Starbucks Supply Chain and Comprehensive Forecasting Methodologies

Global Supply Chain Challenges and the Role of Forecasting

Starbucks Corporation represents the largest coffeehouse company globally, maintaining over $17,000$ stores in more than $50$ countries and serving approximately $50 \times 10^{6}$ customers every week. Forecasting demand for such an enterprise is complex because the product line extends far beyond drip-brewed coffee to include espresso-based hot drinks, cold drinks, coffee beans, salads, sandwiches, panini, pastries, snacks, and merchandise like mugs and tumblers.

Many of these products are seasonal or specific to local store environments. Furthermore, Starbucks-branded items like ice cream and coffee are sold in grocery stores worldwide. Managing this requires a single, global logistics system to handle a far-flung supply chain. The process involves:

• Sourcing: Bringing "green" (unroasted) coffee beans from Latin America, Africa, and Asia to the United States and Europe via ocean containers.

• Storage: Trucking beans from ports to six storage sites located at or near roasting plants.

• Distribution: After roasting and packaging, finished products are trucked to regional distribution centers (DCs). These centers range from $200,000$ to $300,000$ square feet.

• Global Infrastructure: Starbucks operates five regional DCs in the United States, two in Europe, and two in Asia. These warehouses store everything from coffee and cappuccino mixes to furniture for retail outlets.

Fundamental Concepts of Forecasting

Forecasts are the foundation of corporate planning and control. They are essential for every significant management decision across various functional areas:

• Finance and Accounting: Forecasts serve as the basis for budgetary planning and cost control.

• Marketing: Relies on sales forecasting for new product planning and sales personnel compensation.

• Production and Operations: Uses forecasts for supplier selection, process selection, capacity planning, facility layout, purchasing, scheduling, and inventory management.

Strategic vs. Tactical Forecasts

• Strategic Forecasts: These are medium- and long-term forecasts used for decisions related to overall strategy and estimating aggregate demand. They occupy a timeframe of three months to greater than two years. They are appropriate for decisions regarding capacity, manufacturing process design, sourcing, and location/distribution design.

• Tactical Forecasts: These are short-term forecasts, typically covering a few weeks or months. The goal is to estimate demand to meet customer lead-time expectations and ensure product availability. They serve as inputs for day-to-day decisions like inventory replenishment and employee scheduling.

Decoupling Points and Buffering

Decoupling points are specific locations within the supply chain where inventory is positioned to allow different entities to operate independently.

• Inventory acts as a buffer between the customer and the manufacturing process.

• Impact of Lead-Time: The closer the decoupling point is to the customer, the shorter the lead time.

• The Trade-off: Quicker response times usually require a higher inventory investment, as finished goods are more expensive than raw materials.

• Role of Accuracy: Accurate forecasts allow for precise inventory levels. High uncertainty in demand forecasts necessitates extra inventory to cover the potential error.

Components of Demand and Trend Types

Demand for services or products is typically categorized into six components:

1. Average Demand: The baseline requirement for a specific period.

2. Trend: The overall upward or downward movement of the data over time.

3. Seasonal Element: Repeating patterns of demand increase or decrease during specific times of the year.

4. Cyclical Elements: Recurrent periods of activity that may be several years long, often influenced by political elections, economic conditions, or sociological pressures.

5. Random Variation: Unexplained portions of demand caused by chance events. Once all other factors are subtracted, the remaining value is random.

6. Autocorrelation: The persistence of occurrence; the value expected at any time is highly correlated with past values (e.g., waiting line lengths).

Common Types of Trends

• Linear Trend: A straight continuous relationship.

• S-Curve Trend: Typical of product growth and maturity cycles; characterized by a transition from slow growth to fast growth and then back to slow growth.

• Asymptotic Trend: Starts with high demand growth at the beginning and tapers off as the market becomes saturated.

• Exponential Trend: Common in products with explosive growth, suggesting sales will increase at an ever-increasing rate.

Quantitative Forecasting Models: Time Series Analysis

Time series analysis predicts future demand based on past data. The choice of model depends on the time horizon, data availability, required accuracy, budget, and personnel qualifications.

Simple Moving Average

A moving average is useful when demand is relatively stable (no clear trend or seasonality). It removes random fluctuations by averaging demand over the most recent periods.

• Formula:     $F_t = \frac{A_{t-1} + A_{t-2} + A_{t-3} + \dots + A_{t-n}}{n}$     where $F_t$ is the forecast for the coming period, $n$ is the number of periods, and $A_{t-i}$ represents actual occurrences in prior periods.

• Trade-off: Shorter period lengths follow trends more closely; longer time spans provide a smoother response but lag the trend.

Weighted Moving Average

This method allows for varying levels of importance to be assigned to each component, as long as the sum of weights equals $1$.

• Formula:     $F_t = w_1 A_{t-1} + w_2 A_{t-2} + \dots + w_n A_{t-n}$

• Sum of Weights:     $\sum_{i=1}^n w_i = 1$

• Application: More recent data is usually weighted more heavily unless demand is seasonal (e.g., weighting July sales higher for bathing suits in the next July forecast).

Simple Exponential Smoothing

This is the most widely used forecasting technique because it is accurate, easy to formulate, and requires minimal computer storage. It only needs three pieces of data: the most recent forecast, the actual demand for that period, and the smoothing constant alpha ($\alpha$).

• Formula:     $F_t = F_{t-1} + \alpha(A_{t-1} - F_{t-1})$

• Smoothing Constant ($\alpha$): Determines the response rate. Stable products use low $\alpha$ ($0.05$ to $0.10$). Growth products use higher $\alpha$ ($0.15$ to $0.30$).

• Approximation: $\alpha$ can be approximated by $\frac{2}{n+1}$, where $n$ is the number of periods in a corresponding moving average.

Exponential Smoothing with Trend (FIT)

To correct for the lag in exponential smoothing during upward or downward movements, a trend adjustment is added using a second smoothing constant, delta ($\delta$).

• Unadjusted Forecast ($F_t$):     $F_t = FIT_{t-1} + \alpha(A_{t-1} - FIT_{t-1})$

• Trend Estimate ($T_t$):     $T_t = T_{t-1} + \delta(F_t - FIT_{t-1})$

• Forecast Including Trend ($FIT_t$):     $FIT_t = F_t + T_t$

Linear Regression Analysis

Linear regression assumes a functional relationship between two correlated variables, following a straight line ($Y_t = a + bt$). This is primarily used for long-term forecasting of product families.

• Equations for Least Squares Method:     $b = \frac{\sum ty - N \bar{t} \bar{y}}{\sum t^2 - N \bar{t}^2}$     $a = \bar{y} - b \bar{t}$     where $Y_t$ is the dependent variable, $a$ is the y-intercept, $b$ is the slope, and $t$ is the time period index.

• Standard Error of Estimate: Measures how well the line fits the data:     $S_{yt} = \sqrt{\frac{\sum_{t=1}^N (y_t - Y_t)^2}{N-2}}$

Time Series Decomposition

Decomposition involves separating data into fundamental components like trend and seasonality. There are two types of seasonal variation:

• Additive Seasonal Variation: Forecast Including Trend and Seasonal = Trend + Seasonal Factor.

• Multiplicative Seasonal Variation: Forecast Including Trend and Seasonal = Trend $\times$ Seasonal Index. Multiplicative is more common as variations typically increase with the trend.

Measurement of Forecast Error

Forecast errors, or residuals, are the difference between actual occurrences and forecasts. Errors are classified as Bias (consistent mistakes, such as using the wrong trend line) or Random (unexplained variation).

Error Measurement Formulas

• Mean Absolute Deviation (MAD): The average forecast error using absolute values. Provides a measure of dispersion.     $MAD = \frac{\sum_{t=1}^N |A_t - F_t|}{N}$     Relationship to standard deviation: $1 \, \text{standard deviation} \approx 1.25 \, MAD$.

• Mean Absolute Percent Error (MAPE): Gauges the error as a relative percentage of demand.     $MAPE = \frac{100}{N} \sum_{t=1}^N \left[ \frac{|A_t - F_t|}{A_t} \right]$

• Tracking Signal (TS): Determines if the forecast is keeping pace with demand changes. A consistently high or low signal indicates bias.     $TS = \frac{RSFE}{MAD}$     where $RSFE$ is the Running Sum of Forecast Errors.

Causal and Qualitative Techniques

Causal Relationship Forecasting

Estimates demand based on independent variables that are leading indicators (e.g., housing start permits causing future carpet sales).

• Multiple Regression Analysis: Examines the effects of several variables (e.g., marriages, housing starts, income, and time trend) on demand ($S = A + B_m M + B_h H + B_i I + B_t T$).

Qualitative Techniques

Subjective methods used when historical data is unavailable, such as during new product launches:

• Market Research: Uses surveys and interviews to identify product preferences and ideas.

• Panel Consensus: Open meetings between different management levels to develop forecasts. A risk is that lower-level employees may be intimidated by executives ("Executive Judgment").

• Historical Analogy: Using existing products as models for new ones (e.g., using VCR history to forecast DVD player growth).

• Delphi Method: A structured process using secret questionnaires to ensure all participants have equal weight. It involves several rounds of summarizing results and redistributing them to experts until a consensus is reached.

Web-Based Collaborative Forecasting (CPFR)

Collaborative Planning, Forecasting, and Replenishment (CPFR) is an internet-based tool used to coordinate demand forecasting, production, and inventory between supply chain partners.

The Five Steps of CPFR

1. Creation of a Front-end Partnership Agreement: Specifies objectives (inventory reductions, etc.), resource requirements, and confidentiality expectations.

2. Joint Business Planning: Partners design a joint calendar for planning activities and set criteria for handling forecast variances.

3. Development of Demand Forecasts: Uses point-of-sale (POS) data from retailers; often utilizes simple procedures like moving averages modified by expert knowledge.

4. Forecast Sharing: Retailers (order forecasts) and vendors (sales forecasts) post data to a shared server. Exception notices are issued if differences exceed a safety margin (e.g., $5 \, \%$).

5. Inventory Replenishment: Agreed-upon forecasts become actual orders, triggering the replenishment process.

Implementation obstacles include a lack of trust between maximizing profit (vendor) and minimizing cost (customer), and concerns regarding the security of strategic data.