Both forecasting and managing customer demand to reach operational and financial goals
Step 1: Define users and processes
Step 2: Identify data sources
Step 3&4 Select and document forecasting tools
Step 5 : Monitor and continuously improve the
Time Series Analysis
Forecasts are based solely on past historical demand
Identifies and measures patterns in data
Assumes that patterns will continue in the future (Black Box approach)
Assumes demand is influenced by one or more factors
Identified important factors and measures their effects
Characteristics of Qualitative Forecasts
Can integrate soft data
Can be used for new products
Better with one off events
Characteristics of Quantitative Forecasts
Can handle large data sets
Requires hard data
Advantages of Causal Modeling
Best for long range forecasting
Best for predicting turning points or shifts in data patterns
Can generate a better understanding of the mechanisms influencing sales/demand etc.
Disadvantages of Causal Modeling
Requires more data
Requires special expertise that many managers do not have
Time Series Analysis Characteristics
Multiple identifiable patterns in demand:
Random fluctuations, Trends, Seasonality, Cyclical
Typically multiple patterns coexist and they can be separated and identified. The goal is to
Simple Moving Average
Simple average places the same weight on all past demand periods.
This method works well when the demand is fairly stable over time.
Decreasing the number of periods in forecast, creates more responsive data.
The forecast lags the demand because of averaging effect
This method doesn't do a good job forecasting when trends are present in a data
Weighted Moving Average
Allows greater emphasis to be placed on more recent data points to reflect changes in the demand.
Information used can be based on the forecaster's experience.
Decreasing # of periods in forecast or increasing the size of the weights creates a more responsive forecast.
The forecast still lags the demand because of averaging effect
This method does not do a good job of forecasting when trends are present in the data.
Simple & Weighted Moving Average
A simple or weighted moving average does not "hold to the past" will be forgotten by a moving average. Thus, outliers and extreme observations will eventually cease to negatively affect a moving average forecast.
Simple Exponential Smoothing
Like other time serious models, it is suitable for data that show no trend patterns. Due to its simplicity, exponential smoothing is a very popular forecasting technique. however the forecast will also lag any trend in the actual data. , Smoothing Methods
a) Simple moving averages
b) Simple exponential smoothing
Quality of forecast depends on accuracy and bias. Forecasting errors are absolute or relative.
Mean Forecast Error
Average of forecasting errors from period 1 to period current period. MFE is a measure of bias, but not accuracy. Smaller MFE means relatively unbiased forecasts.
Larger MFE means biased forecasts which indicates overestimation or underestimation
Mean Absolute Deviation
Average of absolute values of forecasting errors. MAD measures the dispersion(variance) of the forecasting errors. positive and negative forecast errors do not cancel each other out.
Smaller MAD means relatively more accurate forecasts
Larger MAD means relatively less accurate forecasts
Mean Squared Error
Average of squared values of forecasting errors. It is dispersion measure of the forecasting errors.
Positive and negative errors do not cancel each other
Smaller MAD means relatively more accurate forecasts, larger MAD means relatively less accurate forecasts.
The difference between MSE and MAD is that MSE penalizes larger forecast errors to a greater degree. And MSE is sensitive to extreme forecast errors.
Mean Absolute Percent Error
Standardizes the forecast errors with respect to the actual demand. Measure the average relative magnitude of the forecasts errors. Expressed as a % of the actual demand.
Short term forecast are more accurate than long term.
Aggregate forecast are more accurate than detailed forecasts.
Information from more sources yields a more accurate forecast
Fluctuating Customer Demand
cause operational inefficiencies. Timing and level of activities determine operational costs.
Smoother operations means lower costs and higher profits
Fluctuating operations means higher costs and lower profits.
Lead to expansion/contraction of capacity (equipment, labor, materials)
Backlog (delayed fulfillment of orders)
System Buffering (safety stock, safety lead time, capacity cushions)
Collaborative planning, Forecasting, and Replenishment: Focuses on information sharing among trading partners, and eliminates typical order processing. The goal to ensure effecient replenishment
CPFR in process
Step1: Market/item knowledge store planing item planing by individual stores(marketing and promotional input from sales/marketing)
Step2: POS Data(production planners validate item -level forecast)
Step 3:Create Item level forecast and special even calendar. (forecast drives productions)
Step 4: Create purchase order item (product shipped purchase order specification)
Value of CPFR
an exchange of forecasting information
not from complex algorithms to forecasting accuracy
Better information leads to forecasting accuracy
Better forecasting accuracy leads to better decisions, better resource allocation, better utilization,better coordination.
Benefit of CPFR
Strength partner relationship
Provides analysis of sales and order forecasts upstream and downstream
Uses POS data, seasonal activity, promotions, new product intros, and store openings or closing to improve forecast accuracy
Manages the demand chain by exception and proactively eliminates problems before the appear
Allows collaboration on future requirement and plans