Analysis: Predictive analytics is new edge in shipping

Analysis: Predictive analytics is new edge in shipping

To be distinguished from competitors, the capability to exploit data is critical. Photo credit: Shutterstock.

Having access to data is key in the industry today. To be distinguished from competitors — who also have access to a vast amount of data — the capability to exploit this data becomes critical in achieving a competitive advantage. This, unfortunately, is much easier said than done.  

For example, when it comes to predictive analytics — a much talked about buzz phrase these days in the maritime industry and beyond — there are literally countless ways to arrive at predictions. Not all predictions are the same. Some predictions are more useful than others.

In fact, every single person in the industry has most probably engaged in predictive analytics in the past (without referring to it as such), but most likely using a very wide range of different models and methods. Some might simply calculate historical averages and use that as a forecast. Others might employ something fancier, such as linear regression, decision trees, or even so-called neural networks.

Indeed, predictive analytics is really only an umbrella term for a vast collection of models and methods to arrive at predictions that include all of the above (and much more). To further complicate the choice, new and improved predictive models are regularly being invented in academia that, unfortunately, might not be readily accessible to practitioners unless they are endowed with the appreciation for complex mathematical equations. Thus, the question becomes: although the industry agrees that predictive analytics can be useful, and that it can help alleviate many of the challenges the industry is facing, which model/method will give it the best predictions when faced with a particular problem? 

Unfortunately, there is no clear-cut answer to this question. Instead, it all depends on the specific challenge one is facing. For instance, two areas of application of predictive analytics that have recently been documented by are chassis demand forecasting and the prediction of when import containers are available for pickup at marine terminals. Although the chassis predictive model was able to significantly reduce the chassis repositioning cost, due to more-accurate chassis demand projections, nothing guarantees that the same class of models will give useful predictions when applied to forecast the time of container availability. The dynamics are completely different.

In other words, the particular challenge at hand dictates what can be used, as every predictive model comes with its unique set of assumptions/conditions under which it can be used. (It is to be emphasized that if a method can be used, that does not necessarily guarantee that it will actually lead to useful predictions.) Herein also lies a danger of predictive analytics: sometimes the industry sees predictive models/methods being used as a black box, without exactly knowing the assumptions/conditions under which the predictions are valid. It is true that black boxes will output numbers, but how much value should be attached to these “forecasts”? Consistently useful predictions are typically the result of a combination of industry knowledge, technical understanding of the model/method used, and mathematical modeling skills, i.e., the ability to translate real-world phenomena into mathematical equations.

One key fundamental question remains: can predictive analytics really tell the industry what the future holds? The bad news is that it cannot. Nothing can. No one can predict the future. Although this is rather disappointing, one has to keep in mind that the benchmark should not be the unattainable truth, but rather the predictions that currently inform the decisions made in practice. In other words, predictions have value as long as they are able to paint a clearer picture of the future than the image that is currently in place. Regarding the work in chassis demand forecasting: although the team was not able to predict chassis demand with certainty, compared with the chassis pool operator’s own predictions, the predictive model gave more-accurate predictions. As a result, chassis repositioning cost could be substantially reduced (by 80 percent). 

Thus, although the predictions were not perfect, they are perhaps good enough until someone else comes around the corner with an even better prediction. To conclude with some good news for the new year's start: there might indeed be a lot of room for predictive analytics to make a positive impact on the state-of-the-practice in our industry, provided that it is used creatively and correctly. 

ManWo Ng is an Assistant Professor of Maritime and Supply Chain Management at Old Dominion University.