by Hollis Nolan | Dec. 13, 2022

Logistics is one of the planet’s most essential – and most complicated – industries. There’s lots of room for error when you’re dealing with a (literal) million moving pieces scattered around the planet, but data science gives logistics companies the opportunity to connect those dots.

The challenge

A cross-border trucking company based out of Laredo, Texas, offers full truckload freight service between all major metropolitan economic zones across two countries. Their fleet is composed of 250 trucks and 500 trailers. 

Prior to working with Insight Softmax, the company had a large amount of data, such as trip metrics, status updates, and dispatch information, but lacked company-wide access to the information. Separate departments were siloed from one another, which led to inefficiencies, duplication of work, and missed opportunities.

They needed a solution that would democratize the data and analyze it in a way that would create efficiencies, help the company achieve a goal of self-managed teams, and even improve driver morale.

The ISC approach

Making changes in the way data is used starts with an understanding of what’s available. To begin, ISC analyzed data points including driver retention and efficiencies. We asked questions around various factors, like the reasons drivers quit or the challenges that impact delivery efficiency, then built AI/ML-based models to answer them.

One discovery from our initial modeling was that the ideal road situation is one driver putting in 10 hours at 60mph. Calling that “100% efficiency” that would equal 600 miles per day, and 3,000 miles per week. What the actual historical data showed, however, is that drivers were actually only driving 2,434 miles per week – an 81% efficiency. This gave the client a clear business metric to track day to day, week to week, and a clear goal of what to shoot for. 

In addition to efficiency, driver retention is an important metric for long-term success. The relationship between scheduling, overtime, idle time, routing, and overall driver retention is complex. A Bayesian approach enabled us to build an AI/ML model that starts with the client’s existing knowledge as a jumping-off point and learns those complex relationships between variables. The model was able to estimate efficiency day to day, and driver turnover rate in the long term.

The client’s team could use the model to enter different sets of input parameters – fuel costs, wear-and-tear costs, driver overtime, and paid hours for idle time, to name a few – and see their impacts on predicted efficiency. IoT data from fleet vehicles enabled them to identify inefficiencies in individual pieces of equipment. AI allowed them to “play with the numbers” in order to see what worked and what didn’t in a hypothetical environment.

We also discovered during this process that the historical data was siloed, which meant that only a fraction of the company even knew this efficiency existed. To remedy this, we made the data shareable across the entire organization via a dashboard that allowed disparate teams to track progress not only for themselves, but compared to their peers.

Everyone working together toward raising that efficiency score evolved from a singular challenge to a group project that involved sharing techniques, offering cross-departmental advice, and solving the task at hand.

The AI/ML difference

Here’s how it works: In its initial framework, the company drew data from various inputs, such as status updates, trip data, telematics, dispatch and accounting. Those data points were fed through different analytics systems – telematics through GPS only and accounting through finance only, for example – and then outputted in various business intelligence reports that were good for their individual silos, but missed out on opportunities to cross-check pain points.

Implementing AI/ML programs didn’t necessarily change the input sources, but it had a marked effect on the output. Instead of running data points through separate funnels, everything was routed through a cloud-based, machine-learning program that was able to compile every data point into a data lake and analyze it in a new way. Instead of the previous reports, the new-and-improved ML-powered reports were able to deliver leveled-up data to teams across the organization in entirely new ways.

In addition to this new overarching framework, several self-initiated AI experiments were put into place, including rate prediction for freight pricing, document-vision AI (OCR and Microsoft Form Recognizer), and vehicle tag AI using the Google AI Vision API.

Real-World Results

By combining big data, advanced analytics, IoT, and geospatial systems with AI/ML, we saw some notable results:

• Creating a cleaned, organized data lake with a one-way sync of ongoing events allowed the source data to remain safe in its original format while at the same time enabling unprecedented visibility. Removing the silos that previously existed around various power structures within the organization allowed us to democratize access to the data. And with this unfettered access, the business was able to realize a company-wide goal of moving to self-managed teams.
• Analyzing driver preference scores per work activity (long-haul vs. short-haul, local vs. international), resulted in improved and better-aligned scheduling, as well as increased job satisfaction and employee retention.
• Calculating equipment efficiency scores led to the root cause of issues. This level of understanding led to an increase in equipment and service utilization, which translated to lower costs and increased billables.
• Streamlining the sales process reduced sales-quote generation time from 4 hours to 2 seconds.

Adopting an AI/ML Mindset

ISC worked in partnership with the trucking company to not only build and deploy AI/ML programs, but also to launch an internal program aimed at fostering curiosity and innovation around the opportunities it creates. This next step in the partnership led to a variety of pilot projects, including the creation of a freight-rates prediction engine that used machine learning to draw from rate history and other features to make predictions about future prices before the market changed.

While the process is still in its early stages, the adoption of data science mindset has thrown open the doors to opportunities and growth across the organization. 

If you’d like to explore how you can unlock the transformative value of your company’s data, drop us a line.