RaceTrac

RaceTrac's journey into the world of AI and ML took about three years to complete. Before we could venture into this space, we had to take an assessment of our current environment to ensure that we had a solid foundation. By this, we had to assess the following:

• Technology Stack

• People

• Processes

From a technological standpoint, we moved from being 100% on-premise to a hybrid cloud environment. One of our key considerations in planning this migration was our people. We identified our internal technical strengths and weaknesses and provided proper training where necessary. We also brought in technology partners to build a robust team.

This assessment was not only within IT but also our overall data culture. We established a data and AI education program to educate all data users on proper data definition and usage, data visualization, and a basic understanding of our new technology stack (cloud, data lake, etc.).

“The implementation of AI/ ML is essential to the success of any company. When used properly, it can increase revenue, decrease expenses, and increase productivity, possibly leading to an increase in market share”

Data processes are critical to the success of an AI initiative. Without reliable and trustworthy data, your efforts will be in vain. Therefore, we established processes and procedures to improve data velocity, accuracy, and governance (data catalog, data owners, and stewards).

After establishing a “solid foundation”, we were confident that we were ready to jump into AI/ML. One of our most successful AI/ML efforts is our Fuel Pumps Down initiative. This was one of our first AI/ML projects. After completing the rolling out of IoT devices on our pumps, we began streaming data into our cloud environment in real-time. We use ML to identify patterns of each pump handle within a certain timeframe. From there, we established control limits based on several variables and factors. Once a fuel pump was outside of that control limit, the system would create a work order for a technician to investigate. This use case started off small in scope, but due to its success, it has grown to add additional use cases to address maintenance issues. We have a fleet of approximately 18,000 fuel pumps, and prior to this effort, 5% of the fleet would be down for maintenance. In less than a year, we were able to reduce this to less than 0.1%. This effort has also reduced the number of work order calls received by our help desk. Approximately 50% of our fuel pump work orders are now being created through AI/ ML. More importantly, this takes the workload off of our team members, allowing them to focus on creating a better experience for our customers.

As with all initial AI/ML efforts, challenges were encountered and expected. Outside of the new technology, one of our challenges was education. Very few people within the organization were familiar with this new initiative, causing skepticism. Education and communication were critical to the success of this effort. This education must span across all stakeholders, consumers, and project participants. We made sure everyone was properly educated, therefore confident in the success of this effort. We also had challenges with numerous false positives. False positives are bad, it's an opportunity for improvement. Through education and communication, we made sure that all business users understood that AI/ML is not an exact science, and the need to consistently train and re-train the models due to changing business factors. Some of these false positives uncovered business processes that the core team was not aware of, leading to improvements in business processes and data lineage and governance.

We are now adding features such as restarting the fuel pump remotely under certain conditions, as well as, evaluating computer vision AI at the edge. The implementation of AI/ML is essential to the success of any company. When used properly, it can increase revenue, decrease expenses, and increase productivity, possibly leading to an increase in market share. These are just some of the benefits of AI/ML. However, before you can make that leap, you must have a solid data framework and foundation in place.