Chapter 12: Demand Planning & Forecasting

Components of Demand

-stable pattern -seasonality and cycles -trend -shift or step change -auto correction -forecast error

Root Mean Squared Error

gives an approximation of the forecast error standard deviation -provides a good approximation of the standard deviation.

Demand Forecasting

is a decision in which managers predict demand patterns

Forecast Accuracy

measures how closely the forecast aligns with the observations over time. *every error whether positive or negative reduces accuracy*

Designing a Forecasting Process

1. identify users and decision-making processes that the forecast will support 2.Identify likely sources of the best data inputs 3. select forecasting techniques 4. Document and apply the proposed technique to the data gathered for appropriate business processes 5. Monitor the performance of the forecasting process for continuous improvement

Executive Judgement

Forecasting techniques that use input from high-level, experienced managers. -better for making judgments regarding long-term sales or business patterns. -high level managers have more access to source information to base their judgement

Weighted Moving Average model

a forecasting model that assigns a different weight to each period's demand according to its importance. -an adjustment to the moving average model -this model assigns a different weight to each period's demand according to its importance. ex. giving more recent periods more importance than earlier periods -more weight is given to the most recent demand because this is thought to capture market effects that are more relevant to the current demand situation.

Moving Average Models

a forecasting model that computes a forecast as the average of demands over a number of immediate past periods. -a way to create forecasts that reflect changes in demand dampening or smoothing out erratic movements is to forecast future demand as a simple average of the past demand values. -this model is used when the demand pattern is relatively stable WITHOUT trend or seasonality. -a smaller value of n produces forecasts that are more sensitive to changes in demand -a lager value of n tends to smooth out demand changes.

Marketing Research

a forecasting technique that bases forecasts on the purchasing pattern and attitudes of current or potential customers. -customer surveys, interviews, focus groups -click-streams and internet data.

Historical Analogy

a forecasting technique that uses data and experience from similar products to forecast the demand for a new product. -ex. when next generation electronics are introduced, manager use sales patterns for previous generations along with other information to predict the life cycle stages for the new products.

Mean Squared Error (MSE)

a more sensitive measure of forecast errors that approximates the error variance. -MAD and MAPE don't recognize that forecasts that are really far off the mark can be more harmful to the user than missing the actual demand by a small amount. -*gives more weight to larger and larger errors* -the variance of errors would use the actual forecast errors and the mean of the the forecast error.

Exponential Smoothing

a moving average approach that applies exponentially decreasing weights to each demand that occurred farther back in time. -each weight is a certain percentage smaller than the weight assigned to demand data for the previous period. -an other time series model for stable patterns -assigns moving average calculations in a systematic way -each new forecast is implicitly built upon many past actual demands, each of which receives less and less weight as one goes back in time

Naive Model

a simple forecasting approach that assumes that recent history is a good predictor of the near future. -the most simple of time series forecasts - simply assumes that tomorrow's demand will be the same as today's. ex. if a restaurant serves 55 customers, managers might expect to serve 55 customers the following day as well. -this approach ignores the trend, seasonal, or other components of past demands.

Seasonal Index

an adjustment factor applied to forecasts to account for seasonal changes or cycles in demand. -a "season" can occur daily, weekly, or in large periods. -is computed by dividing each period's actual demand by an estimate of the average (or base) demand across all periods in a complete seasonal cycle: that is the average demand that would be expected if no seasonality existed. ex. if there are 4 periods in a compete season cycle, the one would compute the average demand across the 4 periods in the cycle. -alternatively, the average demand can be estimated using a time series regression model, because it creates estimates of average demand all across the time horizon.

Judgement-based forecasting

are built upon the estimates and opinions of people,most often experts who related sales or operational experience. -seek to incorporate factors of demand that are difficult to capture in statistical models. *this is useful when there is a lack of quantitative historical information* ex. when a new product is about to launch or when past information may not support future decisions

seasonality and cycles

are regular patterns of repeating highs and lows. Seasonality may be daily, weekly, monthly or even longer ex. restaurants experience seasonal patterns during the day peaks. Banks typically experience a monthly seasonal pattern with peaks when employees get paid. -economic, political, demographic, and technological factors influence these patterns

Simulation Models

are sophisticated mathematical programs that offer forecasters the ability to evaluate different business scenarios that might yield different demand outcomes. this help to better understand how different variables and drivers of demand relate to on another.

Casual Models

assumption that a relationship exists between the item being forecast and other factors. -these models concentrate on external factors that are thought to CAUSE demand. ex. the amount of household disposable income in an economy might be a good leading indicator of the sales of luxury items like sailboats.

Times Series Analysis Models

compute forecasts using historical data arranged in the order of occurrence. Forecasting models that are based only on a series of past demands assume that a demand pattern of the past will continue in the future. Forecasts are generated by summing weighted values of past demands, and the weighting schemes range from very simple to very complex. *the type of weighing used depends on the demand patterns*

auto correlation

describes the relationship of current demand with past demand. -if values of demand at any given time are highly correlated with demand value from the recent past then we say that the demand is highly autocorrelated.

Delphi Method

develops forecasts by asking a panel of experts to individually and repeatedly respond to a series of questions

trend

identifies the general sloping tendency of demand, either upward, or downward in a linear or nonlinear fashion. -new products in the growth phase of the life cycle typically exhibit an upward, nonlinear trend.

shift or step change

in demand is is a one-time change, usually due to some external influence on demand such as a major product promotional campaign, or sudden economic shock.

stable pattern

is a consistent horizntial stream of demands. Mature consumer products for example shampoo or milk often exhibit the type of pattern.

Demand Management

is a proactive approach in which managers attempt to influence the pattern of demand. -involves the use of pricing and promotional activities.

Grassroots Forecasting

is a technique that seeks inputs from people who are in close contact with customers and products. ex. a marketing study may ask sales representatives for their sales estimates and comments on current market conditions. -a major limitation of this technique is that "experts" may unconsciously base their forecasts on their most recent experiences, rather than their entire set of experiences. -*most useful for short-term forecasts for individual products*

Forecast error

is simply the "unexplained" component of demand that seems to be random in nature. *the difference between a forecast and actual demand* a good forecasting process will produce a small error.

Forecast Bias

is simply the average error. -indicates the tendency of a forecasting technique to continually over predict or under predict demand. *Bias= Mean Forecast Error (MFE)* -does not allow easy comparison across products when the average demands are different.

Demand Planning

is the combined process of *forecasting* and *managing* customer demands to create a planned pattern to create a planned pattern of demand that meets the firm's operational and financial goals. -you're forecasting for both the quantity and timing or demand. -

Regression Analysis

is the most commonly used method for estimating relationships between leading indicators and demand. -Simple linear regression is a technique that finds "optimal" values for the parameters a and b. These parameters that will most closely equate the independent variable t, and the dependent variable d, over a set of values.

Artificial Intelligence

refers to learning and decision making capability that stems from software algorithms. -combines time-series analysis, casual modeling, simulation and focused forecasting techniques -instead of requiring manager inputs (as focused forecasting does), learning algorithms that are embedded in forecasting software are able to develop rules and heuristics on their own -the software learns how to weight and adapt various inputs

Mean Absolute Percentage Error (MAPE)

the MAD represented as a percentage of demand. -for purposes of comparability across products, forecasts need to adjust the MAD to create a related metric. -the MAPE indicates how large errors are relative to the actual demand quantities.

Mean Absolute Deviation (MAD)

the average size of forecast errors, irrespective of their directions. Also called *mean absolute error* -*the simplest measure of forecast accuracy*

Assessing the performance of the forecasting process

the primary measure of forecasting performance is forecast error. *positive forecast error*- an overly pessimistic forecast *negative forecast error*- indicates an overly optimistic forecasts Forecast errors can be examined to determine 2 primary aspects of forecast performance over time: *1. forecast accuracy* *2. forecast bias*

Statistical Model-Based Forecasting

transforms numerical data into forecasts using one of three models. 1. *time series analyses*, which extrapolate forecasts from past demand data. 2. *Casual studies*, which look for casual relationships between leading variables and foretasted variables. 3. *Simulation models*, which try to represent past phenomena in mathematical relationships and then evaluate data to project future outcomes 4. *Artificial intelligence*, in which a "smart" computer program "learns" from a combination of casual and simulation analyses using a wide array of data.

smoothing coefficient

where a is a constant between 0 and 1 -this parameter indicating the weight given to the most recent demand.

Planning Activities

*forecasting activities* integrate information gathered from the market, from internal operations and larger business information gathered from the market. This information includes: -past demand -past forecasts and their associated errors -business and economic metrics -judegments of experts -demand management plans that specify the firm's pricing and promotional strategies. The demand management system in turn creates uses these forecasts as inputs for future demand management planning. -these systems are used to manage resources and operating systems.

The Role That Demand Planning Plays in Operations Management

Demand planning drives almost all other activities in operations management. -managers have to anticipate demand and plan what materials and resources they will need well in advance of actual orders. -Cost of making forecasts too high include money lost in holding inventory that is never sold, lost capacity that is spent making forecasts that no one wants to buy and lost wages paying for workers not needed. -Cost of making forecasts too low include lost sales and lower product availability for customers.

determining trend factors

The trend component of a time series normally results from some market force that causes a general rise or decline in values over time. a linear trend results when demand rises or falls at a constant rate, describing a straight line on a graph. to estimate a trend you should begin by graphing the data.

Focused Forecasting

a combination of common sense inputs from front line personnel and a computer simulation process. -simulation-based approach -uses "rules of thumb" ex. one rule might be "we will probably sell 10% more product this month than we did in the same month last year" -predicting demand from past data -*this approach has delivered better results* than those provided by exponential smoothing or other time-series based models -however it requires more preparation and user involvement

Mean Percent Error (MPE)

average error represented by a percentage of demand. -better for comparability sake