Validate Trailer Status with Machine Learning

Background

Our platform relied on third-party sensors and driver checks to verify trailer status before dispatch. However, drivers would sometimes pick up partially loaded trailers, leading to rejections at delivery sites where customers expected fully loaded trailers. These rejections reduced our profitability through unnecessary fuel costs paid to drivers, while damaging customer relationships. We needed a reliable way to validate empty trailers before pickup.

Goal

Ensure accurate trailer status verification before dispatch to reduce customer complaints and operational costs.

Why is this worth solving?

When trailers aren’t validated correctly, it disrupts our service and leaves customers frustrated with incorrect deliveries. Improving trailer status accuracy helps us avoid these issues, build customer trust, and cut down on the costs of fixing dispatch errors. It also makes life easier for our operations team by reducing the number of communications they have to handle.

Problem

We identified key issues in our trailer validation process:

• Third-party sensors failed to detect partial loads

• Drivers inconsistently performed trailer checks

• Manual verification was time-consuming and error-prone

• Customer complaints increased due to delivery rejections

Solution

We hypothesized that a photo-based machine learning validation system could ensure accurate trailer status reporting before pickup. To implement this, we integrated a workflow in the mobile app that required drivers to capture a photo of the trailer before hooking it to their truck. This photo was processed through a machine learning model to confirm whether the trailer was fully loaded or empty. If the model detected that the trailer was not loaded, it prevented the driver from proceeding, flagged the load, and raised a ticket for the operations team to manually resolve.

We closely monitored implementation with feedback from operations, fine-tuning the model by adding more images to achieve high accuracy.

Outcome

The machine learning-powered trailer validation system reduced customer complaints related to trailer status to less than 2% of total loads.