Machine Learning (ML) has revolutionized the supply chain. Every day, companies realize significant benefits, from improving the quality and speed of supply chain decisions to increasing the degree of automation and enabling supply chain teams to focus on strategic initiatives that drive growth and service. Data is at the heart of all supply chains and companies need the ability to quickly analyze large, diverse data sets to discover patterns in the supply chain, identify and minimize risk, provide insights and prescribe recommendations, and help them gain more visibility and insights from their dynamic supply chains.
The digitalization of the supply chain is quickly changing the business landscape, creating a significant advantage for organizations that have embarked on a digital transformation journey. Today we have the ability to gather and analyze multiple sources of data such as IoT, point of sales (POS), weather, social sentiment, and more to drive greater insights and visibility into the supply chain.
With this explosion in volume, variety and veracity of data many companies have turned to machine learning to make sense of it all. The use of ML in supply chain is not new, innovators like John Galt Solutions, have been delivering solutions for more than a decade. In this post we will explore the top five use cases for machine learning in supply chain based on our extensive work with clients across multiple industries.
1) Demand Sensing
Demand sensing helps planners refine their demand forecast based on near real-time information in the supply chain. Using automated pattern recognition algorithms to capture, harmonize, and sort through masses of real-time data, ML can determine the influencing factor for each signal to predict for example customer orders.
An example of demand sensing in action is from the world’s largest packaged ice manufacturer. This company has placed IoT sensors in its iceboxes across grocery stores, gas stations and convenience stores. Each sensor measures how full iceboxes are throughout the day to provide real-time inventory levels. This information is then combined in the Atlas Planning Platform with point-of-sale (POS) and weather data to forecast each location’s hourly consumption level and plan their deliveries as precisely as possible.
2) Causal Forecasting
To handle increasing variety and complexity of variables that can influence future demand, causal forecasting helps planners measure the relationships between different variables to predict demand. These can include:
- Advertising and marketing efforts
- Macroeconomic
- Market intelligence
- Lead times like purchasing, manufacturing and shipping
- New product introductions, promotions and end-of-life
- Challenges with operating business scenario
- Seasonality, holidays and special events
3) Clustering and Segmentation
This is a popular use case for machine learning to help intelligently and automatically cluster various type of data such as products, customers, and attributes to accelerate and enhance decision making. Companies can leverage clustering and segmentation together with machine learning algorithms to establish relationships and find contextual information that can be used to create strategies. An example is the clustering of like products to understand halo or cannibalization impact. For instance, during the pandemic, an oil & gas distributor adjusted forecasts as residential propane use increased as more people worked from home and commercial use fell as less people worked from offices.
4) New Product Forecasting
New product forecasting allows companies to bring in multiple product attributes including category, style, channel, customer, and geography along with a variety of historical, market and competitive information into a single place. Machine learning analyzes this data to help companies understand key decisions including when consumers like Product A, they will likely purchase Product B.
Recently a fashion footwear brand utilized this ML capability to help them understand consumer preferences and determine which new collection will most likely to sell at the greatest margin potential from season to season, from boots to shoes to sandals.
5) Improving the Quality of the Supply Chain Model
Supply chain models must be dynamic and represent the real world. However, far too often we find supply chain decisions are based on stale, static data. Access to real-time, real-world data removes latency from the decision-making process and ensure the digital twin is an accurate reflection of the network. Greater visibility through improved quality drives increased accuracy of the supply chain planning model by creating a virtual supply chain blueprint that provide accurate data about the relationships between supply chain entities. For example, leveraging machine learning capabilities, supply chains can quickly evaluate inventory cost, service level, lead times and environmental impact across the supply chain (manufacturing, distribution centers, warehouses, etc.) and product types (finished goods, work-in-process, and raw materials) to best identify where and on what product types to hold inventory to best fulfill customer demand.
Another powerful use case is the use Multi-Echelon Inventory Optimization (MEIO) to automatically adjust inventory positions. MEIO seeks the optimal balancing of inventory at each stocking location and SKU level by looking at all the nodes in the entire supply chain and all the variability at each node such as cycle times, lead times, WIP, etc of the supply chain sources and all the stocking locations.
Recently a global flavor and spice manufacturer, whose supply chain is characterized by combinations of shard and unique ingredients, from raw materials through finished goods delivered to customers, implemented MEIO to optimize inventory associated with each retail customer.
What is Next in Machine Learning?
Machine learning is an essential element in the future of supply chains. Advances in machine learning and uses cases will continue to accelerate as companies synchronize their entire supply chain ecosystem to remove silos, maximize resources and gain end-to-end visibility.
