At this moment, the lights are on around the country in homes and workplaces. Most don’t realize, perhaps take it for granted, that those lights come on because of the hard work of dispatchers operating one of the most important man-made systems ever built, the national electric power grid.

Without oversimplifying the process too much, this happens because of a complex network of thousands of coordinated individuals in energy control offices across the country. Each office serves as a part of an electric grid nervous system of interconnected generators and transmission lines that deliver power.

For utilities, building accurate estimations of the electricity demand of their customers is critical. Inaccurate forecasts can result in significant financial losses and/or outages for consumers, but unfortunately, the ability to accurately predict demand, even within the very short-term, is extremely difficult to do. Operators, schedulers and marketers must make this kind of high-stakes decisions every day, often based on tedious, manual and error-prone methods.

Like many utilities across the country, Independence Power & Light found itself in this difficult position. The city’s department is responsible for predicting the amount of power IPL’s customers would consume each day and buy the required energy at the lowest prices possible. Ultimately, their goal was to ensure customers’ needs were met, while maximizing cost savings.

To do this, the IPL’s system operators used an off-the-shelf product alongside historical data. Every morning before the markets opened, they would study weather forecasts, recent activity and historically similar day data to build a load profile for the next day. This was time-intensive, manual work, but all too common in the utility industry. Due to the dynamic nature of power demand, it was also necessary for system operators to constantly update their predictions.

Ultimately, IPL was running an 8 percent average error rate. One of the system operations supervisors knew there had to be a better, easier way to build load forecasts that would save the city time and ultimately money.

They turned to a product from Pattern Recognition Technologies Inc., a company that provides load, price, wind and solar forecasts based on highly adaptive and dynamic machine learning algorithms. Using this artificial intelligence system meant IPL received highly accurate load forecasts updated several times per hour.

Within two weeks, IPL’s error rate dropped from 8 to 6 percent. The improvements continued as PRT’s algorithms were able to learn from past load behaviors and build increasingly accurate forecasts.

Within a year, the error rate was down to within 4 percent, resulting in annual savings of more than $100,000. With more accurate forecasts and their team’s time used more efficiently, IPL is creating value and cost savings that they can pass on to the consumers.

Nearly every industry has its periods of monumental advancements that transform how things were done, resulting in a more efficient, less expensive solution. The tractor improved farming, the automobile enhanced transportation, and I believe artificial intelligence is on the cusp of revolutionizing power. But these historical moments didn’t happen in an isolated vacuum – they took collaboration and communication.

It’s been said that those who cannot remember the past are doomed to repeat it. If we learn from others, we need not commit the same costly mistakes.

– Sogand Shoja is senior vice president of Drillinginfo. She holds BS and MS degrees in industrial engineering from Purdue University.