Aggregate Planning and Forecasting in Make-to-Order Production Systems – Exam Notes

Introduction

This paper delves into the critical area of Aggregate Production Planning (APP) and examines how different forecasting techniques impact two key performance indicators: customer-service levels and inventory efficiency within Make-to-Order (MTO) environments. The study specifically uses the context of a multi-stage automotive supplier, providing concrete, real-world relevance, but its findings are designed to be broadly applicable to other MTO businesses.

There are two primary motivations behind this research:

• Increasing operational complexity: Modern industries face rapidly rising product complexity, shorter product life cycles, and notoriously volatile demand patterns. These factors significantly amplify the need for robust and effective mid-term production planning to ensure operational stability and profitability.

• Addressing a common practitioner dilemma: It's notably common for practitioners to bypass the APP stage entirely in their planning hierarchy, opting to jump directly from initial demand forecasts to Master Production Scheduling (MPS). This study rigorously investigates whether this omission is ever strategically justified or if it consistently leads to detrimental outcomes.

Hierarchical Production Planning (HPP)

HPP is a structured approach to production planning that systematically breaks down the overall planning horizon and data granularity into several interdependent layers. This allows for a more manageable and detailed approach to complex manufacturing operations:

1. Aggregate Production Planning (APP):

   • Horizon: Operates over a mid-term horizon, typically spanning $3$ to $18$ months.

   • Granularity: Focuses on balancing overall capacity with aggregated demand at a high level. Decisions are made for product families (e.g., all sedans, all SUVs) and resource groups (e.g., assembly line 1, paint shop 2), rather than individual products or machines.

   • Purpose: Aims to establish production rates, workforce levels, inventory levels, and backlogs to meet anticipated demand while considering available resources.

2. Master Production Scheduling (MPS):

   • Horizon: Operates over a short-term horizon, typically a few weeks.

   • Granularity: Disaggregates the higher-level family plans from APP down to specific, finished individual item quantities. For example, if APP decided on a certain number of "sedans," MPS breaks this down into exact quantities of "Model A sedan, red, with sunroof."

   • Purpose: Acts as the crucial bridge between high-level forecasts and actual customer orders. It dictates what needs to be produced, when, and in what quantities to meet specific order commitments.

3. Material Requirements Planning (MRP) & Shop-floor control:

   • Horizon: Operates over an operational horizon, typically days or hours.

   • Granularity: Deals with the minute details of production.

   • Purpose: Involves lot-sizing (determining optimal batch sizes for production), material netting (calculating exact raw material and component needs), and detailed sequencing and dispatching of individual jobs to specific workstations.

The classical HPP workflow, as meticulously followed in this study, can be conceptually described as a sequential flow of information and decision-making:

• Forecasts provide the initial demand estimates.

• These estimates feed into APP, where an LP optimiser is used to create an aggregate production plan.

• The aggregate plan undergoes disaggregation to translate family-level plans into specific product and quantity targets.

• This disaggregated plan informs the MPS, which incorporates a frozen zone (a period where the schedule is fixed and cannot be changed).

• MPS outputs then drive MRP, which handles netting (calculating exact component needs by subtracting on-hand and on-order quantities from gross requirements) and lot-sizing (determining production or purchase quantities).

• Finally, production control manages the actual execution on the shop floor, where a simulation model introduces stochasticity (random variability) related to real-world operational issues like machine set-ups, unexpected breakdowns, and maintenance requirements.

Make-to-Order (MTO) vs. Make-to-Stock (MTS)

The fundamental distinction between MTO and MTS environments profoundly influences how planning and forecasting are approached:

• Make-to-Stock (MTS):

  • Primary buffer: Relies heavily on finished goods inventory to absorb demand fluctuations and ensure immediate customer availability.

  • Dominant driver: Sales forecasts are the absolute cornerstone of planning, as production occurs before actual customer orders are received.

• Make-to-Order (MTO):

  • Primary driver: Real customer orders (actual confirmed demand) become the dominant input for lower-level planning activities (like MPS and MRP).

  • Role of forecasts: Even in MTO, forecasts remain crucial for upper-level capacity planning (APP). They serve to anticipate future demand trends, allowing the firm to proactively secure sufficient capacity (e.g., labor, machinery, raw materials) to avoid bottlenecks later on when real orders arrive. Without accurate aggregate forecasts, an MTO firm risks having insufficient capacity to fulfill incoming orders, leading to long lead times and missed delivery promises.

Mathematical APP Model

The study employs a classical Linear Programming (LP) model for Aggregate Production Planning. This model aims to optimise resource allocation and production to meet demand while minimising costs. Here are the key components:

Sets and aggregated parameters:

• $P$: Represents the set of product families (e.g., types of car bodies, engine families).

• $T = {1, \dots, n}$: Represents the set of planning periods (e.g., months, quarters) within the planning horizon.

• $J$: Represents the set of machine groups or resource types (e.g., stamping presses, welding robots, paint booths).

• $d_{pt}$: Denotes the forecast demand for product family $p$ in period $t$. This is the anticipated volume the company expects to sell.

• $K_{jt}$: Denotes the available capacity for machine group $j$ in period $t$. This is the maximum output that resource can provide.

• $a_{pj}$: Represents the capacity coefficient, indicating the amount of capacity of machine group $j$ required to produce one unit of product family $p$. This links product families to the resources they consume.

• $h_p$: Represents the holding cost per unit of product family $p$ per period. This includes costs such as storage, insurance, and obsolescence.

• $m_p$: Represents the back-order cost per unit of product family $p$ per period. This includes costs associated with delayed deliveries, lost goodwill, and expedited shipping.

Decision variables per family $p$ and period $t$:

• $x_{pt}$: The production quantity of product family $p$ in period $t$. This is what the model determines to be produced.

• $l_{pt}$: The inventory level of product family $p$ at the end of period $t$. This represents units held in stock.

• $n_{pt}$: The back-order level of product family $p$ at the end of period $t$. This represents unmet demand carried over to future periods.

Objective function:

$\min \sum<em>{p\in P}\sum</em>{t\in T} \big( h<em>p l</em>{pt}+ m<em>p n</em>{pt}\big)$

This objective function aims to minimise the total aggregate cost over the entire planning horizon. This total cost is a sum of two components: the total holding cost for any inventory carried and the total back-order cost incurred from unmet demand. The model seeks to find a balance between these two conflicting costs.

Subject to (constraints):

1. Capacity constraint:
   $\sum<em>{p\in P} a</em>{pj} x<em>{pt} \le K</em>{jt} \quad \forall j\in J,\; t\in T$
   This constraint ensures that the total capacity consumed by all product families produced in period $t$ on machine group $j$ does not exceed the available capacity for that machine group in that period. Essentially, you can't produce more than your machines allow.

2. Inventory balance equation:
   l{pt}-n{pt}=l{p,t-1}-n{p,t-1}+x{pt}-d{pt} \quad \forall p\in P,\; t>1
   This is the mass balance equation for inventory and back-orders. For each product family $p$ and period $t$ (after the first period), the net inventory position at the end of the current period ($l<em>{pt}-n</em>{pt}$) must equal the net inventory position from the previous period ($l<em>{p,t-1}-n</em>{p,t-1}$) plus the production in the current period ($x<em>{pt}$) minus the demand in the current period ($d</em>{pt}$). This ensures that inventory and backlogs are accurately tracked over time.

3. Non-negativity constraint:
   $x<em>{pt},\;l</em>{pt},\;n_{pt}\ge0$
   This standard constraint simply states that all production quantities, inventory levels, and back-order levels must be non-negative (cannot be less than zero).

The LP model is solved to optimality using the XpressMP solver, which is a powerful commercial optimization software. The plan is then updated periodically on a rolling horizon basis. This means that as new demand forecasts become available and actual production progresses, the planning window moves forward; the oldest period drops off, and a new future period is added, with the LP model being re-solved to adapt to updated information.

Literature Highlights

This study builds upon established theoretical foundations and addresses contemporary debates in production planning:

• Original HPP Framework: The concept of Hierarchical Production Planning was originally systematised by Hax & Meal (1975). Subsequent research has significantly advanced HPP by focusing on enhancing its robustness under uncertainty (making plans less susceptible to unexpected changes), incorporating fuzzy sets (to handle imprecise or subjective data), and integrating capabilities like workforce flexibility (allowing for adjustments in labor capacity).

• Impact of Forecasting Errors: Research by Enns (2002) and Xie et al. (2004) clearly demonstrates that forecasting errors have direct and significant negative impacts on key performance indicators such as total cost, customer service levels, and the stability of production schedules. Inaccurate forecasts can lead to overproduction (and excess inventory) or underproduction (and stockouts/lost sales). This underscores the importance of effective forecasting techniques discussed in this paper.

• Practitioner Skepticism: There has been a notable skepticism among practitioners regarding the practical relevance and utility of formal APP, as highlighted by Buxey (2005). Many perceive it as overly complex or disconnected from real-world shop-floor dynamics. This study directly addresses this perception by rigorously investigating APP's value in a simulated stochastic environment. Crucially, few quantitative studies have previously managed to combine the analytical power of APP models with the realistic complexity of stochastic shop-floor simulation, making this research particularly valuable.

Forecasting Techniques Considered

To evaluate the impact of demand forecasting on APP and overall performance, the study considers four distinct forecasting approaches:

1. Simple Moving Average (SMA):

   • Methodology: Calculates the mean (average) of the actual demand from the last $n=4$ months. All past data points within this window are given equal weight.

   • Characteristics: It's straightforward and easy to compute. It tends to smooth out random fluctuations in demand but can lag significantly behind actual trends or sudden shifts.

2. Exponential Smoothing (ES):

   • Methodology: Uses a weighted average of the most recent actual demand and the previous forecast. The formula is: $\hat d<em>t = \alpha d</em>{t-1} + (1-\alpha)\hat d<em>{t-1}$ where $\hat d</em>t$ is the new forecast, $d<em>{t-1}$ is the actual demand from the previous period, and $\hat d</em>{t-1}$ is the forecast from the previous period. The smoothing factor $\alpha$ is set to $0.2$ in this study.

   • Characteristics: Gives more weight to recent data, making it more responsive to changes than SMA. However, a low $\alpha$ (like $0.2$) means it reacts slowly to sharp demand spikes or drops. It's good for stable demand patterns without strong trends or seasonality.

3. Descriptive / Judgmental Forecast (DF):

   • Methodology: This forecast is derived from historical demand series that have been deliberately distorted by systematic noise. This noise is introduced to reflect the kind of managerial intuition or biases that might influence a human forecaster's assessment of future trends, often leading to overreactions to perceived market shifts.

   • Characteristics: Simulates scenarios where forecasts are influenced by qualitative factors or human judgment, which can sometimes introduce inaccuracies or amplify minor fluctuations into significant perceived trends.

4. No APP (NAP):

   • Methodology: This is not a forecasting technique in itself, but rather a control policy or a quasi-myopic production strategy. Under NAP, capacity is NOT pre-allocated based on mid-term forecasts. Instead, production is triggered and resources are allocated only by actual customer orders as they arrive.

   • Characteristics: Represents the scenario where a firm effectively skips the APP layer and operates on a purely responsive, order-driven basis. While seemingly agile, it risks encountering capacity constraints and bottlenecks when actual demand exceeds immediately available resources.

Simulation–Optimisation Framework

To rigorously test the various forecasting techniques and planning strategies, the study employed a sophisticated simulation–optimisation framework:

• Platform Integration: The core of the framework is AnyLogic 6.5, a powerful discrete-event simulation software platform. This is integrated with XpressMP, the commercial LP optimiser used for solving the APP model iteratively.

• Simulated Plant Environment: The study models a multi-stage automotive supplier plant with the following characteristics:

  • Production Stages: Consists of 5 distinct production stages, simulating a typical manufacturing flow (e.g., stamping, welding, painting, assembly, final testing).

  • Products: Produces 12 different finished products which require 16 distinct components.

  • Bill of Materials: Features a divergent bill of materials structure, meaning a few common components can eventually be assembled into a wide variety of finished products. (The detailed structure is presented in Fig. 2 of the original article).

  • Realistic Operational Dynamics: The simulation incorporates real-world complexities such as machine set-up times (time needed to switch production from one product to another), processing times, potential machine failures (breakdowns), and scheduled maintenance periods. This introduces the crucial element of stochasticity.

• Performance Metrics: The effectiveness of each planning strategy is evaluated based on two key performance metrics:

  • Service Level ($\beta$): Defined as the ratio of on-time deliveries to total deliveries. A higher service level indicates better customer satisfaction and reliability.

  • Inventory Level: Calculated as the daily on-hand sum of inventory across the entire planning horizon. Lower inventory levels generally indicate greater efficiency and reduced carrying costs.

• Parameter Tuning: Critical operational parameters, such as lot sizes (how many units to produce in one batch), lead-times (time from order placement to delivery), and Work-In-Process (WIP) safety stock levels, were carefully calibrated to realistic values. This tuning was achieved using a sophisticated metaheuristic optimisation algorithm called Variable Neighbourhood Search (Gansterer et al. 2013), ensuring that the simulation reflects an optimally managed system under each scenario.

Demand Scenarios Tested

The study tested each forecasting and APP strategy across four distinct and challenging demand scenarios over a $36$-month simulation horizon (with the first $6$ months serving as a warm-up period to allow the system to reach a steady state). These scenarios represent different demand profiles a manufacturing firm might encounter:

1. High Demand:

   • Characteristic: Represents a scenario with steady and consistently high volume orders.

   • Capacity Utilisation: The firm operates at approximately $95\%$ capacity utilisation, indicating very tight capacity and little slack.

   • Challenge: Managing consistently high throughput and avoiding bottlenecks.

2. Low Demand:

   • Characteristic: While order frequency might be similar to High demand, the quantities per order are very small, leading to an overall low volume.

   • Capacity Utilisation: The firm operates at only approximately $10\%$ capacity utilisation, indicating significant surplus capacity.

   • Challenge: Managing idle capacity and ensuring cost-effectiveness despite low volumes.

3. Few Demand:

   • Characteristic: Characterised by intermittent ordering, meaning long periods without orders followed by bursts. Average utilization is moderate.

   • Capacity Utilisation: Averages around $60\%$ utilisation.

   • Challenge: Dealing with sporadic demand, which can make planning difficult and lead to either excess inventory during lulls or stockouts during bursts.

4. Peaks Demand:

   • Characteristic: This is the most volatile scenario, featuring a baseline of random low months punctuated by sharp, unpredictable spikes in demand.

   • Capacity Utilisation: Averages around $60\%$ utilisation, but with extremely high volatility due to the sudden peaks.

   • Challenge: The primary challenge is reacting quickly and efficiently to unexpected demand surges without over-investing in capacity or accumulating excessive inventory during quiet periods.

In all scenarios, the customer-requested lead-time (the time customers expect their orders to be delivered) followed a Lognormal distribution with a mean ($\mu$) of $30$ days and a standard deviation ($\sigma$) between $0$ and $1$. This realistic distribution adds another layer of complexity to the service level analysis.

Aggregate Planning Strategies vs. Demand (Results)

The study conducted $10$ replications for each scenario, with highly consistent results (coefficients of variation $\le 4\%$). The aggregate results and interpretations are crucial for managerial decision-making:

Detailed Interpretation:

• Surplus Capacity Scenario (Low Demand):

  • Finding: When a firm consistently operates with significant surplus capacity (e.g., 10% utilisation), the study conclusively shows that proactive APP is detrimental. The No APP (NAP) policy surprisingly delivers the best service level ($\approx97\%$) in this context.

  • Explanation: In such environments, committing to aggregate production plans based on forecasts (as APP does) leads to unnecessary stock accumulation and surprisingly degrades service levels. This is because the ample capacity means the firm can respond efficiently to actual orders as they arrive without pre-committing resources. APP in this case simply generates forecasts that bind resources that aren't truly needed, leading to idle inventory and potentially slower response to real, immediate demand compared to a purely order-driven approach. For instance, MA-based APP significantly increased inventory by $38\%$ and still lost $2.79\%$ service level compared to NAP.

• Tight Capacity or Volatile Demand Scenarios (High, Few, Peaks):

  • Finding: When capacity is tight (High demand) or demand is highly volatile (Few, Peaks), forecasting-driven APP is absolutely essential for maintaining high service levels. In these situations, Simple Moving Average (SMA) generally outperforms other forecasting techniques.

  • Detailed Breakdown:

    • High Demand: With near-full capacity, omitting APP (NAP) caused a substantial $7.55\%$ drop in service level. This indicates that without proactive planning, the system quickly gets overwhelmed. SMA, ES, and DF all perform well, achieving over $99\%$ service, but SMA is superior in inventory efficiency, requiring at least $11\%$ less stock than ES to achieve similar service. This implies SMA's smoothing effect is beneficial in stable, high volume environments as it avoids unnecessary inventory build-up.

    • Few Demand: SMA is again the best performer for service (98%). Omitting APP (NAP) leads to a $3.32\%$ service level loss and surprisingly a $10\%$ reduction in inventory. While lower inventory might seem good, the concurrent service loss suggests that the firm is missing opportunities to serve intermittent high demand quickly. Exponential Smoothing (ES), despite being responsive, resulted in higher stock levels with lower service, indicating it struggled to manage the intermittent nature effectively.

    • Peaks Demand: This scenario is the most challenging due to high volatility. SMA emerges as the clear winner in service (82%). Omitting APP (NAP) results in a severe $6.97\%$ service level loss and an astonishing $1,290\%$ increase in inventory! This highlights the disastrous consequences of reactive planning during highly volatile periods. Without proactive capacity allocation, the firm attempts to chase the peaks, leading to massive inventory overshoots in anticipation of demand that may not materialise immediately. The Descriptive/Judgmental Forecast (DF) performs the worst, leading to the largest service loss ($-7.85\%$), likely because managerial intuition over-amplifies or misinterprets the peaks, leading to poor capacity allocation decisions.

• Specific Forecasting Technique Insights:

  • Exponential Smoothing (ES): Tends to add unnecessary inventory in smooth, high demand environments like the 'High' scenario. Paradoxically, it also under-performs in highly volatile 'Peaks' scenarios, likely because its slow response (due to the low $\alpha$) cannot keep pace with sharp changes, leading to either stockouts or over-corrections that build inventory.

  • Judgmental Forecast (DF): Demonstrates that human-biased or purely descriptive forecasts can easily exaggerate volatility patterns, leading to misallocation of capacity, particularly evident in the 'Peaks' scenario where it performed worst.

Managerial Recommendations

Based on these robust simulation results, the study offers clear, actionable recommendations for managers in MTO environments:

1. Diagnose Capacity Slack Accurately: Before deciding on an APP strategy, it is paramount for managers to first determine their prevailing capacity situation:

   • If there is persistent surplus capacity: The data suggests managers should consider skipping formal APP. Instead, they should rely on a highly responsive, order-driven execution system. In such scenarios, proactive planning can tie up resources unnecessarily and hinder agility.

   • If capacity is scarce or tight: Managers should definitively implement APP. Proactive planning is crucial to balancing demand and supply, preventing bottlenecks, and ensuring customer service when resources are constrained.

2. For Stable or Moderately Volatile Demand, Prefer SMA-based APP: For demand patterns that are either consistent (like 'High') or exhibit moderate, less extreme volatility (like 'Few'), an APP strategy that employs Simple Moving Average (SMA) forecasting is highly recommended. This approach consistently achieves a strong balance between high service levels and moderate inventory levels, signifying efficient resource usage.

3. ES is Acceptable, But Be Mindful of Inventory: Exponential Smoothing (ES) with a low alpha value ($\alpha=0.2$ in this study) can provide acceptable service levels, especially in relatively stable environments. However, managers should be aware that it may lead to inflated inventory levels compared to SMA due to its inherent smoothing and slower reaction to true demand changes.

4. Exercise Caution with Purely Descriptive or Biased Forecasts: Managers should avoid relying solely on purely descriptive or human-biased judgmental forecasts unless the underlying future trends are exceptionally well-understood, reliably quantifiable, and objectively verified. Such forecasts can introduce significant errors and lead to poor planning decisions, particularly in volatile market conditions.

5. Beware of Pure Chase Strategies: A

Introduction

This paper delves into the critical area of Aggregate Production Planning (APP) and examines how different forecasting techniques impact two key performance indicators: customer-service levels and inventory efficiency within Make-to-Order (MTO) environments. The study specifically uses the context of a multi-stage automotive supplier, providing concrete, real-world relevance, but its findings are designed to be broadly applicable to other MTO businesses.

There are two primary motivations behind this research:

• Increasing operational complexity: Modern industries face rapidly rising product complexity, shorter product life cycles, and notoriously volatile demand patterns. These factors significantly amplify the need for robust and effective mid-term production planning to ensure operational stability and profitability.

• Addressing a common practitioner dilemma: It's notably common for practitioners to bypass the APP stage entirely in their planning hierarchy, opting to jump directly from initial demand forecasts to Master Production Scheduling (MPS). This study rigorously investigates whether this omission is ever strategically justified or if it consistently leads to detrimental outcomes.

Mathematical APP Model

The study employs a classical Linear Programming (LP) model for Aggregate Production Planning. This model aims to optimise resource allocation and production to meet demand while minimising costs. Here are the key components:

Sets and aggregated parameters:

• $P$: Represents the set of product families (e.g., types of car bodies, engine families).

• $T = {1, \dots, n}$: Represents the set of planning periods (e.g., months, quarters) within the planning horizon.

• $J$: Represents the set of machine groups or resource types (e.g., stamping presses, welding robots, paint booths).

• $d_{pt}$: Denotes the forecast demand for product family $p$ in period $t$. This is the anticipated volume the company expects to sell.

• $K_{jt}$: Denotes the available capacity for machine group $j$ in period $t$. This is the maximum output that resource can provide.

• $a_{pj}$: Represents the capacity coefficient, indicating the amount of capacity of machine group $j$ required to produce one unit of product family $p$. This links product families to the resources they consume.

• $h_p$: Represents the holding cost per unit of product family $p$ per period. This includes costs such as storage, insurance, and obsolescence.

• $m_p$: Represents the back-order cost per unit of product family $p$ per period. This includes costs associated with delayed deliveries, lost goodwill, and expedited shipping.

Decision variables per family $p$ and period $t$:

• $x_{pt}$: The production quantity of product family $p$ in period $t$. This is what the model determines to be produced.

• $l_{pt}$: The inventory level of product family $p$ at the end of period $t$. This represents units held in stock.

• $n_{pt}$: The back-order level of product family $p$ at the end of period $t$. This represents unmet demand carried over to future periods.

Objective function:

$\min \sum<em>{p\in P}\sum</em>{t\in T} \big( h<em>p l</em>{pt}+ m<em>p n</em>{pt}\big)$

This objective function aims to minimise the total aggregate cost over the entire planning horizon. This total cost is a sum of two components: the total holding cost for any inventory carried and the total back-order cost incurred from unmet demand. The model seeks to find a balance between these two conflicting costs.

Subject to (constraints):

1. Capacity constraint:
   

   $\sum<em>{p\in P} a</em>{pj} x<em>{pt} \le K</em>{jt} \quad \forall j\in J,\; t\in T$
   

   This constraint ensures that the total capacity consumed by all product families produced in period $t$ on machine group $j$ does not exceed the available capacity for that machine group in that period. Essentially, you can't produce more than your machines allow.

2. Inventory balance equation:
   

   l{pt}-n{pt}=l{p,t-1}-n{p,t-1}+x{pt}-d{pt} \quad \forall p\in P,\; t>1
   

   This is the mass balance equation for inventory and back-orders. For each product family $p$ and period $t$ (after the first period), the net inventory position at the end of the current period ($l<em>{pt}-n</em>{pt}$) must equal the net inventory position from the previous period ($l<em>{p,t-1}-n</em>{p,t-1}$) plus the production in the current period ($x<em>{pt}$) minus the demand in the current period ($d</em>{pt}$). This ensures that inventory and backlogs are accurately tracked over time.

3. Non-negativity constraint:
   

   $x<em>{pt},\;l</em>{pt},\;n_{pt}\ge0$
   

   This standard constraint simply states that all production quantities, inventory levels, and back-order levels must be non-negative (cannot be less than zero).

The LP model is solved to optimality using the XpressMP solver, which is a powerful commercial optimization software. The plan is then updated periodically on a rolling horizon basis. This means that as new demand forecasts become available and actual production progresses, the planning window moves forward; the oldest period drops off, and a new future period is added, with the LP model being re-solved to adapt to updated information.

Aggregate Planning Strategies vs. Demand (Results)

The study conducted $10$ replications for each scenario, with highly consistent results (coefficients of variation $\le 4\%$). The aggregate results and interpretations are crucial for managerial decision-making:

Detailed Interpretation:

• Surplus Capacity Scenario (Low Demand):

  • Finding: When a firm consistently operates with significant surplus capacity (e.g., 10% utilisation), the study conclusively shows that proactive APP is detrimental. The No APP (NAP) policy surprisingly delivers the best service level ($\approx97\%$) in this context.

  • Explanation: In such environments, committing to aggregate production plans based on forecasts (as APP does) leads to unnecessary stock accumulation and surprisingly degrades service levels. This is because the ample capacity means the firm can respond efficiently to actual orders as they arrive without pre-committing resources. APP in this case simply generates forecasts that bind resources that aren't truly needed, leading to idle inventory and potentially slower response to real, immediate demand compared to a purely order-driven approach. For instance, MA-based APP significantly increased inventory by $38\%$ and still lost $2.79\%$ service level compared to NAP.

• Tight Capacity or Volatile Demand Scenarios (High, Few, Peaks):

  • Finding: When capacity is tight (High demand) or demand is highly volatile (Few, Peaks), forecasting-driven APP is absolutely essential for maintaining high service levels. In these situations, Simple Moving Average (SMA) generally outperforms other forecasting techniques.

  • Detailed Breakdown:

    • High Demand: With near-full capacity, omitting APP (NAP) caused a substantial $7.55\%$ drop in service level. This indicates that without proactive planning, the system quickly gets overwhelmed. SMA, ES, and DF all perform well, achieving over $99\%$ service, but SMA is superior in inventory efficiency, requiring at least $11\%$ less stock than ES to achieve similar service. This implies SMA's smoothing effect is beneficial in stable, high volume environments as it avoids unnecessary inventory build-up.

    • Few Demand: SMA is again the best performer for service (98%). Omitting APP (NAP) leads to a $3.32\%$ service level loss and surprisingly a $10\%$ reduction in inventory. While lower inventory might seem good, the concurrent service loss suggests that the firm is missing opportunities to serve intermittent high demand quickly. Exponential Smoothing (ES), despite being responsive, resulted in higher stock levels with lower service, indicating it struggled to manage the intermittent nature effectively.

    • Peaks Demand: This scenario is the most challenging due to high volatility. SMA emerges as the clear winner in service (82%). Omitting APP (NAP) results in a severe $6.97\%$ service level loss and an astonishing $1,290\%$ increase in inventory! This highlights the disastrous consequences of reactive planning during highly volatile periods. Without proactive capacity allocation, the firm attempts to chase the peaks, leading to massive inventory overshoots in anticipation of demand that may not materialise immediately. The Descriptive/Judgmental Forecast (DF) performs the worst, leading to the largest service loss ($-7.85\%$), likely because managerial intuition over-amplifies or misinterprets the peaks, leading to poor capacity allocation decisions.

• Specific Forecasting Technique Insights:

  • Exponential Smoothing (ES): Tends to add unnecessary inventory in smooth, high demand environments like the 'High' scenario. Paradoxically, it also under-performs in highly volatile 'Peaks' scenarios, likely because its slow response (due to the low $\alpha$) cannot keep pace with sharp changes, leading to either stockouts or over-corrections that build inventory.

  • Judgmental Forecast (DF): Demonstrates that human-biased or purely descriptive forecasts can easily exaggerate volatility patterns, leading to misallocation of capacity, particularly evident in the 'Peaks' scenario where it performed worst.