3 questions to consider before investing in AI

This article was originally published in the second issue of FreightWaves’ Supply Chain Playbook.

By leveraging AI technologies, shippers, carriers and logistics providers can reduce costs, drive operational efficiencies and, in the end, provide a customer service experience that has consumers coming back for more.

While there is significant opportunity for greater productivity by using AI tools, these are not systems that you add to your operations today and expect a return tomorrow. The success of these complex technologies requires careful planning and preparation.

Here are three considerations to make before infusing AI technology into your logistics operations.

What are you trying to solve?

Shippers and logistics providers should first identify their most pressing operational and service obstacles and work with their management teams to thoroughly examine where AI could make a difference.

“Generally, companies tend to not know what they need or what data we can give them that will help them be more efficient,” Matt Dzugan, former director of data science at visibility company project44 and now director of data for Muck Rack, told FreightWaves.

Dzugan explained that when working with customers on their ocean visibility needs, it takes deep discussions into customer service operations to find what aspects of their work could use AI-driven efficiencies.

“Let’s say you want to know where your ocean container is. No problem. Here is its location. But why did you want to know that? Maybe it’s because you want to know if it’s running late so you can take action for your customer,” he said.

According to Dzugan, the process of building AI models to meet those requirements can vary significantly.

“If I am trying to build a model that can help you flag an anomaly or I am trying to help build a model that will give you the ETA down to the nearest decimal point or I am trying to build a model that tells you how many deliveries are going to arrive at your warehouse on a certain date, I would build those all different and I would approach those three problems in three different ways,” he explained. “You should be very thoughtful and intentional about what you are actually trying to improve with AI.”

Identifying the most pressing business problems is crucial for shippers and logistics providers seeking to implement AI technologies. A thorough analysis of current operations, including customer needs, can help leadership find where AI can have the most significant impact, such as optimizing shipping routes, reducing inventory costs and improving demand forecasting. The in-depth analysis also can help prioritize your investment, delivering tangible benefits and demonstrating a return on investment to stakeholders.

How expansive and clean is your data?

Once you understand the areas of your operations that could utilize AI technologies, it’s time to tackle your company data strategy.

AI tools rely on vast amounts of data to identify patterns and make appropriate predictions. Shippers and logistics providers must ensure they have access to high-quality data that is relevant to their business needs.

Hunter Kreshock, senior full stack developer at FreightVana, explained that these data strategies can also include how you collect, store and manage your data so that it stays clean and unbiased.

“For example, let’s say you are building AI predictive modeling for hiring, but the majority of your new hires are men. That predictive model would likely become biased towards men over women recruits,” Kreshock said.

This is why you want to collect as much data as possible around the operations you are solving. Another example is if you are building a pricing algorithm but all of your data consists of van loads running from the Southeast to the Northeast. The pricing tool will not accurately price reefer loads running from the Northwest to the Midwest.

Once you have that data, you will need to continue to feed your AI predictive models with more data as time goes by. Much like a human who does not continue gaining knowledge throughout his or her life span, your AI tools can become partisan.

“You do not want your [AI tools] to get used to your data because it can become extremely predictive and become biased. We call this overfitting,” said Kreshock. “The tools basically become stuck in this bias and give you super consistent outputs, but it’s not really giving you the results that you want. It goes too far in one direction because you gave it basic data.”

Are your employees trained for your AI future?

A big factor in obtaining clean, unbiased data, and acquiring more and more of it as your AI models grow and learn, is building a data culture within your company, from entry-level to C-suite roles.

Training and development are critical to AI tool success. Shippers and logistics providers must ensure that their employees have the necessary skills to work with AI tools effectively, and this may require additional investment in training programs and the hiring of new talent with expertise in data science and machine learning.

Kershock also points out the importance of cross-training and having both customer- and carrier-facing roles working with a data scientist on solving these problems.

“If you have hired someone who knows machine learning, get them involved in your business,” he said. “Get them involved with the type of data they will be handling so they can understand your problems. It’s very important to understand these problems or else you are not going to get the results that you want.”

Furthermore, it is crucial that employees understand their roles in collecting this data. For example, if a shipper pays a carrier $100 for detention but adds that to the load’s linehaul and not as a separate line item of its own, future predictive pricing tools could be greatly affected.

Dzugan expounded on that problem, pointing out data-driven roles that have been created to solve this exact scenario.

“There is now a whole subfield that exists for this called data governance because we have realized as an industry that solving data science problems can be very valuable but boy, you can’t do it if you have people going willy-nilly in forms and entering data in all caps or adding commas in random fields,” he said.

“Not only do we have to enforce that from a people side, we also have to enforce it technically. We have to build out software to detect if this data is high quality or not. When we were in data science school, no one was telling us this. Now if you search for data governance roles, they are huge for this very reason.”

While these AI technologies have the potential to transform the shipping and logistics industries, their success takes careful and strategic planning. By asking these questions, you can position yourself to compete in the evolving digital economy.