Production optimization software applies mathematical optimization methods to production planning and scheduling decisions finding the combination of which products to make, when to make them, on which resources, and in what sequence that maximizes a defined objective (throughput, margin, on-time delivery) subject to operational constraints. It overlaps with production scheduling software but extends the scope beyond schedule generation to broader production-system decisions.
The category is broader than scheduling alone. Production optimization can cover product mix decisions (which orders to accept given limited capacity), campaign planning (how to group products into manufacturing campaigns), resource allocation (which machines to dedicate to which products), and yield optimization (how to operate within process parameters to maximize throughput). Scheduling is one application of optimization; production optimization is the broader discipline.
This page covers what production optimization actually does, the mathematical methods involved, where it pays back, and how it relates to but extends beyond production scheduling.
Horizon's production optimization extends beyond scheduling to cover product mix, campaign planning, resource allocation, and order-acceptance decisions. The optimization engine uses MILP for tractable problems, constraint programming for scheduling-heavy problems, and metaheuristics for large problems where speed matters more than mathematical optimality.
The optimization integrates with adjacent planning functions: demand planning provides the order pipeline, supply planning provides the production plan, the scheduling layer provides feasibility validation. Optimization decisions flow through these layers a product mix decision changes the production plan, which changes the schedule, which changes material requirements.
For order-acceptance specifically, Horizon supports interactive order promising: a sales team can ask "can we accept this order at this date?" and the system responds with feasibility, margin impact, and the changes to existing commitments. This converts order-acceptance from a tribal-knowledge decision into a data-driven one.
The honest scope: Horizon's production optimization is most mature for discrete and process manufacturers with the constraint structures common to those modes. Process parameter optimization for continuous processes (refineries, petrochemicals, large-scale chemical operations) is typically better handled by specialized process control tools. Reinforcement learning for process optimization is an active research area we don't currently claim that capability in production deployments.
Scheduling answers a constrained question: given the work orders that exist, when should they run on which resources? Optimization answers a broader question: across the full set of decisions about what to make, when, where, and how what combination produces the best business outcome?
The difference matters because many of the most impactful production decisions aren't scheduling decisions. Examples: which customer orders to accept when capacity is short. Which product mix to run when margin varies by SKU. Which campaigns to run to minimize total setup. Which machines to dedicate to which products. How to operate a continuous process (temperatures, flow rates, residence times) to maximize yield. These decisions are upstream of scheduling scheduling answers "given these decisions, what's the schedule?" while optimization asks "what should these decisions be?"
The financial impact of getting optimization right is typically larger than getting scheduling right. A scheduled plant that's running suboptimal product mix may have a great schedule but be making the wrong things. A scheduled plant running suboptimal campaign structure has the right products but inefficient changeover patterns. The schedule can be optimized within a fixed structure; production optimization improves the structure itself.
When capacity is constrained and not all orders can be made, which orders should be accepted? Product mix optimization weighs orders by margin contribution per bottleneck hour (the appropriate metric when capacity is the binding constraint) and recommends which orders to prioritize.
Example: A plant with 1,000 bottleneck hours per month receives orders totaling 1,400 hours. Product A contributes $80 margin per hour, Product B contributes $60, Product C contributes $40. Optimization recommends accepting all Product A orders, then Product B until capacity fills, declining or deferring Product C. This often differs from FIFO order processing, which would accept whichever orders arrived first regardless of profitability.
For process operations where changeovers between product families involve significant cost (cleaning, equipment reconfiguration), campaign planning decides how to group products into runs. The decision balances changeover cost against inventory cost long campaigns minimize changeovers but build inventory; short campaigns minimize inventory but increase changeovers.
Example: A food processor runs 8 product variants on a shared line. Each changeover costs 4 hours of cleanout. Campaign planning optimizes the cycle: how often to run each variant, in what sequence, with what campaign length. The optimal cycle differs from intuitive sequencing it accounts for cycle stock requirements, demand patterns, and changeover matrix.
When multiple resources can produce the same products, which products should run on which resources? Resource allocation optimization considers resource efficiency variations (some machines are faster on some products), maintenance schedules, operator skills, and tool availability.
Example: A discrete manufacturer has three parallel CNC machines. Machine 1 is fastest on small parts; Machine 2 is most efficient on aluminum; Machine 3 handles complex setups well. Allocation optimization assigns work to maximize total throughput rather than evenly loading the three machines.
For continuous and process operations, the operating parameters (temperatures, pressures, flow rates, residence times) affect both yield and throughput. Process parameter optimization finds the parameter combinations that maximize the right objective often yield in pharma and chemicals, throughput in commodity processing.
When a customer requests an order, can it be promised? At what date? Order-acceptance optimization considers the current schedule, the impact of adding the new order on other commitments, and the margin implications. Bad order-acceptance promises orders that can't be made, creating downstream chaos. Good order-acceptance protects existing commitments while accepting profitable new business.
