Shipping and delivery schedules were already being planned out in the 1600’s between England and the Netherlands. Crates, barrels, or other non-standard containment systems were used until the 1950s when the standard shipping container was developed and later standardized in the late 1960s. Until recently, this has been the most significant development, but data is changing all of that. Even a small increase in efficiency carries huge benefits in terms of safety and savings.
Airfare is segmented into a few different domains. There is commercial airfare, private airfare, and air cargo. There is overlap amongst these segments. For clarity’s sake, the following solutions are segmented into travel and freight.
Get your ticket early. It’s an age-old piece of advice that we all hear, but this seemingly useful tidbit doesn’t actually reflect reality. You’ll be surprised to learn that the best time to purchase your ticket is typically either eight weeks or two weeks before your flight, but data can be used to pinpoint a more exact timeframe. Ticket prices that vary according to a given flights’ demand are far fairer for consumers and lucrative for airlines. A flight has to complete its trip regardless if it is filled to capacity or not, and this has an effect on ticket prices. Data can be used to maximize booking rates, increase margins, and give thrifty consumers the chance to save more by booking at the perfect time.
Global consumption is not random. We can, with relative ease, predict demand for food, clothing, technology, and other common items. Veblen goods are tied less to the economy and more to social behavior. And, with this data, we know how much people want to buy. When it is analyzed, corporate purchases are often found to be less than projected sales; this is partly due to long sales windows and careful decision-making. Strategic plans can shed light on these plans in advance. Nevertheless, freight and logistics companies have even more data than just global trade at their disposal. Their entire history can be used to train models that predict demand. Moreover, idiosyncratic events in the past can serve as guides for dealing with similar future events.
Certain containers are intended for particular items. These may be liquid, hazardous, or simply general purpose. The containers intended for a specific usage like liquid chemicals have longer turnaround times. In some cases, they may have to be shipped empty in order to be used because of a lack of demand in particular parts of the world. By optimizing routes, these containers’ downtime can be minimized. As a result, when these containers are washed and prepared, there will be an intended client ready to use them next. Siemens recently invested in data science and has already seen returns. You can see a demonstration of how you would apply a data-based solution using Keras to assist shipping logistics.
Cities predict their public transits demand to determine how many buses/trains they need, where to build new lines, and how often clients need their services. You can read more about smart cities here. Train companies work on similar problems but usually abide by stricter schedules. There are obviously issues with other rail companies using particular tracks, but schedules are rarely so tight as to not leave room for change. Certain clients expect trains to run at a particular time, but planning train schedules without learning more about their passengers’ behavior is inefficient. Companies can learn their clients’ behavior to better plan the number of cars that they need to provide at particular times. They can increase or decrease the usage of particular routes and optimize their schedules to serve as many clients as possible, while simultaneously minimizing waste.
Deep learning with R and Keras – Build a Handwritten Digit Classifier in 10 Minutes.
Courier companies have been working on optimizing every aspect of their process to provide their services at the lowest cost possible. Companies like FedEx, DHL, or UPS are constantly looking for ways to improve. Smaller companies must be aware of such methods and react accordingly. In order to remain competitive, they must work on their own methods of optimization. A local produce supplier could save minutes a day by properly arranging his stock to minimize the time it takes to unload. Saving just 15 minutes a day adds up to about two weeks of work time throughout the course of a year. For one year, this can translate into 5-10% more business. If properly managed, additional customers would allow for faster growth, resulting in a snowball effect that could dramatically change the trajectory of a company.
Trucking is an interesting business. Just as in many aspects of society, a small portion of individuals tend to look for their own personal gain at the expense of others. Fraudulent behavior, namely with regard to fuel, costs companies millions a year. Ultimately, these costs are paid by the consumer, leading to an increase in product costs and a decrease in logistic efficiency. Economically speaking, lower transformational efficiency results in less cooperation, and eventually, a weak economic structure. But, this type of fraudulent behavior can be identified through data science.
In many countries, regulations arm trucks with countless sensor devices for safety reasons. These sensors are ready to collect data that can be explored for further insights.
GPS and IoT data allow us to monitor the behavior of any given truck. While there are techniques that allow for fraud, such behavior can be detected by sensors. Furthermore, the modern logistic legislature requires that there be countless sensors on a truck. This means that little effort is needed to discover and extrapolate such insights. All we need to do is take a look at data that is already being recorded.
Automatic Identification Systems (AISs) have become standard over the past decade. They are officially mandated on larger ships but have recently become much more commonplace. When moving, vessels send their data as frequently as every two seconds and around every two minutes while anchored. Ships are expected to report on their predicted route, destination, and ETA, but these are set manually and are often inaccurate. The route a given vessel chooses can help us understand the potential time it takes to reach the destination it will be heading to. Machine learning, in combination with this data, can significantly improve vessel tracking and provide more accurate ETAs.
Officially speaking, AIS data is supposed to include information relating to the planned destination. It would be quite nice if this information was up-to-date and provided a valid overview, but that is not often the case. Thankfully, routes are fairly specific to particular destinations. Even on open seas, we can take advantage of historical data to predict the statistical likelihood of potential destinations. For example, when a ship is transporting itself through the Strait of Gibraltar, we have, at least to a certain extent, relative certainty of the list of potential destinations.
The crane – the most significant tool in a port’s arsenal. These mighty tools are the main way of loading and unloading a ship. There is software that tries to optimize many aspects of this process, including load distribution and final destination as well as trying to select the optimal order to minimize load times. But there is something missing – artificial intelligence, or, as it is known in tech, machine learning. Heuristic or rule-based systems are currently used to optimize the order and location of containers during loading. The Cold War taught us those heuristic systems are only useful for simple tasks, and the scale of an entire vessel’s cargo is complex. The way this would work is actually quite simple to explain. First, we need to aggregate historic container data. We’d have an even easier time if we had data relating to particular cranes. This data can then be fed into either an unsupervised system, where anomalies are automatically identified and an algorithm makes sense of the data without human interference or a supervised system, where we use said data to train our algorithm. Both of these methods have their benefits. In short, if we have a significant amount of data and a complex system, then we should look to machine learning to increase performance.
Global trade is a marvel of the modern age. There are many, even hundreds, of touchpoints involved in transporting items across the globe. When approaching a port, tides, traffic, and weather issues occasionally force a ship to wait before re-attempting to dock after a few hours. Once a ship does dock, it needs to be unloaded, containers must be transported, passed through customs (if necessary), and loaded onto the delivery vehicle. Every step along the way has room for potential delays and issues. As a trader, it is important to be up to date in real-time on the particulars regarding a given ship, its content, and the containers involved. The most accurate way to do this involves taking advantage of AIS data and tracking IMO codes. Data science allows us to take this a step further and predict the types of issues the shipment might encounter.
The corollary to destination predictions involves the destinations themselves. When aggregating the predicted destination data, we are, in effect, taking a few steps into the future. Though this data is not 100% accurate, it still provides us with information that is adequate enough to be able to gauge demand. Historical prediction is most useful for the short term. When predicting the future demand of a port, we must adopt a different tactic. Economic drivers affect the demand and supply of various products. Trade agreements and technological development further complicated this dynamic. Creating different models that take a particular look at these factors and coalescing them into a macroscopic view allows for the most accurate predictions.
The standard 20’ and 40’ containers have been around for nearly half a century. In that time, they have eased the transport of billions of tons of cargo around the world. And yet, there is still room for improvement. First off, there are codes unique to each vessel, but they can be obstructed or damaged. Deep learning can allow us to recognize individual containers to improve tracking and minimize the number of containers that are lost in transit. Additionally, increased monitoring can identify common points of failure and help optimize future workflows.
Predictive maintenance is a fascinating use case, as it can be applied across countless industries. The particulars come down to the specific component or system that is being monitored, but algorithmically speaking, there is a significant overlap. These kinds of algorithms are solutions that can help avoid catastrophes by identifying potential failures before they happen. They can also decrease maintenance costs by identifying subtle changes and component idiosyncrasies. Catastrophe prevention speaks for itself, but the savings from lowered maintenance adds up to millions over a given product’s lifetime. Another example is that data science can help identify fuel consumption anomalies. Fuel consumption is second to crew costs when looking at operational expenses. There are a few aspects in shipping where fuel anomalies arise. For instance, we can take a look at potential malicious behavior like skimming fuel. There are systems in place intended to counteract such practices, but monitoring isn’t precise enough to catch every occurrence. The AI methods that we mentioned above can be fine-tuned to identify such behavior and flag long-term patterns. These patterns can also point out inefficiencies in your engines or steering behavior. Your engine may only need minor maintenance to output a 20% increase in fuel efficiency, but how were you supposed to know otherwise?
There are long-trusted suppliers that do their work well enough, but have you worked with them for so long that you find it hard to imagine actually changing them? Are you sure about the quality of their work? Are they really providing you with the work that you are paying for? It’s worth taking a look at. You can compare the work they have done in the past and with other factors that may provide insight into their current level of quality. We can compare their initial work from when they first started to another crew that is working on a similar problem. Insights from data are not the only indicator, there may be other factors you may not be aware of. This is why it is important to look for the causes of any issues before blindly trusting data points.
Vessel Monitoring Systems (VMSs) allow us to track fishing vessels within the economic exclusion zone of a given country or region. National laws and regulations vary, but overall, these kinds of systems are very useful for monitoring shipping vessels. They are much more accurate than AISs, as they transmit constantly. Taking years’ worth of data from an entire country’s fishing vessels gives a significant overview of the extent to which certain fish and areas are being farmed. Such an overview allows agencies and private companies to plan their future fishing behavior and minimize negative environmental impact.
Warehouses that stock inventory that will never be purchased are wasting their shelf space. More importantly, they are inadequately forecasting their demand. Proper forecasts can help reduce the chance of shelves being understocked by more accurately predicting demand. Such solutions apply not only to finished products but also to the manufacturing and assembly process itself. Complex engineering products may travel thousands of kilometers before being completed. Inefficiencies quickly add up, leaving much to be desired. Decreasing the number of steps is a manufacturing challenge and being able to extrapolate the demand for particular components based on the finished products’ demand can serve as an idealized goal. It can be used as a reference point when training models and looking for areas to optimize.
We’ve written an entire post on dynamic pricing. We better reflect actual demand and allow consumers to take advantage of limited-demand products while allowing businesses to increase their profits for high-demand items.
Amazon has shown that a conveyor belt worked for the third industrial revolution, but it is not the most optimal solution for warehouses and factories in the fourth one. Robots are only a small part of the solution. To delve deeper into our Amazon example, consider the structured chaos that is item bin management. Things are not stocked on shelves based on their category, but rather, they are scattered in small, individual bins that have been algorithmically optimized to provide workers the ability to quickly and efficiently find and package customer orders. Their adoption of technologies causes short-term creative destruction that will lead to major improvements in the long run.
A fluid business that iterates on its processes looking to reimagine its future is providing itself with the best chance of being relevant and profitable. Companies that stand still are taking a step backward. Their competitors are not waiting for them to catch up. Land, air, or sea, the method varies, but the nervous system of global trade has begun taking advantage of machine and deep learning to optimize and automate their processes. Dynamic Pricing, demand forecasting, route optimization, and inventory management are just a few solutions today’s technological innovations provide. Set up a call now to find out if your business is ready for the future.