AI in the trucking Industry: Forget about self-driving; the real value of artificial intelligence in trucking has nothing to do with replacing drivers.

It is hard to go online without seeing “AI,” or artificial intelligence, everywhere. On the hype-o-meter scale of 1 to 10, AI is about a 100 right now. It seems that nearly every day, another company is touting that they are using AI to change the world. It is not all that different in trucking, especially when it comes to self-driving. Although we could fill up this entire section with a conversation on self-driving, AI, and trucking, let’s talk about how AI matters in the non-self-driving part of trucking — which I think is much more interesting.

First, a small primer on AI. At its core, AI is just a mathematically driven prediction model — albeit a very complex one that, up until five years ago, required a super-computer and trillions of complex mathematical calculations.  This is why you hear so much about AI now, because you don’t need a super-computer anymore.

Prediction models are nothing new; they have been around for centuries using legacy statistical methods. Every time a truck picks up a new load, some computer process or person has “predicted” that the load is going to make money for the trucking company — even though you don’t know if weather, maintenance or any of hundreds of things could affect how profitable that trip is.

What makes AI different from legacy statistical models is how many inputs can inform the prediction. I know that last sentence probably made no sense, but stay with me here. Once you have an “a-ha” moment on what AI is, you will see its potential in a whole new light.

In any prediction model, you take a set of inputs and use those to generate some kind of prediction. For example, with the weather, you can take yesterday’s temperature, barometric pressure and precipitation to predict what today’s weather will be. That is a fairly simple model. Each of these inputs will inform what the predicted output will be, in this case whether (pun intended!) it will rain or not. The challenge with legacy statistical models is that with each new input, your model becomes exponentially more complex — so you are limited on how many inputs you can use.

In the case of the weather, super- computers and AI started being used decades ago because there were so many inputs: not just the weather yesterday, but how about the weather last year, the year before and decades of historical weather data;  the direction of the wind yesterday, last year and 50 years back; the weather a few hundred miles away; the weather the day before yesterday, and the day before that, and on and on. Pretty soon there are thousands of inputs that can be used to predict the weather.  You can’t have that many inputs in a legacy statistical model — it is just too complex to manage. With AI, though, it is a different story — you can have hundreds, thousands, even hundreds of thousands of inputs in your prediction model.

Think about that for a second. Predicting anything is very difficult because life is complex and there are hundreds and thousands of factors that affect everything. Until AI, there was not a way to take into account all of those factors. With AI, though, we can now predict things that were not predictable before.

So why does all this matter to trucking? Well, simply put, trucking is one of the most complex industries in the world.  Think of a simple thing like picking which load to pick up after you have dropped off a certain load. Wouldn’t you just take a load that is close to the one you dropped off? Well, it depends; maybe that load is going to a city that has a bad spot market right now, maybe the driver needs to get home for a certain event, maybe the truck needs maintenance, maybe the weather risk is really bad in a certain part of the country even though the rates are really good. Just that decision alone is super-complex, with numerous factors that have to be considered. Now, take a company that is doing that for 30-40 (or even hundreds of) trucks every day — how do they decided the best assignments to make?

Because assigning loads to drivers is so complex, mathematicians call it a combinatorial optimization problem — which means that there are literally an infinite number of possible solutions. Because there are so many factors (i.e., inputs) that can affect the ultimate success and profitability of a certain trip, AI is uniquely capable of helping to solve that problem. 

Trucking is unique in that there are many problems of significant complexity: In addition to assigning loads to drivers, deciding on optimal maintenance schedules, balancing time on the road versus lowest cost operating speed (i.e., if you go faster and drop off the load sooner, will that ultimately make you more money?), predicting how long it will take to deliver a certain load — and the list goes on and on.

Now, you can’t just throw all this data at an AI engine and have it tell you the answer; it is much more complex than that.  You have to gather very large amounts of historical data and “train” the AI to the outputs (in this case, profitability and productivity) that you want. For example, what were all the possible load assignments over the past year, what was chosen, and how did each of those trips turn out? Gathering that data (if it even exists) takes a lot of time and organization. In fact, the biggest challenge of many AI projects is getting the right data so that you can effectively use AI to solve your problem.

Over the coming years, you will see AI take a bigger and bigger role in trucking. The current driver shortage only increases the pressure to have AI solutions help companies be more efficient to make up for not being able to hire enough drivers. In one study, a company’s average daily mileage was 465 miles per day, yet their top drivers were averaging 650 miles per day — and there was no noticeable difference in the loads assigned or the routes taken. Imagine if that company could get every driver producing 650 miles per day. They could do the same number of loads with significantly fewer drivers, which goes a long way toward addressing the significant driver shortage problem in trucking.

It is an exciting time to be in trucking. The demand to move goods and supplies across the road has never been higher.  At the same time, new technologies like AI will enable trucking companies to deliver more loads faster and more safely — and leading trucking companies are already starting to gather their data in a way that can leverage AI. If you have thought AI doesn’t apply to trucking, it might be time to think again and position your company to be ready — because your competition is already doing that.

Cory Linton is the CEO of Edify.ai, a Utah software company that is using artificial intelligence to help trucking companies maximize efficiency and productivity.  He can be reached at cory.linton@edify.ai.