In the realm of supply chain management, time series forecasting plays a crucial role in guiding businesses through uncertain futures. This article targets stakeholders in the supply chain who need to ascertain the quantity of finished goods required across various planning timeframes. Beyond just determining how many units are necessary, organizations must also consider the geographical distribution of inventory to optimize availability.
The Fragile Balance of Supply Levels
When manufacturers produce insufficient parts or finished goods, they face the risk of undersupply. This can force them into difficult decisions regarding the rationing of their limited resources among partners or internal business units. Consequently, purchase orders may be rejected more frequently, leading to reduced profit margins. On the consumer end, if retailers lack adequate stock to meet demand, they can disappoint customers due to out-of-stock situations. Such shortfalls can drive shoppers to alternative retailers or brands, which poses a churn risk if these substitutes become the new go-to options.
Conversely, an oversupply of products can incur its own set of penalties. Excess inventory must be held until sold, and while a certain level of safety stock is necessary to mitigate demand fluctuations, carrying surplus inventory creates inefficiencies that can erode an organization’s bottom line. This is especially crucial for perishable goods, where an oversupply may lead to lost investments. Even non-perishable items represent idle resources while stored, resources that could otherwise contribute to the balance sheet as liquid assets or be utilized for other investments. Organizations must work within the constraints of their warehouse and logistics capacities, ensuring efficient resource use.
Faced with the dilemma of oversupply versus undersupply, many organizations tend to lean towards the former by choice. The costs associated with undersupply can often be significantly higher, sometimes by several multiples, compared to the expenses of carrying excess inventory, as we will explore further on. The primary motivator behind this inclination is the desire to maintain customer goodwill and avoid the intangible costs of unavailability. Manufacturers and retailers alike focus on nurturing long-term customer relationships and brand loyalty, which in turn shapes their supply chain strategies.
In this section, we’ve discussed the challenges of resource allocation stemming from demand planning processes. Next, we will delve into time series forecasting and how demand predictions can align with item-level supply strategies.
Traditional Approaches to Sales and Operations Planning
Traditionally, forecasting has relied on statistical methods that yield point forecasts—predictions that offer a most-likely estimate for future trends. These methods typically utilize moving averages or linear regression to create models based on ordinary least squares. A point forecast represents a singular average prediction, where, statistically, the actual value will exceed this mean about 50% of the time, leaving the other 50% for values falling below it.
While point forecasts can be informative, they often lead retailers to face stockouts on essential items half of the time if followed without thorough scrutiny. To prevent disappointing customers, supply and demand planners frequently apply manual adjustments or modify point forecasts using safety stock formulas. Companies may adapt their interpretations of these formulas to ensure that product availability is maintained through uncertain short-term horizons. Ultimately, planners must decide whether to adjust their mean point forecast estimates up or down based on their subjective views and interpretations of future demand.
Modern Time Series Forecasting for Informed Decision-Making
To address real-world forecasting challenges, AWS offers a comprehensive suite of capabilities that modernize time series forecasting. Their machine learning services include Amazon SageMaker Canvas, Amazon Forecast, and built-in algorithms from Amazon SageMaker. Additionally, AWS has developed AutoGluon, an open-source software package that facilitates various machine learning tasks, including those related to time series forecasting. For further insights, check out this resource on effective forecasting with AutoGluon.
The previously discussed point forecast may not capture the complexities of real-world data effectively. Given the imbalance between over and undersupply, relying solely on a single point estimate is insufficient. AWS services meet this need through the application of machine learning models and quantile regression, enabling businesses to choose from a spectrum of planning scenarios rather than limiting themselves to single-point forecasts. These quantiles provide a range of choices, which we will elaborate on in the following sections.
Forecasts That Prioritize Customer Needs and Business Growth
Visual representations of time series forecasts can illustrate multiple potential outcomes through quantile regression. For instance, a red line marked as p05 indicates a 5% probability that the actual value will fall below that line, suggesting that 95% of the time, the true value will likely exceed it. Similarly, a green line labeled p70 reveals that the actual value will fall below this line about 70% of the time, with a 30% chance it will surpass it. The p50 line serves as a midpoint, indicating a balanced probability distribution around the mean.
In the next section, we will explore how to quantify oversupply and undersupply based on historical data.
Assessing Oversupply and Undersupply Using Historical Data
In the previous section, we showcased a graphical approach to understanding predictions; another method involves presenting data in tabular format. When developing time series models, a portion of the data is typically reserved from the training phase to generate accuracy metrics. Although future outcomes remain uncertain, the accuracy observed during a holdback period serves as the best approximation for future predictions, all else being equal.
The table presents true historical values alongside various quantile predictions ranging from p50 to p90 in increments of ten. For instance, over the past five time periods, the actual demand was recorded at 218 units, with quantile predictions spanning from a low of 189 units to a high of 314 units. For additional insights into this topic, visit this excellent resource that delves into related issues.
In conclusion, effective supply chain management hinges on balancing customer satisfaction with growth goals. By leveraging advanced forecasting techniques and understanding the intricacies of supply levels, organizations can better navigate the complexities of market demand.
Leave a Reply