Multi-Site, Supply Chain and Sustainable Planning

Introduction

Aggregate production planning (APP) traditionally balances demand, workforce, inventory and production capacity for a single organization over a medium-term horizon. Contemporary supply networks make this perspective incomplete. A firm may operate several plants, contract manufacturers, regional warehouses, suppliers, repair centres and recycling partners. Decisions made at one location affect capacity, inventory, transport, service, cash flow, emissions and social outcomes elsewhere.

Multi-site and sustainable planning therefore treats the supply chain as an interconnected system rather than a collection of independent facilities. The central managerial problem is to coordinate production, procurement, inventory, logistics and capacity while satisfying customer demand and respecting financial, environmental and social constraints. The problem is strategic because it influences facility roles, supplier relationships, resilience investments, technology choices and the long-term configuration of the network.

A useful planning horizon is typically 3–24 months, although strategic capacity commitments may extend beyond it. Plans are usually revised periodically through a rolling-horizon process. The plan should be sufficiently aggregated to support managerial decisions but sufficiently detailed to reveal bottlenecks, interdependence and sustainability trade-offs.

Learning objectives

After studying this chapter, students should be able to:

1. Explain why multi-site aggregate planning requires system-wide coordination.
2. Formulate a multi-period, multi-site planning model involving production, sourcing, inventory, transport and capacity.
3. Evaluate trade-offs among cost, service, resilience, carbon emissions, energy use, circularity and social sustainability.
4. Apply scenario analysis, robust planning and rolling-horizon control to supply-chain uncertainty.
5. Interpret a practical planning case and recommend governance and implementation practices.

1. From single-plant planning to network planning

In a single-site model, production is often represented by a balance equation: opening inventory plus production minus demand equals closing inventory. In a network, the same logic must be applied at each facility and product-location pair, while flows connect plants, suppliers, warehouses, customers and recovery facilities.

The principal sources of complexity are:

• Geographical dispersion: facilities have different costs, capacities, labour rules, lead times, energy mixes and exposure to disruption.
• Shared resources: plants may compete for common suppliers, transport lanes, tooling, engineering capacity or scarce components.
• Material differentiation: products may be made from virgin, recycled or remanufactured material with different yields and quality constraints.
• End-to-end effects: a low-cost production decision can create expensive freight, excess inventory or downstream emissions.
• Uncertainty: demand, supply reliability, energy prices, transport availability and regulatory conditions vary.
• Multiple objectives: the least-cost plan may not be the most resilient, low-carbon or socially responsible plan.

Consequently, local optimization can be globally harmful. A plant that minimizes its own unit cost may produce distant demand, increasing transport emissions and inventory. Conversely, concentrating production in one efficient plant may reduce cost but increase vulnerability to disruption.

The network planning hierarchy

A coherent planning system links three levels:

The levels should not be optimized independently. Tactical plans must respect strategic capacities and should provide feasible inputs to operational scheduling. Operational data, such as actual yields and supplier reliability, should feed back into aggregate planning.

2. A systems model for multi-site APP

A practical model normally distinguishes suppliers, production sites, distribution centres, customers and recovery facilities. Let:

• i index products;
• s index production sites;
• w index warehouses or demand regions;
• t index time periods;
• p index suppliers;
• r index recovery facilities.

Material balance

For a production site, a basic product balance is:

where I is closing inventory, Q is production, R is receipts of recovered or transferred material, U is internal use or shipments to warehouses, and X is other outflow. All quantities should use the same unit, such as tonnes, units or pallets, and the equation assumes inventory is measured at the same point in the process.

At a warehouse serving demand D_{w,i,t], the balance may be written:

Need avoid malformed. Use correct equation with inbound shipment Y and demand D and backorders B perhaps.

This form permits backorders B. If backorders are prohibited, set B_w,i,t=0. If lost sales are used instead, lost demand must be represented explicitly and penalized appropriately.

Capacity and workforce

Production capacity may depend on regular time, overtime and subcontracting:

Here a_s,i is output per labour-hour or effective capacity coefficient, H is available regular labour-hours, O is overtime production and C is subcontracted production. This expression is appropriate only when the production relationship is linear over the planning range. If setups, minimum batches or learning effects matter, a mixed-integer or nonlinear representation is more appropriate.

A shared capacity constraint can be expressed as:

where h_s,i is required resource time per unit and K_s,t is available resource time. The left and right sides must have identical units, such as machine-hours.

Procurement and logistics

Supplier allocations must respect supplier capacity, minimum commitments, qualification and lead time. A shipment from site s to warehouse w may be constrained by transport capacity:

where v_i is volume per unit and L is available transport volume. Freight cost should reflect distance, mode, weight, volume, fuel surcharge and potentially emissions charges.

Objective function

A cost-minimizing deterministic model can use:

The terms represent production, procurement, inventory holding, transportation, labour adjustment, shortage, setup and carbon-related costs. Each term is a sum over relevant products, facilities and periods. Units should be monetary units over the planning horizon. Double counting must be avoided: for example, if a carbon tax is included in C^carbon, the same tax should not also be embedded in production cost unless clearly separated.

A cost objective is not automatically sustainable. A firm may instead use a weighted objective:

The weights α, β, γ, δ must be documented, normalized where appropriate, and tested through sensitivity analysis. Weighted sums can hide unacceptable performance on one dimension; therefore, explicit constraints and Pareto analysis are often preferable.

3. Sustainability as a planning dimension

Sustainability should be incorporated into the decision model rather than treated as a report prepared after the plan is selected. Three dimensions are particularly important.

Environmental performance

Relevant measures include greenhouse-gas emissions, energy use, water consumption, waste, local pollutants and material intensity. A simplified emissions calculation is:

where A_a is activity, such as kWh, tonne-kilometres or kilograms of material, and EF_a is the corresponding emission factor. Units must be compatible; for example, kWh multiplied by kg CO₂e/kWh produces kg CO₂e.

For transport, activity can be approximated by:

This tonne-kilometre estimate is useful for planning but may omit empty returns, load factors, refrigeration and fuel-specific effects. Emission factors should come from a credible and current source, and the organizational boundary should be stated.

Energy constraints can be represented as:

where e_s,i is energy per unit and E^avail is available energy. If renewable energy certificates or on-site generation are used, planners must distinguish physical energy, contractual instruments and actual emissions.

Circularity

Circular planning includes reuse, repair, remanufacturing, recycling and recovery. A recovery stream is not equivalent to virgin supply because collection rates, quality, yield and timing are uncertain. If recovery yield is ρ_r, then usable recovered material is:

where ρ_r lies between 0 and 1. The model should also include sorting, processing capacity, contamination and quality restrictions. Circularity metrics should distinguish absolute recycled volume from the percentage of input that is recycled, since percentage improvement can coexist with rising total material use.

Social sustainability

Social constraints may include minimum employment, safe working hours, living-wage policies, supplier audit requirements, inclusion targets and limits on excessive overtime. A simple overtime constraint is:

but responsible planning should also evaluate whether the limit is safe and lawful, not merely mathematically feasible. Supplier selection should consider labour rights, forced-labour risk, freedom of association and grievance mechanisms. Social indicators are often difficult to aggregate into a single score; qualitative due diligence and escalation procedures remain necessary.

4. Resilience and uncertainty

Efficiency and resilience are related but distinct. Efficiency seeks low resource use under expected conditions; resilience seeks the ability to absorb, adapt to and recover from disruption. A network with one low-cost supplier may be efficient but fragile.

Scenario planning

Scenarios may vary demand, supplier availability, energy prices, border delays, extreme weather and regulatory requirements. A stochastic model minimizes expected cost:

where ω indexes scenarios and p_ω is the probability of scenario ω, with probabilities summing to 1. This approach is appropriate when probabilities are credible and decisions can be separated into those made before and after uncertainty is revealed.

A robust model may instead minimize worst-case or bounded-regret performance. Robustness is useful when probabilities are disputed, but excessive conservatism can produce expensive underutilized capacity. Scenario discovery and stress testing should therefore accompany any optimization.

Resilience levers

Important levers include:

• dual or multiple sourcing;
• geographically diversified capacity;
• strategic safety stock for critical items;
• postponement and product commonality;
• qualified alternate materials;
• flexible equipment and cross-trained labour;
• supplier development and visibility;
• emergency logistics contracts;
• cyber and operational continuity controls.

Resilience should be measured by service recovery time, demand served during disruption, time to recover capacity and financial exposure—not simply by counting suppliers.

5. A worked illustrative model

Consider two plants, East and West, supplying one regional distribution centre over one month. Demand is 1,000 units. Opening inventory is 100 units. East has regular capacity of 600 units at ₹40 per unit and a maximum overtime capacity of 200 units at ₹55. West has capacity of 700 units at ₹46. Transport costs to the distribution centre are ₹4 per unit from East and ₹2 per unit from West. Ending inventory is required to be at least 50 units, and shortages are not allowed.

The required production is:

The delivered unit costs are ₹44 for East regular production and ₹48 for West production. East regular production is therefore preferred, followed by West, provided capacities are respected. East can produce 600 regular units; the remaining 350 units should be produced in West. The plan is:

Ending inventory is 100 + 950 − 1,000 = 50 units, satisfying the requirement. Total production and transport cost is:

If West capacity falls by 100 units, East overtime becomes necessary. The revised plan is East regular 600, East overtime 100 and West 250. Its cost is:

The disruption increases cost by ₹700, but the network remains feasible. A complete model would also compare emissions, labour implications and the probability of the disruption. The example demonstrates why local production cost is insufficient: transport changes the effective cost and capacity flexibility has option value.

6. Applied case: illustrative GreenMotion Components

Background

GreenMotion Components, a fictional manufacturer of electric-vehicle thermal-management modules, operates plants in India and Eastern Europe. It buys aluminium housings from three suppliers, ships modules to four markets and receives used modules for remanufacturing. Management historically planned each plant separately using annual volume targets.

Problem

Demand became volatile, the European plant faced energy-price increases, and one supplier experienced repeated port delays. The company also committed to reducing scope 1 and 2 emissions by 30% in four years and increasing remanufactured output. Local plans met their financial targets but caused long-distance emergency freight, high overtime and inconsistent inventory.

Analysis

The planning team mapped bills of material, lead times, capacity, supplier qualification, energy intensity and transport lanes. It identified three bottlenecks: a casting operation in India, testing capacity in Europe and a limited remanufacturing line. It then created demand, disruption and energy-price scenarios. Carbon was calculated for production energy, inbound materials and transport; social criteria included overtime limits and supplier labour-audit status.

Approach and solution

The company implemented a rolling six-month tactical model. It used:

1. a common demand and inventory data set;
2. capacity reservations for critical suppliers;
3. a minimum dual-source rule for high-risk components;
4. a carbon budget by product family;
5. explicit constraints on overtime and approved supplier status;
6. a recovery-yield parameter for returned modules;
7. monthly scenario stress tests;
8. weekly exception management rather than weekly re-optimization of the entire network.

The solution did not minimize cost alone. Management reviewed a Pareto set showing cost, service, emissions and disruption exposure. A slightly more expensive plan was selected because it reduced single-supplier dependence, lowered expedited freight and stayed within the carbon budget.

Outcome

In the first planning cycle, the firm reduced emergency shipments and improved visibility of capacity conflicts. Remanufacturing increased only after collection and testing capacity were expanded; the team rejected an unrealistic target based solely on expected returns. The company also discovered that a renewable-energy purchase would improve reported emissions but would not remove a physical electricity bottleneck, illustrating the need to distinguish accounting instruments from operational capacity.

Learning points

• Network boundaries must include suppliers, logistics and reverse flows where material.
• Sustainability targets require operational variables and credible data.
• Resilience has a cost, but the value of flexibility appears during disruption.
• A Pareto view is often more informative than a single composite score.
• Governance, data ownership and escalation rules are as important as the optimization algorithm.

7. Technology, analytics and governance

Enterprise resource planning, advanced planning systems, transport-management systems, energy meters, supplier portals and product-lifecycle systems provide the data foundation. Digital twins can simulate alternative network configurations, while machine learning can improve demand forecasts, lead-time estimates and anomaly detection. However, predictive accuracy does not guarantee a good decision: an optimization model can produce a precise answer to an incorrectly specified problem.

Artificial intelligence should therefore be governed through data lineage, explainability, human approval, model monitoring and access controls. Planners should test for systematic bias—for example, whether a supplier-risk model penalizes small suppliers because of missing data rather than actual poor performance. Cybersecurity is part of resilience because production and logistics decisions depend on connected systems.

A strong implementation sequence is:

1. define the decision scope and planning calendar;
2. establish a single master-data structure;
3. validate capacities, yields, lead times and emissions factors;
4. build a transparent baseline model;
5. add sustainability and resilience constraints incrementally;
6. pilot one product family or region;
7. compare model recommendations with expert judgment;
8. institutionalize scenario review and post-plan learning.

Governance should specify who owns demand assumptions, capacity data, sustainability factors, supplier risk ratings and final trade-offs. Internal audit should be able to reproduce the logic behind a significant plan.

8. Critical issues and limitations

Multi-objective planning has no universally correct solution. Weight selection can reflect managerial power rather than social value. Carbon factors vary by geography and accounting method. Scope 3 emissions are difficult to allocate without double counting. Social performance is not fully captured by numerical indicators. Scenario probabilities may be unreliable, and robust models can be too conservative.

Aggregation also creates risk. Product families with different routings, yields or customer priorities may not be safely combined. Average capacity can conceal changeover requirements, maintenance and skill constraints. A mathematically feasible plan may be operationally infeasible if it ignores sequence-dependent setups, labour agreements, quality release or transport cut-off times.

Managers should therefore combine optimization with feasibility workshops, sensitivity analysis, stress testing and post-implementation measurement. Every material assumption should have an owner, a refresh frequency and a validation method.

Best-practice checklist

• Plan the network, not isolated facilities.
• Use delivered cost and total-impact measures rather than factory cost alone.
• Separate strategic, tactical and operational decisions while linking their data.
• State system boundaries, units and emission-accounting rules.
• Represent capacity, lead time, yield, minimum lots and qualification constraints explicitly.
• Use scenarios for disruption and energy uncertainty.
• Protect critical flows with resilience policies, not only extra inventory.
• Treat circular material as a supply stream with yield and quality uncertainty.
• Include lawful, safe and auditable social constraints.
• Use rolling horizons and measure forecast, plan and execution error.
• Validate model outputs with local operating expertise.
• Review trade-offs transparently through Pareto analysis and governance forums.

Chapter Summary

Multi-site aggregate production planning coordinates production, procurement, inventory, transport and capacity across a network. Its material-balance and capacity equations must be extended to represent suppliers, warehouses, reverse flows, shared resources and time-phased constraints. The relevant objective is broader than production cost: service, cash, carbon, energy, circularity, labour conditions and disruption exposure all influence plan quality.

Sustainability can be operationalized through emissions and energy equations, carbon budgets, circular-yield constraints and social safeguards. Resilience is supported by multi-sourcing, geographic diversification, flexibility, postponement, strategic inventory and scenario analysis. Stochastic models use expected performance when probabilities are credible; robust models protect against uncertainty when they are not.

The GreenMotion case shows that successful planning depends on data governance, transparent assumptions, cross-functional decision rights and rolling-horizon execution. Optimization is a decision-support capability, not a substitute for operational knowledge, ethical judgment or accountability.

Key Terms

Aggregate production planning: Medium-term coordination of demand, capacity, workforce, inventory and supply.

Multi-site planning: Coordinated planning across geographically separated facilities.

Supply-chain network: Interconnected suppliers, plants, warehouses, customers and recovery entities.

Material balance: Accounting relationship linking opening stock, inflows, outflows and closing stock.

Capacity constraint: Limit on available machine, labour, energy, transport or supplier resources.

Rolling horizon: Planning process that periodically updates and extends the planning horizon.

Delivered cost: Total cost of supplying a destination, including production and logistics.

Carbon budget: Maximum permitted emissions for a product, facility or planning period.

Emission factor: Emissions associated with one unit of an activity or material.

Circularity: Retention of products and materials through reuse, repair, remanufacture or recycling.

Recovery yield: Usable output obtained from collected or returned material.

Resilience: Ability to withstand, adapt to and recover from disruption.

Scenario analysis: Evaluation of plans under alternative future conditions.

Robust optimization: Planning that seeks acceptable performance across uncertainty sets.

Pareto frontier: Set of solutions where improving one objective worsens at least one other.

Scope 3 emissions: Indirect value-chain emissions outside an organization’s direct operations.

Postponement: Delaying product differentiation until demand information is clearer.

Social sustainability: Protection of labour rights, safety, equity and community interests.

Data lineage: Traceable record of a data item’s source, transformation and use.

Review Questions

Short Answer Questions

1. Why can optimizing each plant independently produce an inferior supply-chain plan?
2. Distinguish strategic, tactical and operational planning horizons.
3. What information is required to calculate transport-related emissions?
4. Explain the difference between efficiency and resilience.
5. Why must recovered material be modelled with a yield parameter?

Long Answer Questions

1. Develop and explain a multi-site aggregate planning formulation containing inventory, production, transport and capacity constraints.
2. Critically evaluate weighted-objective and constraint-based approaches to sustainable planning.
3. Explain how scenario analysis and robust planning can support supply-chain resilience.
4. Discuss the data, governance and ethical requirements for using AI in aggregate production planning.
5. Design a rolling-horizon planning process for a multinational manufacturer with reverse logistics.

Analytical / Application Questions

1. A network has two plants with different production costs, transport costs, capacities and emission factors. Construct a plan that minimizes delivered cost subject to demand and a carbon budget, and explain the trade-off.
2. A single-source supplier is cheap but has a 15% disruption probability. Compare dual sourcing, safety stock and excess capacity as resilience responses using scenario analysis.
3. A remanufacturing target is based on expected returns, but actual collection and recovery yields are uncertain. Recommend model changes and performance measures that prevent an unrealistic plan.

Multiple Choice Questions with Answers

Question 1

Which decision is primarily tactical in a multi-site aggregate planning system?

A. Selecting the location of a new plant

B. Allocating monthly production volumes across existing plants

C. Sequencing individual jobs on a machine

D. Designing a long-term corporate ownership structure

Correct Answer: B. Allocating monthly production volumes across existing plants

Question 2

If opening inventory is 100 units, demand is 1,000 units and desired ending inventory is 50 units, required net production is:

A. 850 units

B. 900 units

C. 950 units

D. 1,150 units

Correct Answer: C. 950 units

Question 3

In the emissions equation E = ∑_a A_a EF_a, what does EF_a represent?

A. Available facility capacity

B. Emission factor for activity a

C. Expected financial cost

D. Finished-goods inventory

Correct Answer: B. Emission factor for activity a

Question 4

Which practice most directly reduces dependence on a single disrupted supplier?

A. Single sourcing at a larger volume

B. Dual or multiple sourcing

C. Reducing supplier qualification

D. Eliminating safety stock for critical items

Correct Answer: B. Dual or multiple sourcing

Question 5

A recovered-material yield of 0.80 applied to 500 collected units produces usable material of:

A. 100 units

B. 400 units

C. 500 units

D. 625 units

Correct Answer: B. 400 units

Question 6

A transport constraint based on ∑_i v_iY_i ≤ L requires compatible units for:

A. Product value and labour hours

B. Volume per unit, shipment quantity and transport capacity

C. Carbon factor and monetary cost

D. Demand probability and inventory age

Correct Answer: B. Volume per unit, shipment quantity and transport capacity

Question 7

When credible scenario probabilities are available, a stochastic model commonly minimizes:

A. Expected objective value

B. Only the best-case objective value

C. The number of suppliers regardless of cost

D. Historical average demand only

Correct Answer: A. Expected objective value

Question 8

Which statement best describes a Pareto solution?

A. It is always the least-cost solution

B. It improves every objective simultaneously

C. Improving one objective would worsen at least one other objective

D. It ignores sustainability objectives

Correct Answer: C. Improving one objective would worsen at least one other objective

Question 9

Why is a mathematically feasible plan potentially operationally infeasible?

A. Mathematics cannot represent any capacity

B. It may omit setups, skills, maintenance or quality-release constraints

C. Feasible plans never include inventory

D. Optimization always assumes zero demand

Correct Answer: B. It may omit setups, skills, maintenance or quality-release constraints

Question 10

Which is an appropriate governance practice for AI-supported planning?

A. Remove human review entirely

B. Hide model assumptions from decision makers

C. Maintain data lineage, explainability and model monitoring

D. Use incomplete data without validation

Correct Answer: C. Maintain data lineage, explainability and model monitoring