Reducing transportation problems and delays using Artificial Intelligence

We at Quicargo are connecting thousands of companies with transport needs to a transport company each quarter. We are handling thousands of transport requests each month, carrying goods worth hundreds of thousands of Euros.

As part of our vision, we would like to offer a service that is as flawless as possible. However, supply chains are delicate and complex systems. A problem in one link can cause a ripple effect, and it could eventually hit the consumer.

Importance of transportation incidents and problems

To understand this better, consider this simple example, which could be generalized to almost any modern product:

You want to buy a birthday cake from your favourite supermarket chain. The cake comes in a plastic package. The supermarket chain buys the plastic package from a plastic wholesaler, that is supplied by a plastic producer. If there is a problem in the transport between the plastic producer and wholesaler, like a delay of one day, it could affect the production of the birthday cake! You may think that the supermarket has a huge stockpile of plastic packages for the rainy day, but it takes a lot of space and effort to do so for every product. So oftentimes stockpiles are reduced to a bare minimum.

Transportation problems

In addition, when something goes wrong in a transport, additional time has to be spent to fix the problem. There are inevitable costs that have to be absorbed by shippers, carriers, Quicargo, or all.

Preventive measures

Transport incidents are costly, both in terms of time, and their effect on the rest of the supply chain. Therefore we strive to prevent these problems from happening in the first place. However, the question is, how do we know if a transport is going to be problematic before it really takes place?

If we only had a very experienced planner who could oversee each and every transport and predict for us which ones are going to be problematic, we could pay more attention to those transport and prevent the time and money consuming incidents.

Imagine a transport planner that has observed hundreds of thousands of transports, and has the processing speed and memory of a super-computer. When the imaginary super-planner observes a new transport order, they can immediately notice if something is suspicious about it. If the order has a tight pickup time window and is planned to be picked up in a less known area, the super-planner will raise a red flag: “This is the recipe for disaster!”

In reality, we don’t have such a super-planner who recognizes the looks and characteristics of problematic transports beforehand. But we can teach a computer to think like the imaginary super-planner. This is what we call machine learning.

Machine Learning and Artificial Intelligence to the rescue

We at Quicargo, have developed and implemented a machine learning model that predicts if a transport is going to contain an incident or a problem, like a wrong address, long waiting times, delays, or missing pallets. Here is an overview of the steps we took to achieve this:

Step 1: Statistical analysis

We studied our past transports meticulously. What could be the characteristics of problematic orders? Here is one example of such analyses:

We can see which characteristics of an order are more correlated with operational problems and incidents.

Step 2: Training the model

Let the machine learning algorithm learn the characteristics of the past problematic transports. We show both problematic and non-problematic past orders, so that the algorithm can find the decision rules that distinguish the two.

transportation problem analysis

Step 3: Predict

When a new transport order is introduced to the system, we show it to the previously trained model. The model will assign a “risk” of having incidents to the new transport order.

Step 4: Act and Prevent

When a transport order is predicted to be “risky”, we create tasks for our operation team, so that they can review the risky order with more care. We also create separate tasks on the day of pickup and delivery to ensure that everything is in order.

Results and outcomes

The result is a reduction in operational incidents and problems via smart prediction and prevention.

Using this machine learning model we can predict and label more than 63% of all problematic transports as risky, the moment that they are created, way before the ignition switch key is inserted into the truck and before the transport is even planned!

By being proactive and taking preventive actions on these 63%, we are able to reduce operational problems and incidents by at least 20%!

Going forward, we would be improving this machine learning model to make it even more smart and precise, and achieve an even greater reduction in transportation problems.

Written by Nima Maleki, Data Scientist at Quicargo.