Demand Forecasting and Product Usage: We need to know how much product is demanded on average each period to know how much to order.
Exponential smoothing: A forecasting technique that can be applied to smooth time series data and calculate accurate forecasts. This technique allows for fluctuations in demand without creating an overabundance in safety stock while maintaining high service level.
Product Average (Product’s Usage): The average is calculated using Exponential Smoothing which focuses more weight on the most recent data.
For a normal moving horizontal product, when a new period begins, K3S will take 90% of the current average and add it to 10% of the most recent period’s demand. The 10% is considered the smoothing factor. The smoothing factor is determined by the forecast error of the product which is how much each data point of history is above or below the average.
Forecast Error: Refers to any period of history that is different from the average. Typically, every period of history will have some level of error compared to the product’s four-week or monthly average. For example, if a product has an average of 5 units per month and there are 6 units in demand for the last month, there will be a forecast error of 1.
Mean Absolute Deviation: The Mean Absolute Deviation (MAD) of a set of data points is the average distance between each data value and the mean (product average). The Mean Absolute Deviation allows us to filter out fluctuations for forecasting and correctly flag exceptions for our users.
There are five different PE Checks:
PE1 Check: Demand Spikes occur when the demand for a product for a given period is much higher than the product’s average. Every product is unique in that it will have a different level where it is considered to be a PE1 Check.
PE2 Check: Demand Dips occur when the demand for a product for a given period is much lower than the product’s average. Like PE1 checks, every product is unique and has its own level where it is considered an exception.
PE3 Check: Upward Trend occurs when the forecast error is more often above the average than below the average in recent periods. Here the tracking signal has increased and we will place more weight on the most recent period of history.
PE4 Check: Downward Trend occurs when the forecast error is more often below the average than below the average in recent periods. Here the tracking signal has increased, and we will place more weight on the most recent period of history.
PE5 Check: Product missed its Service Level Target
Seasonality occurs when there are specific times of the year where a product’s demand increases or decreases. Knowing when these patterns occur is helpful to be ahead of the change to adjust the forecast. K3S uses Seasonal Profiles to predict when these patterns occur for certain products like ice cream, soup, snow tires, and more.
Seasonality: A characteristic of inventory where the data experiences regular and predictable changes which recur every calendar year. Any predictable change or pattern in inventory levels that recurs or repeats over a one-year period can be said to be seasonal.
Demand Pattern Line: The Demand Pattern Line is built from two years of past history (or three years if it exists), and is used to predict how future years will map seasonally. We use the history only if there is history for every period (zeroes count).
The Demand Pattern Line is built using algorithms that weigh the most recent Year 1 more heavily than Year 2, and weigh Year 2 more heavily than Year 3. These algorithms also understand business growth or business decline from year to year. If less than two years of history exists for a product, then the Composite line is built from the most recent year of history (Year 1).
Seasonal Profile Line: The Seasonal Profile Line (displayed in green) tells you what percentage of the Average the demand is expected to be for that particular month. It appears under the Seasonal Profile Line as a percentage. Remember that the Seasonal Profile Line does not keep any of the changes you make to it during simulation; only the values in the Factor line get saved when you generate a new profile. Multiply your average by the factor to get your expected demand for that period.
You are also able to see how the Average and Deviation will change if you accept the profile. This change reflects a more accurate based average (where the Factor = 1.00). This change reflects how well the seasonal profile matches past history and will enable us to carry less safety stock.
Article last revised on March 27, 2018