Constraint-Aware MPS Generation

A master production schedule (MPS) is the bridge between an approved demand plan and executable factory work. When that bridge is built by hand, planners re-key demand into ERP planning runs, chase capacity calendars across plants, and reconcile BOM lead times in spreadsheets. The result is slow cycle time, version drift between S&OP and the shop floor, and schedules that look complete but violate a constraint that only surfaces after release.

Constraint-aware MPS generation automates the translation step. An agentic system reads the approved demand plan, evaluates capacity, lead times, and BOM structure, and proposes a time-phased MPS for human approval. The goal is speed with feasibility: fewer days spent keying and reconciling, and fewer releases that fail because a work center, component, or lot size rule was ignored.

This page describes how a master scheduler should run, review, and govern that workflow. It assumes demand has already been agreed in the planning cycle. Upstream forecast construction is covered in Multi-Signal Demand Forecasting. Downstream recovery when the live schedule breaks is covered in Real-Time Schedule Reoptimization.

The agent should not invent missing planning data. Before any MPS is proposed, it must assemble a complete constraint set. Treat the following as hard prerequisites.

Demand plan. Firm or approved independent demand by SKU, location, and period. Include any firm customer orders or allocations that the MPS must honor ahead of forecast. Reject soft or draft demand that has not cleared the planning gate.

Capacity. Finite capacity by work center or resource group for the planning horizon: available hours, shifts, planned downtime, and any shared bottlenecks across product families. If capacity masters are incomplete, stale, or missing for a demanded family, stop. Use Capacity Bottleneck Identification when the question is which resources bind first; this use case consumes those answers as capacity truth, not as a substitute for missing calendars.

Lead times. Manufacturing, purchasing, and transportation lead times that affect when lower-level demand must land. Include frozen zones and minimum planning horizons so the agent does not propose starts inside a lock window.

BOM and lot rules. Active multilevel BOMs, effectivity dates, scrap or yield factors, lot-size policies, and make-or-buy flags. Incomplete or contradictory BOMs (missing components, expired effectivity, circular structures) are a hard stop.

Inventory and open supply. On-hand, quality holds, open production orders, and open purchase orders that already consume capacity or cover demand. Without this netting picture, the MPS will overstate new starts.

When any of capacity or BOM inputs fail completeness checks, the agent returns empty or blocked. It should state which SKUs, plants, or resources are missing data, and it must not post a partial MPS that looks executable. Empty or blocked is the correct outcome: incomplete constraint data is not a signal to guess.

With inputs complete, the agent converts approved demand into a proposed master schedule through a constrained planning pass rather than a free forecast dump.

1. Net demand. Explode or attach dependent demand where the MPS scope includes make items with multilevel structure, then net against inventory and open supply. Preserve firm orders and allocations.

2. Time-phase starts and receipts. Apply lead times so planned receipts land in the periods demand requires, and planned starts fall outside frozen zones. Respect lot-size rules so proposed quantities are manufacturable, not merely arithmetically neat.

3. Load capacity. Assign starts to resources that can produce the item. Compare load to available capacity by period. When a period is overloaded, shift within allowed windows, split lots only if policy allows, or flag unmet demand. Do not silently drop volume.

4. Surface trade-offs. The proposal should make constraints visible: overloaded resources, late receipts relative to demand, materials that cannot arrive in time, and SKUs blocked by BOM or capacity gaps. Schedulers need the conflict list as much as the schedule grid.

5. Produce a review package. Output a time-phased MPS by SKU and plant, capacity load by resource, exception list, and a clear mapping back to the demand-plan version used. That package is what the scheduler reviews; it is not yet ERP truth.

The agent’s value is consistent application of rules at speed. It does not replace judgment on which demand to delay, which overtime to authorize, or which alternate BOM to use when policy allows more than one path.

The master scheduler remains the control point. The agent proposes; the scheduler approves; only then does the system post to ERP.

Review should focus on:

• Alignment with the approved demand-plan version and any firm-order commitments.
• Capacity realism on known bottlenecks, including shared resources and planned downtime.
• Lead-time and freeze-zone compliance (no starts or receipts inside locked periods unless policy explicitly allows a break).
• BOM and lot-size validity for every proposed start.
• Exceptions the agent could not clear: unmet demand, overloaded periods, or SKUs left blocked.

Approval is explicit. Until the scheduler signs off, nothing writes planned orders or MPS rows into the system of record. After approval, the integration posts to the planning layer used by the plant (for example SAP S/4HANA PP planned orders or equivalent objects in Kinaxis RapidResponse or PlanetTogether) with audit of who approved, which demand version, and which constraint snapshot applied.

If review finds bad inputs or an unacceptable trade-off, reject or request a replan. Do not patch the ERP by hand to “fix” an agent proposal that was built on incomplete capacity or BOM data. Fix the masters, re-run, and approve a clean package.

This pattern typically sits beside existing APS and ERP planning stacks rather than replacing them.

SAP S/4HANA PP. Use approved MPS output to create or update planned orders and related MRP-relevant objects after human approval. Keep capacity and BOM masters in ERP as the authoritative source the agent reads; avoid a shadow BOM in the agent layer.

Kinaxis RapidResponse. Fit well when the demand plan and scenario comparison already live in the planning control tower. The agent can generate a constrained scenario for scheduler review, then promote only the approved plan into the live planning model.

PlanetTogether. Useful when detailed finite scheduling and the MPS handoff must stay close. Keep the MPS approval gate above detailed dispatch so plant-floor reoptimization does not silently rewrite the master plan without scheduler consent.

Across vendors, keep three operating rules stable: read constraint masters from the systems of record, block on incomplete capacity or BOM data, and require scheduler approval before any ERP or APS write.

Constraint-aware MPS generation assumes a trustworthy demand signal and clear capacity truth, and it feeds a schedule that will still need live recovery.

• Build or refresh the demand input with Multi-Signal Demand Forecasting.
• Locate binding resources before or during the capacity load step with Capacity Bottleneck Identification.
• Recover when execution invalidates the approved plan with Real-Time Schedule Reoptimization.

Used together, these flows keep the path from demand plan to master schedule fast enough for planning cycles, while leaving feasibility and release authority with the master scheduler.