We’re often asked by our clients to explain the difference between “predictive” and “prescriptive” analytics. We thought that this simple example could clear up the distinction:
Imagine you’re getting ready for a road trip. You’re trying to get to your destination as quickly as possible, so you’ll want to know where you’re likely to run into traffic. After looking at the map and listening to some traffic reports, you pick out the highways that will avoid major cities and construction sites. Your predictive analytics help you anticipate which route will be fastest, helping you to predict what’s going to happen on your way.
For many businesses dealing with asset maintenance, this is where the analytics life cycle begins and ends. When a piece of machinery fails—an oil well, a wind turbine, or a jet engine—data is produced that helps predict future events; predictive maintenance allows users to take preventative or mitigative action.
With predictive analyses for your road trip, it’s up to you to take the traffic data and try to figure out the best route. What if instead, a program could analyze all of the data —construction sites, proximity to cities, accident reports, speed limits, and so forth— and use it to predict the best route for you? Accurately anticipating what will happen is one thing, but using that information to define a best course of action is another. And this is the essence of prescriptive analytics: using machine learning techniques to analyze traffic data in real time, Google Maps goes beyond simple predictions, determining the optimal course of action in response.
Artificial intelligence is a watershed for prescriptive maintenance. Natural language processing-powered algorithms can race through vast quantities of maintenance data, interpreting and organizing information from sensors and reports at the speed of light. With that data, neural networks can analyze millions of potential associations, identifying precisely what caused a particular event and why. In this way, a company’s previous maintenance analytics “Prepare for 75% chance of turbine failure in next three months,” can become, “Alert: anomalous behavior in turbine two. Likely cause: blade retaining key failure (95% likelihood). Replacement required.”
It was this type of analysis that allowed SparkPredict® to identify a turbine malfunction for a major energy company in February 2017, using sensor data to point out anomalous behavior in a single element of the machine. Traditional predictive models lacked the specificity necessary to detect the failure, which would have almost certainly gone on to cause catastrophic damage.
Machine learning-enabled tools like SparkPredict mean that companies shouldn’t settle for less than prescriptive analytics and maintenance. For more on how these tools are helping businesses be better, check out the full case study on failure detection in a combustion turbine.