for finite capacity planning implementation?
For the selected bottleneck, start with orders and operations, machine calendars, standard times, major setups and confirmed maintenance stops. Add labor skills and material availability dates that actually change delivery decisions. Reproduce recent orders and check whether the reason an order cannot be planned is correct. Record missing fields, provisional values, owner and deadline rather than concealing gaps.
Are a production capacity planning system and APS the same thing?
Capacity planning is the business process of comparing demand with resource capability. APS describes a family of software tools that may support it. An APS label alone does not tell you which equipment, labor, material and maintenance rules can be checked together, at what detail, or how ERP/MES data is exchanged. Test the named scenarios.
Can finite capacity scheduling automatically reflect maintenance and absences?
It can when the source has confirmed data and the update timing and owner are defined. The design must distinguish tentative from confirmed stops, breakdowns from preventive maintenance, and absence from missing qualifications. Include outage alerts, manual fallback and message replay in acceptance tests.
How should the cost and duration of APS implementation be estimated?
Estimate by the number of constraint types, item-operation-machine combinations, master-data gaps, ERP/MES/maintenance interfaces, languages, training, parallel running and support, not just by line count. A uniform price or schedule before data audit is unreliable. Give suppliers the same RFP and test cases and compare their assumptions and exclusions.
Conclusion
Implementing finite capacity planning means checking equipment, labor, material and maintenance against the same order and explaining both feasible plans and infeasibility. Choose a bottleneck and management priorities, assign data sources and freshness rules, and put exception scenarios in the RFP. FAT and SAT should test constraint compliance, change impact, approvals, recovery and users’ understanding. After go-live, review deviations and exceptions regularly and use them to adjust resource investment and operating rules.
If your Thai plant is still defining which rules must be hard constraints and where the data comes from, contact TOMAS TECH. One current planning sheet and one real order can be enough to begin structuring an RFP and acceptance criteria.
