How emerging technologies can deliver efficiency across the supply chain
Are you puzzled about the differences between Artificial Intelligence (AI) and Machine Learning (ML)? Are you wondering how these technologies can enhance your role in inventory planning?. Don't worry, you're not alone. Let's explore how these technologies can be beneficial.
What exactly is AI, and how does it differ from ML? When people want to understand these terms, they often turn to Google, which can sometimes add to the confusion.
One source might describe deep learning, a neural network methodology, as a subset of machine learning, which itself is a subset of AI. Another source might argue that deep learning is already part of AI because it mimics human brain functions, unlike machine learning.
Some sources categorize machine learning into two types: supervised and unsupervised. Others expand this to four types: supervised, unsupervised, semi-supervised, and reinforcement learning.
There are also differing opinions on whether reinforcement learning falls under machine learning or AI.
Traditionalists might refer to many of these methods as "statistics," though not all of them fit this label.
The terminology can be influenced by both emotion and marketing. If a software vendor thinks you want to hear "AI," they might use the term to appeal to you.
To avoid getting lost in the hype, it makes sense to focus on the results provided by these technologies, regardless of their labels. Several analytical tasks are relevant to inventory and demand planners, including clustering, anomaly detection, regime change detection, and regression analysis. These methods are often classified as machine learning but can also stem from classical statistics.
Clustering involves grouping similar items and separating them from dissimilar ones. Sometimes, clustering is straightforward, like sorting customers by state or sales region. For more complex problems, data and clustering algorithms can automate the process, even with large datasets.
Figure 1: Clustering items based on the shapes of their cumulative demand
For example, Figure 1 shows a cluster of "demand profiles," dividing a customer's items into nine clusters based on their cumulative demand curves. Cluster 1.1 includes items with declining demand, while Cluster 3.1 includes items with accelerating demand. Clustering can also be applied to suppliers. The number of clusters is typically user-defined, but ML can guide this choice. For instance, a user might request four clusters, but ML might reveal six distinct clusters for analysis.
Traditional demand forecasting uses time series extrapolation. Simple exponential smoothing, for example, finds the "middle" of the demand distribution and projects it forward. However, sudden, one-time demand changes can skew near-term forecasts and affect safety stock calculations.
Planners might want to identify and remove such anomalies, but manually scanning thousands of demand plots is impractical. Anomaly detection algorithms can automate this process using statistical methods. You could call this "artificial intelligence" if you prefer.
Regime change detection is a more advanced form of anomaly detection. It identifies sustained shifts in a time series' characteristics, such as mean demand, volatility, or distribution shape.
Figure 2 illustrates an extreme regime change, where demand for an item drops significantly around day 120. Inventory control policies and forecasts based on older data would be inaccurate.
Figure 2: An example of extreme regime change in an item with intermittent demand
Statistical algorithms can address this issue, and it would be fair to label them as "machine learning" or "artificial intelligence." Using ML or AI to detect regime changes allows demand planning software to automatically use relevant history for forecasting, eliminating the need for manual adjustments.
Regression analysis relates one variable to another through an equation. For example, you might predict window frame sales based on the number of building permits issued a few months earlier. Although regression analysis has been part of statistics for over a century, it can also be considered "machine learning" since an algorithm determines how to convert one variable into a prediction of another.
It's natural to be curious about machine learning and artificial intelligence. The buzz around technologies like ChatGPT is intriguing, but it doesn't directly apply to the numerical aspects of demand planning or inventory management. The numerical aspects of ML and AI are relevant, but it's important to see through the hype and focus on their practical applications. If classical statistical methods can achieve your goals, you might use them and label any automated processes as ML if desired.
Learn how Epicor inventory management solutions can help your business thrive.
