Machine learning is the new hot buzzword in supply chain. In a space striving for efficiency, the logistics and supply chain industries are ripe with opportunities to utilize machine learning. As global supply chains are growing in complexity and scale, technologies such as machine learning (ML) and artificial intelligence (AI) can help the industry continue to modernize as well as save time and money. But what are some actual use cases for machine learning in supply chain? How can logistics providers actually use ML? Continue reading to find out.
While many people tend to use AI and ML interchangably, there is actually a difference. To be clear, ML is a subset of AI, while AI encompasses a number of other technologies, including neural nets, deep learning, etc. Specifically defined, machine learning “enables a system to learn from data rather than through explicit programming.”
The secret sauce of ML is that computers can cull through seemingly endless data to identify patterns - without a human having to specify notable parameters and spend time analyzing data. The technology can identify important variables on its own, and can see opportunities for increased business efficiency without the need for human intervention.
The beauty of machine learning is that users do not have to tell the algorithm what to look for. Instead, the technology identifies important data and patterns on its own without human biases.
A great example of this is BlueDot. Months before the World Health Organization alerted people to the novel coronavirus in early 2020, BlueDot's AI platform was able to notice anomalies in Wuhan, China by analyzing global airline ticketing data and identifying an increased number of one-way flights out of the Wuhan area. Many human analysts may have overlooked this variable, but BlueDot's algorthim was able to anticipate a problem and notify their clients of a risk well before others considered COVID-19 an issue. BlueDot had noticed “unusual pneumonia” cases happening around a market in Wuhan, all by analyzing data many humans had overlooked.
The power of machine learning is that it sees patterns and data points humans cannot see. Machine learning algorithms can analyze extremely large amounts of data, and pick up on correlations and patterns with no human biases.
In order to use machine learning algorithms, a model must first be developed and trained using existing data. A machine learning model is an output generated when a user trains a machine learning algorithm with data. Once trained, inputs can be provided to the algorithm, and the algorithm can then generate an output.
A machine learning algorithm can identify and extract significant and useful patterns from a dataset, enabling better forecasts than possible with just human analysis. Trained algorithms can create better forecasts because they can look at more information and analyze at a deeper level than humans can. The speed and accuracy of machine learning algorithms can determine the important values and notice patterns based on the data to create more accurate forecasts.
One of Germany’s largest drugstores Drogerie Markt (DM) uses machine learning algorithms to predict future demands. DM’s six distribution centers serve over 3,300 international stores, and product availability and quality are top priorities. However, like many companies, DM struggles with high storage costs and capital tied-up in inventory. DM trained a ML algorithm using six-months of SKU-level data to create better weekly demand forecasts. Because of this algorithm, DM could better plan their inventory using these more precise demand forecasts. The company’s delivery and product reliability improved dramatically due to the introduction of machine learning.
Another area machine learning can help supply chains is in monitoring container and shipping data to alert a supply chain managers when a product or shipment will be late.
Using a predictive algorithm, machine learning can determine if a shipment will be late, using factors determined by the algorithm, including patterns humans may not be able to notice on their own. For example, a machine learning algorithm may be able to show that every seven weeks, shipments tend to be late. This delay may be tied to weather patterns, such as increased wind speeds which slow down ships transporting the containers. While a human may not connect these variables and see this pattern, a machine learning algorithm can objectively notice these data points and create accurate predictive analyses.
ML algorithms can also be used to predict when equipment should be serviced, providing early warnings of equipment failure before they happen. This can help companies know when equipment will not be available due to maintenance, helping them accurately plan production schedules and inventory / delivery schedules.
Another example of predictive machine learning algorithms is the Airbus Skywise platform. This system predictively tracks Delta Air Lines’ aircraft and monitors component conditions, predicting when aircraft will need maintenance and allowing Delta to better prepare for maintenance and downtime.
Machine learning algorithms can also use customer- and driver-submitted data to create up-to-date, real-time optimal delivery routes, while also taking into account live data from road conditions, traffic, weather and other factors.
For example, UPS uses a route optimization algorithm with AI-powered GPS called ORION (On-Road Integrated Optimization and Navigation) that helps drivers find the most efficient routes to their destinations. The algorithm optimizes delivery routes using factors like distance, fuel and time, helping to solve the notorious Traveling Salesman Problem. The algorithm takes into account tracking equipment, like IoT vehicle sensors as well as drivers’ mobile devices to capture data related to vehicle routes, such as idling times and driving speed, in order to produce the most optimal route between locations.
Anomaly detection is another primary example of how supply chain operations can use machine learning. Using anomaly detection strategies, ML algorithms can automate many quality inspection processes to analyze product defects. Machine learning improves quality audits and reduces the chances of delivering defective goods to customers.
Often, ML enabled quality audits are accomplished utilizing image recognition and categorization techniques. A machine learning-supported algorithm can be trained to recognize images that contain product damages. At the end of a production line, a picture of a finished good is taken and automatically analyzed, and instead of simply comparing against a static image of a good part, the algorithm dynamically analyzed the captured image and can better predict if a defect is present or not.
Using ML and AI algorithms can save time and money in your logistics operations - creating a more efficient supply chain. Machine learning algorithms help supply chain managers maintain standard operating procedures and best practices, including more accurate demand forecasts that can decrease holding costs and maintain optimal inventory levels, as well as predictive analyses that can create better delivery schedules and maintenance routines.
Machine learning does what humans cannot - finding patterns without knowing what to look for. Machine learning algorithms output actionable insights that help humans quickly problem solve, which results in continual improvement. These algorthms will not decide for humans, and will still require input from humans to optimally operate, but can supplement and aid associates by finding patterns humans would not easily identifiable.
With machine learning and AI, supply chains can increase efficiency throughout the entire process, saving time and money and better serving their end customers.