Artificial Intelligence in Logistics: Two Approaches to Improve Planning

Artificial Intelligence in Logistics: Two Approaches to Improve Planning
7 min read

The real-world use cases of Artificial Intelligence (AI) are expanding rapidly: from e-commerce to healthcare to security and fraud detection, AI seems to have something to offer virtually for every industry. It’s a bit surprising that AI’s potential hasn’t become more realized in core logistics operations, but that time is fast approaching.

To best leverage AI in logistics operations, decision-makers need to understand the basics. Best to start with the two AI approaches with the potential to bring the biggest impact on logistics.

Two AI Approaches – Overview

The first AI approach is called Machine Learning, which can best be understood as “statistical AI.” Machine learning is based on the premise that large volumes of historical, current and future data contain important patterns. The problem is they are hard to find with the human eye. Machine learning software is adept at discerning these patterns – though most software require some human input to operate.

The patterns that are deciphered with statistical AI are used to develop models, which become useful predictors of outcomes for businesses. This approach is known as “machine learning” precisely because the more experience and data the software has, the more effective it becomes at predicting.

People are still needed in AI planning – not to teach the AI software, but rather to be the executor of the action taken based on the options that the software produces. Depending on the level of access and authority someone has, they can give approval for decisions, stop processes entirely, or be mere spectators of the process.

AI planning is a second approach that has a useful application for logistics. It involves providing information on present conditions, possible actions to take, as well as outcomes – and it doesn’t require learning from experience the way statistical AI does.

People are still needed in AI planning – not to teach the AI software, but rather to be the executor of the action taken based on the options that the software produces. Depending on the level of access and authority someone has, they can give approval for decisions, stop processes entirely, or be mere spectators of the process.

If companies see the benefits of both approaches, the good news is they don’t have to choose. It’s possible to develop solutions that incorporate elements of both statistical AI and AI planning.

Maximizing Impact of AI in Logistics

To understand the logistics applications for these AI approaches in more concrete terms, it’s worth taking a closer look at a couple of specific uses.  

The first involves the quality of data, assessed by accuracy, completeness, timeliness, and precision. Historically speaking, the logistics sector often deals with data that needs to be cleansed. This is a problem considering so many downstream processes — like customer service, planning, personnel management, and inventory management — are affected by bad data. In that sense, data quality is not just another problem you need to solve: This is the problem you need to solve. In fact, a Deloitte study found that data quality was the main barrier to the effective application of digital technology in logistics organizations for almost half of chief procurement officers surveyed. Lack of data integration was the second biggest challenge.

Machine learning offers a solution for catching and even correcting data quality problems at an early stage. By combining large historical datasets with human feedback, machine learning enables logistics companies to predict actual data values when entry fields are left blank. That is to say, machine learning delivers an exceptional data outcome even when inputs are substandard. And of course, the models for how the data is filled in vary greatly, depending on how data points interact with each other.

AI Logistics

Secondly, let’s talk about planning: It is embedded in virtually every component of logistics, from transport consolidation and warehouse slotting to routing, pick and pack strategies, and procurement of all of the above. Planning is about making decisions in specific environments and for this reason, each process requires its own approach to planning and a relevant IT system.

Consider the complexity of truck planning, for example. Logistics planners have to assign loads to a list of trucks while considering the weight, volume, and layout of the loads going into each truck. They also have to consider services like temperature range, lift gate, and more. It might also be needed to decide which drivers will handle the delivery or to plan for returnable assets like pallets. In fact, there is a virtually limitless (but predictable) combination of loads that a fleet of trucks can carry. Needless to say, while planning the loads, companies are trying to reduce costs while maintaining service level.

This is where AI comes in: In this particular instance, AI planning is more effective than machine learning. The size of data is a bit more modest as well as structured. No less important, there are explicit choices and clear goals. Compared with traditional planning techniques, AI planning is far more efficient and effective at exploring the scope of possible decisions.

Two AI Logistics Approaches in Action

One example of a company that combines both AI approaches is Transmetrics, which offers AI-driven predictive optimization solutions exclusively for the logistics industry. Named among Top 5 AI Startups for Supply Chain Management by Business Insider Intelligence, Transmetrics has years of experience using Machine Learning algorithms to clean, improve, and fix the historical data of many logistics companies worldwide.

Presently, introducing modern data-driven technologies to logistics has its limitations, by and large for the fact that such data is often messy and incomplete. For example, logistics companies might have data about the weight or the density, but not have the dimensions of a particular shipment. In that case, AI-driven algorithms developed by Transmetrics can systematically go through the data, look and learn about how past shipments behaved and then they can come up with precise deductions about all the properties for each shipment. These AI algorithms might require only 5-20% of data being correct in order to create a training dataset, which serves as a basis for data cleansing and enrichment. Once companies have good quality data, it unlocks all kinds of opportunities for optimization based on predictive analytics to achieve much higher levels of operational efficiency and improving the bottom line.

AI Logistics
Picture Credit: Transmetrics

That is where Transmetrics uses AI Planning approach. Based on a mix of the already improved historical data and a set of external factors (e.g. seasonality, public holidays, weather, etc.), Transmetrics generates daily rolling forecasts of upcoming shipment volumes on the most granular level, weeks ahead. These forecasts are used for proactive optimization. By taking into account all customer requirements and operational constraints, Artificial Intelligence, and complex stochastic optimization algorithms help planners and dispatchers with recommendations on how exactly to adjust the operations. Depending on a particular business case, the software helps logistics professionals to precisely determine whether to decrease/increase capacity; proposes the most efficient mix of “own” vs. 3rd-party assets as well as one-way vs. round-trips; calculates the most optimal empty asset repositioning plan, the right levels of “safety stock” at each location, and more. Ultimately, it enables logistics companies to manage their business in a more intelligent way and narrow the supply and demand gap.

By using two AI approaches in parallel, Transmetrics helped Speedy, a member of DPDgroup, to improve their fleet utilization, ultimately reducing their costs by over 7%. In another project with NileDutch, one of the leading shipping companies focused on the Africa region, Transmetrics software helped to significantly lower the total costs for managing empty containers and reduce its container fleet size. All of this, thanks to the intelligent use of Artificial Intelligence as well as cleansed and enriched data.

The More You Know

Understanding the fundamentals of different AI approaches is important for executives who are trying to implement innovative AI-driven software in their logistics companies. Without taking even a rudimentary look “behind the curtain” at these two approaches, it’s hard to know what legacy systems need to be abandoned and where there could be an opportunity for improvement.  

Machine learning is a better approach to problems of large and unstructured data, where greater experience helps to improve software results. AI planning is better used when the business has clear goals and scope of decisions made to achieve them.

Don’t forget to consider the intended output for your problem. Machine Learning predicts while AI planning makes decisions. By understanding the basics between these two core AI techniques, you are more equipped to hit the ground running with the AI implementations you bring to your logistics company.

Transmetrics Demo