By itself, digital information is meaningless. Ask a computer program a question regarding specific data, however, and you’re on the right track. With this in mind, there are three distinct questions that provide varying degrees of insight:
- Where am I today?
- Where will I be in the future?
- What should I be doing?
If you had to choose between the three, my guess is that you’d want to figure out how you should be conducting operations. Essentially, the last query aims to find the optimal method to performing certain functions. For a retailer, it may mean finding the best way to distribute goods. In contrast, a manufacturer may want to identify an enhanced assembly process.
“Prescriptive analytics are fundamentally similar to descriptive and predictive models, but are capable of estimating multiple outcomes based on given situations and actions.”
Descriptive, predictive and prescriptive
While questions No. 1 and No. 2 correlate with descriptive and predictive data analysis, respectively, prescriptive analytics juxtaposes No. 3. Lithium Technologies Chief Scientist Michael Wu, identified the distinctions between the three:
- Descriptive tools condense large amounts of information and summarize it by presenting bare-bones metrics. For example, a social analysis tool following this model may provide information such as the “number of posts, mentions, fans, followers” or page views a company may have aggregated over a certain time period.
- Predictive programs deduce the probability of certain events occurring in the future. They compare and contrast recent circumstances with historical trends using modeling, data mining and machine learning. Wu emphasized that predictive engines can “NOT tell you what will happen in the future … only forecast what might happen.”
- Prescriptive analytics are fundamentally similar to their predecessors, but are capable of estimating multiple outcomes based on given situations and actions. The difference lies in a prescriptive solution’s ability to recommend calculated decisions to users and change advice based on a feedback system.
For those who aren’t familiar with the term, a feedback system describes a loop in which a user’s action is fed through an analytics solution, which deduces how a company’s (or person’s) behavior will impact future outcomes. This is how prescriptive analytics works: It provides organizations with smart recommendations for how they should react to present trends and prepare for potential future outcomes.
How United Parcel Service is putting it into action
Network World Editor in Chief John Dix spoke with Jack Levis, senior director of process management and UPS, on the supply chain management provider’s use of prescriptive analytics. Levis maintained UPS deployed a predictive model in 2003, which served as an “assistant” to its drivers, allowing the organization to save 8.5 million gallons of fuel and decrease the number of miles driven annually by 85 million miles.
Despite these enhancements, the predictive model failed to identify anomalies, obligating drivers to figure out how specific circumstances should be handled efficiently. Furthermore, the analysis tool did a mediocre job of advising truckers on how to prioritize packages under deferred and premium services.
This is where UPS’ On-Road Integrated Optimization and Navigation (ORION) system comes into play. While the program has been deployed to one-fifth of the company’s drivers, it’s incapable of performing amazing feats.
“Putting it into perspective, the advanced math around determining and order of delivery is incredible.” Levis told Dix. “If you had a 120-stop route and you plotted out how many different ways there are to deliver that 120-stop route, it would be a 199-digit number.”
Behind this horsepower is a program that can recommend optimal decisions to drivers – in practice. Levis maintained that ORION is a long-term investment, one that will be equipped with more advanced features as time progresses. One such function would alert drivers to a problem before it even happens and then provide them with ways to avoid discrepancies completely. It’s certainly an exciting development, one that marks the beginning of prescriptive data analysis’ maturity.