How artificial intelligence has revolutionised logistics

Stefan Seufert, CTO/Vorstand EIKONA AG
Older man with gray hair stands in a laboratory and plays chess against artificial intelligence

A buzzword that sounds like science fiction to many has the potential to revolutionise logistics: artificial intelligence (AI). This IT term refers to an application that processes large amounts of data in order to independently improve, make predictions and take intelligent decisions.

AI technologies harbour the potential to significantly improve logistics. Self-learning processes can, among other things:

  • Capture and process data in real time
  • Analyse large volumes of data
  • Recognise patterns and make predictions from them
  • Automate warehouse processes
  • Simulate complex networks and forecast their utilisation levels
  • Make real-time decisions

In logistics, AI solutions have the capabilities to bring about a serious change in value and supply chains. They require tremendous computing power, large amounts of process data and human coaches.

Lay the groundwork

Artificial intelligence needs intensive training

AI applications can extract patterns from similarities in a fraction of a second and use them to determine an ideal course of action. For example, they know the likelihood of a shipment reaching its destination on time. They can identify the routes on which a freight forwarder needs daily direct shipments given the volume of shipments from its customers. Or whether the same service provider would be better off joining a forwarding alliance than working with its own partner network. To make these assessments, the AI first has to be provided with a whole range of parameters and a large amount of historical data. It also has to be trained from the ground up by working its way through simple questions with predetermined answer choices (yes/no, large/medium/small, intact/defective, etc.) until it can assess complex situations. Humans are both teachers and sparring partners for the AI in this process: the machine can learn but cannot set its own goals or give itself feedback on the accuracy of its assessments. For those things, it needs human development partners with a lot of patience and access to relevant, comprehensive data. Training AI is a lengthy process that requires a great deal of expertise, too. The human coach specifies answers and provides appropriate examples, while the AI analyses data, recognises correlations and learns from them. In other words, the human is not needed as a trainer in the traditional sense or a direct counterpart at every learning step. Once you have a good baseline, it is time to begin machine learning (ML), i.e., the independent improvement of AI through pattern analysis and derived conclusions.

Set goals

What insights are important?

The software needs the answers to a number of goal-oriented questions to evolve properly. For example, it first has to learn when a shipment is on time so that it can make suggestions for improving fleet punctuality based on further insights. All this depends on the data that helps the software to learn correlations. The data has to meet strictest standards: It should be both error-free and relevant. Irrelevant data should be removed from the training material from the start. For example, there is no meaningful relationship between on-time delivery and the dispatchers' shoe size or hair color. That information does not need to be collected for an assessment. On the other hand, the more relevant data an AI receives, the more surprising its results seem to be for us humans.

Big data

Add value through data evaluation

The greatest tool for increasing efficiency with AI is the evaluation of unused data. This data holds the potential to improve processes, widen margins and cut costs. It can also be used to increase security of supply. However, this requires the ability to represent logistics processes as correlations that can be represented by questions with suggested responses. This favours the basic structure of logistics per se: It consists of simple, clearly defined processes that can be easily evaluated automatically. They do not become impenetrable to human planners until they are interconnected in a complex network. That is because the planners would have to recognise and determine correlations within the company (vertical) and across company boundaries (horizontal). This may exceed the limits of a human's mental capacity, but it is an ideal training environment for AI. AI applications in artificial neural networks (ANNs) make connections between patterns that they have recognised. As networks branch out further and further, AI can develop solutions to increasingly complex problems. Human trainers are then often unable to understand completely (or at all) how the AI arrived at these solutions. This produces process improvement suggestions that, depending on their degree of maturity, are authorised by the coaches or are approved automatically: from machine learning to deep learning. For example, the delivery speed of a product depends on much more than whether the product is picked up by a truck immediately after it is finished. AI recognises that it is also affected by the time of day, weather, traffic density, vehicle size and other factors. Arrival time forecasts become more reliable that way than, for example, the delivery times currently provided by online stores. In the day-to-day operations of a freight forwarder, it could provide recipients with a delivery schedule using an AI-based historical data analysis. In this schedule, the recipients would then find the most likely delivery time slot during which the local transport provider would deliver certain items to them. The companies, in turn, could adapt their downstream processes to this forecast.

Get there faster

Beat the odds

The goal of complex data analyses is to improve process planning so much that deviations – and thus execution errors – are virtually impossible. The analysis of vast amounts of data thus turns into forecasts that support a trouble-free process. This can be achieved when shippers use their production planning, incoming customer orders and actual production progress to automatically plan shipments and place transport orders. Customers will then receive their shipments in the shortest possible time. If this data is evaluated over a longer period of time with enough meaningful events, freight forwarders can then plan their capacity far more precisely. They will then know the exact number of local transport vehicles required for all pick-ups and deliveries. The most reliable results are achieved when all the partners in the value chain share their knowledge. After all, each additional piece of relevant information increases the probability of a perfect forecast and improves profitability.

Perfect forecast

Anticipate every wish

The face of logistics will change dramatically if everyone participates: accurate forecasts make it possible to perform most logistics processes before a customer has even decided to buy a product. It starts with intelligent warehousing, which holds the required quantity goods near the recipient. It includes dynamic resource-driven pricing as well as superior dispatching and trip planning. Through disruption forecasting, ideal capacity planning with optimal cargo space purchasing and perfect customer service with fully automated chatbots, goods will reach their destination by the most direct route at precisely the required time — while maintaining strict transport quality standards. We owe all this to real-time decisions made by self-learning machines – as long as we humans ask them the right questions.

When will you revolutionise your logistics world?

Stefan Seufert
Stefan Seufert

As a design guru, the software developer delves into logistics service providers' requirements like no other. He is passionate about exchanging information securely and efficiently and thus speeding up the physical logistics process.

Add a comment

Please add 8 and 2.