Traditional Time-Series Forecasting Methods
The moving averages approach creates a series of averages over a period of time. The trend these averages create can be used for future predictions. Moving averages are best used when limited data is available, and influencing factors are unknown or non-existent.
To use a moving average to calculate your next month's demand (P4) you should have the total demand for the previous 3 months (P1, P2, P3) at a minimum.
Weighted Moving Average
Weighted moving average is similar to moving averages, except that it places a weight consideration on recent or influenced data. This difference is helpful when considering fluctuating data over a greater period of time. The emphasis on data is determined by a weighting scale.
Flat scale: A variety of scales can be used for this formula, if there is not a known influencing factor on a specific period (sale, seasonality, etc) then a flat sale can be used. Your scale will be the # of time periods being used to calculate the weighted moving average. With the highest value being the most recent data point. If June, July, & August were the time periods used, August would be 3, July 2, and June 1.
Variable scale: If an influencing factor on a time period needs to be considered, the weight cannot be linear. You must decide on a weight multiplier. In this case, a fraction of 1 could be used. In this instance, June and August had normal sales demand, however, in July there was a blow-out sale which tripled sales volume. Less emphasis should be placed on July. In this example, the following weights could be used for June, July, and August; .5, .15, .5.
W represents the weight, T represents the time period, and D represents the demand.
|Month||Actual Sales||Influencing Factor||Weight Factor|
|4||40||Unforeseen supply shortage, limited sales||.4|
*If your conclusions are vastly out of range with historical data you may need to adjust your weights. Try using historical data without influencing factors to determine a weighting scale that is best for you.
Exponential smoothing is a slightly more advanced and popular method for demand forecasting. This method places an exponentially decreasing weight on older data points. A major limitation to this is that the formula cannot account for influencing real-world factors such as supply shortages, seasonal changes, or promotions.
F Forecasted sales
α Smoothing constant
A Actual sales from the previous period
B Forecasted sales from the previous period
Advanced Demand Forecasting
The above time-series forecasting methods can be quickly exercised in your business with access to historical data and a bit of excel work. However, they do not account for the real-life variables that constantly manipulate the manufacturing industry and can require dozens of hours to prepare reports for each SKU.
For these reasons, Flexfab implemented an AI-driven Demand Forecasting System. Rather than running a report for each formula, this software uses a mixture of 14 leading demand planning algorithms and automatically optimizes the predicting algorithm for each SKU! Additionally, it accounts for external variables such as weather, stock markets, supply chain data, etc. We've been able to exponentially increase the accuracy of our forecasting predictions, connect our findings to specific demand driving variables, and find a number of new opportunities.
And it's easy to use. The software was specifically designed so existing systems don't need to be reconfigured, it can connect to any existing custom or out-of-the-box program. Implementation can be as fast as 2 weeks!
Reduce connected spreadsheets and feed data directly to a cloud or offline dashboard.
Once data has been received, standard implementation can take less than 2 weeks!
SKU-Level optimization is designed to accurately forecast low sales volume SKUs that are still critical to operations.