Forecasting & Capacity Planning: Methods, Models & Real-World Fit

TAKEAWAYS:

• Forecasting is not guesswork—it uses time-series, causal, and judgmental methods to predict demand with statistical rigor.
• Capacity planning bridges supply and demand by matching resources (people, machines, space) to forecasted workloads, avoiding over/underutilization.
• Lead time and safety stock are critical levers in inventory-driven forecasts (e.g., NTC’s network expansion).
• Aggregate planning (e.g., Daraz’s seasonal hiring) balances cost and service levels over medium-term horizons.
• Capacity bottlenecks (e.g., Pathao’s driver shortages) can be identified using Little’s Law or queueing theory (Unit 11).
• Overcapacity (e.g., Ncell’s unused towers) wastes money; undercapacity (e.g., NEPSE’s trading halts) loses revenue.

1. Forecasting: The Art of Predicting Demand

Forecasting is the quantitative process of estimating future demand for products/services. Poor forecasts lead to stockouts (e.g., Daraz running out of laptops during exams) or excess inventory (e.g., banks overissuing credit cards). Businesses use forecasts to:

• Plan production (e.g., Himalayan Java’s coffee roasting schedules).
• Allocate budgets (e.g., NTC’s fiber-optic rollout).
• Optimize supply chains (e.g., Kathmandu’s vegetable market supply).

1.1 Types of Forecasting Methods

Forecasting methods are classified by data type and horizon:

Time-series methods (left) vs. causal methods (right) for Daraz’s demand forecast

1.2 Key Time-Series Models

A. Moving Average (MA)

• How it works: Averages demand over n periods to smooth out short-term fluctuations.
  • Formula: $${\text{MA}}_t=\frac{D_{t-1}+D_{t-2}+\cdots +D_{t-n}}{n}$$
  • Example: NTC forecasts monthly data usage by averaging the last 3 months.
• When to use: Stable demand (e.g., Himalayan Java’s daily tea sales).

B. Exponential Smoothing (ES)

• How it works: Gives more weight to recent data (α = smoothing factor, 0 < α < 1).
  • Formula: $$F_{t+1}=\alpha D_t+(1-\alpha )F_t$$
  • Example: A bank forecasts loan defaults using ES (α = 0.3).

C. Trend Analysis

• How it works: Fits a linear trend to historical data.
  • Formula: $$F_t=a+bt$$
  • Example: Daraz’s order volume grows by 20% annually (b = 0.2).

D. Seasonality Adjustment

• How it works: Decomposes data into trend, seasonality, and residual.
  • Example: Ncell’s data usage spikes in December (festive season).

1.3 Evaluating Forecast Accuracy

Metrics to judge forecasts:

• Mean Absolute Error (MAE): $$\text{MAE}=\frac{1}{n}\sum _{t=1}^n|F_t-D_t|$$
• Mean Squared Error (MSE): $$\text{MSE}=\frac{1}{n}\sum _{t=1}^n(F_t-D_t{)}^2$$
• Example: If Daraz’s MAE for laptop sales is 5 units, it expects a 5-unit error on average.

Worked Example: NTC’s Data Usage Forecast NTC’s last 4 months of data (in GB): 100, 120, 110, 130.

1. Moving Average (n=2):
   • Forecast for Month 5 = (120 + 110)/2 = 115 GB.
2. Exponential Smoothing (α=0.5):
   • $F_5=0.5\times 130+0.5\times 115=122.5$ GB.
3. Which is better?
   • Actual Month 5 demand = 140 GB.
   • MAE for MA = |115–140| = 25 GB.
   • MAE for ES = |122.5–140| = 17.5 GB → ES is more accurate.

1.4 Judgmental Forecasting: When Data Fails

Used for new products (e.g., NEPSE’s crypto trading forecasts) or unique events (e.g., Pathao’s monsoon surge demand).

• Methods:
  • Delphi Technique: Anonymous expert surveys (e.g., Himalayan Java’s new flavor launch).
  • Salesforce Estimates: Frontline staff opinions (e.g., Daraz’s warehouse managers).
  • Market Research: Surveys (e.g., Ncell’s 5G adoption polls).

flowchart TD
    A["Round 1: Experts submit estimates"] --> B["Anonymize & aggregate"]
    B --> C["Round 2: Experts revise based on group data"]
    C --> D["Repeat until consensus"]

2. Capacity Planning: Matching Supply to Demand

Capacity planning ensures resources (people, machines, space) align with forecasted demand. Poor planning leads to:

• Overcapacity: Ncell’s unused 5G towers (wasted investment).
• Undercapacity: NEPSE’s trading halts during peak hours (lost revenue).

2.1 Types of Capacity

2.2 Capacity Planning Approaches

2.3 Aggregate Planning: Balancing Cost and Service

Aggregate planning aligns production, inventory, and workforce over 3–18 months. Methods:

• Level Strategy: Constant output, adjust inventory (e.g., Himalayan Java’s daily roasting).
• Chase Strategy: Match production to demand (e.g., Ncell hiring/training staff seasonally).
• Hybrid Strategy: Mix of level and chase (e.g., Daraz using part-time workers in peak seasons).

Worked Example: Daraz’s Diwali Aggregate Plan

• Forecast: 50% more orders than usual (10,000 vs. 6,500).
• Options:
  1. Level Strategy:
     • Produce 10,000 units daily (cost: overtime pay = ₹50,000).
  2. Chase Strategy:
     • Hire 20 temporary workers (cost: ₹40,000).
  3. Hybrid:
     • Produce 8,000 units daily + hire 10 temps (cost: ₹35,000).
• Best choice: Hybrid (lowest cost).

2.4 Little’s Law: The Capacity-Demand Link

Little’s Law relates inventory (I), throughput rate (R), and flow time (T): $$I=R\times T$$

• Example: If Pathao’s average order flow time (T) is 10 minutes and throughput (R) is 60 orders/hour, then: $$I=60\times \frac{10}{60}=10\text{ orders in the system}$$
  • Implication: If T increases (e.g., due to driver shortages), I (waiting orders) rises → bottleneck.

2.5 Capacity Bottlenecks and Utilization

• Bottleneck: The resource with the lowest capacity relative to demand (e.g., Ncell’s tower maintenance crew).
• Utilization Rate: $$\text{Utilization}=\frac{\text{Actual Output}}{\text{Effective Capacity}}\times 100\%$$
  • Example: If Daraz’s warehouse processes 6,500 units/day with an effective capacity of 8,000: $$\text{Utilization}=\frac{6,500}{8,000}\times 100\%=81.25\%$$

flowchart TD
    A["Step 1: Measure Throughput"] --> B["Step 2: Identify Bottleneck (e.g., Pathao driver shortage)"]
    B --> C["Step 3: Optimize Bottleneck"]
    C --> D["Step 4: Re-measure Utilization"]
    D -->|"Loop"| A

3. Real-World Applications

## In the real world

1. Daraz’s Seasonal Capacity Planning

   • Idea: Uses aggregate planning to balance workforce and inventory during festivals.
   • How: Hires temporary staff (chase strategy) and pre-orders popular items (inventory buffer).
   • Result: Handles 3x normal traffic without stockouts.

2. NTC’s Network Expansion (Forecasting + Capacity)

   • Idea: Causal forecasting links smartphone sales (independent variable) to data demand (dependent variable).
   • How: Expands towers in high-demand areas (e.g., Kathmandu, Pokhara) using strategic capacity planning.
   • Result: Reduced call drops by 40% in 2023.

3. Pathao’s Surge Pricing (Queueing Theory + Capacity)

   • Idea: Uses Little’s Law to detect bottlenecks (e.g., driver shortages during festivals).
   • How: Dynamically adjusts pricing to balance demand and capacity.
   • Result: Maintains service levels even with 2x normal demand.

4. Case Study: Nabil Bank’s ATM Capacity Planning

Scenario: Nabil Bank’s central branch in Thapathali faces long queues during salary days. Management wants to optimize ATM capacity.

Steps:

1. Forecast Demand:

   • Historical data: 500 withdrawals/day on normal days, 1,200 on salary days.
   • Exponential Smoothing (α=0.6): $$F_{\text{salary day}}=0.6\times 1,200+0.4\times 500=920\text{ transactions}$$
   • Adjustment: Add 20% buffer → 1,104 transactions/day.

2. Capacity Analysis:

   • Current ATMs: 4, each handling 25 transactions/hour → 100/hour or 800/day.
   • Bottleneck: 800 < 1,104 → undercapacity.

3. Solutions:

   • Option 1: Add 2 ATMs (cost: ₹200,000).
   • Option 2: Extend branch hours (cost: ₹150,000).
   • Option 3: Hybrid (1 ATM + extended hours, cost: ₹175,000).
   • Best choice: Option 3 (lowest cost).

Outcome: Queue time drops from 30 minutes to 10 minutes.

5. Exam Tip: How to Score Full Marks

1. Structure Your Answer:

   • Start with definitions (e.g., "Forecasting is the process of predicting future demand...").
   • Use formulas (e.g., MAE, Little’s Law) and diagrams (e.g., moving average vs. ES).
   • End with real-world application (e.g., "Like Daraz, banks use aggregate planning to avoid stockouts during salary days").

2. Key Formulas to Memorize:

   • Moving Average: ${\text{MA}}_t=\frac{\sum D_{t-n}}{n}$
   • Exponential Smoothing: $F_{t+1}=\alpha D_t+(1-\alpha )F_t$
   • Little’s Law: $I=R\times T$
   • Utilization: $\frac{\text{Actual Output}}{\text{Effective Capacity}}\times 100\%$

3. Common Pitfalls to Avoid:

   • Ignoring lead time: Always account for supplier delays (e.g., Himalayan Java’s coffee import time).
   • Overlooking seasonality: Assume stable demand unless data shows trends (e.g., Ncell’s data spikes in December).
   • Mixing methods: Use time-series for stable data, causal for external factors (e.g., ads affecting Daraz sales).

4. Case Study Tips:

   • Step 1: Calculate forecasts (use MA or ES).
   • Step 2: Compare actual vs. forecasted demand.
   • Step 3: Identify bottlenecks (use Little’s Law or utilization).
   • Step 4: Propose cost-effective solutions (e.g., hire temps vs. add machines).

5. Visuals That Impress:

   • Draw forecast accuracy graphs (MA vs. ES).
   • Sketch aggregate planning strategies (level vs. chase).
   • Include a Little’s Law diagram showing I, R, and T.

Example Answer Structure for Cases:

1. Forecasting:

   • Used [method] (e.g., exponential smoothing with α=0.4).
   • Calculated forecasted demand = [value].

2. Capacity Analysis:

   • Current capacity = [value], actual demand = [value].
   • Identified bottleneck: [resource] (e.g., driver shortage for Pathao).

3. Recommendation:

   • Solution 1: [option A] with cost = [value].
   • Solution 2: [option B] with cost = [value].
   • Best choice: [option] because [reason].

---