Artificial Intelligence is being discussed more often in the world of logistics as a panacea for a range of critical business challenges. There is also much more talk about a branch of AI research called machine learning (ML).

You hear those terms more often because a combination of improving algorithms and faster computer hardware is allowing us to deploy technology more cheaply and work across much bigger problem sets with greater success.

There are a few things to understand about ML, before working out a budget to pay for the technology and its implementation.

A machine doesn’t learn anything on its own

For ML to be effective, it is recommended that training experiences allowing the machine to learn from the same data that a human would use to perform their task(s). You also need to precisely identify the tasks leading to the decision-oriented function you want to outsource to the machine.

With fierce competition in the logistics space, companies need to be swift and think of AI as the helping hand in the process of digital enablement. Whether you are an established company or a newcomer, everyone can access new technology. A combination of excellent customer service, global network reach, along with harnessing information technology, will provide the optimal combination.

Examples of competitive advantage with better use of technologies include:

  • Instant rates, instant quotes, and instant online booking,
  • Shipper-specific portal/dashboard including real-time shipment tracking,
  • Digital/analogue documentation processing combining ML and robotic process automation (RPA).

Say you chose a task starting from the input of a shipper’s origin and destination information to the output of an instant customer quote (let’s call this OD-to-quote). There should be enough knowledge within your company to ensure training includes OD data being paired with the correct quote. Keep in mind, quotes could consist of multiple transportation modes and value-added services, so it is more complicated than instant rate services provided by ocean carriers.

Let’s now assess your opportunities to harness ML correctly

Since ML suits functions that feature well-defined input leading to a well-defined output, automated “OD-to-quote” would meet this requirement. Ideally, ML will learn to predict accurate quotes associated with any given input and will handle most typical quoting requests. By assessing a history of past quotes and current external data, it will even work out the price point at which the requestor will most probably accept the quote on the first attempt.

ML doesn’t need to know the exact process that happens between the input and output. Clearly described goals are more important than the reasoning process by which your current team achieves the same purpose. After all, mimicking humans might not result in optimal performance because we also make imperfect decisions. In our example, having clearly defined metrics for performance could mean that you ask the machine to optimise the quote with the highest yield rather than to obtain an approved price more speedily.

Diligently review every internal function one-by-one to validate their suitability for ML applications and avoid making pre-emptive assumptions. If you encounter a function that looks too difficult, don’t ignore it. Critically assess why it seems too complex and difficult. Have you made it like that and are afraid to touch it? Did your people performing this function end up developing workarounds with the help of a spreadsheet and a few interns? I have seen it happen all too often. Digitalised freight forwarders don’t have the same qualms, and they won’t pause. You don’t want to come to that gunfight armed with a knife.

Prepare for your ML adventure

Keeping the above observations in mind and armed with some patience, you should be ready to begin your ML adventure. Fitting the right algorithm will take some time, but a skilled team will eventually end up with the accuracy considered optimal for the function you selected. Don’t count on perfection right away. ‘Practice makes perfect’ applies to machine learning too.