Predictive analytics: the ability to use well documented historical and accurate data to improve future logistics or fleet maintenance operations.
Data — especially quality data — is the key to predictive analytics. The good news (also the bad news) is that we have gone from data cups, to data pools, to data lakes, to today, where we are dealing with data oceans. In short, most fleets now have access to data — lots and lots and lots of data.
However, it is important to remember that not all data is created equal. The primary focus to ensuring the usefulness of collected data is to filter out the noise by identifying and eliminating anomalies while searching for repeatable patterns. These patterns are ultimately what your predictive analytics will be based on — detecting and understanding historical data resulting in an actionable decision.
Consider this example: You are reviewing your available data and detect that you are consistently having to replace the alternator on a subset of vehicles. Historic data is telling you that 60% of the time this alternator type, on this specific vehicle vintage, fails in the 65,000 to 85,000 mile range. The question you have to answer is: "Would it be economical to campaign all alternators in that subset of vehicles, before that mileage target band is reached, in an attempt to prevent a road call?" While there might still be life left in some of the alternators you replaced, would the proactive investment be offset by the potential savings of avoiding road calls for a driver, vehicle and product that was disabled on the side of the road?
From the logistics side, another repeatable pattern target would be traffic conditions. You can evaluate historical traffic patterns at specific times of the day (or seasons) to determine if there will be a high probability of traffic-related delays on specific routes that could impact your driver’s ability to get their loads delivered on time. On-time delivery is one of the most critical key performance indicators for most clients, and you can use data to help you gain an understanding of what may be contributing to any delays.
Is the delay caused by road conditions or weather? Does the driver get to the location on time, but can’t get into the loading dock? Can the driver not get to the next stop on time due to a pattern of delays at the previous stop?
You can use your history to understand what has happened in the past and predict what is likely to happen in the future. Armed with data, you can work with your customers and operations teams to reduce delays and adjust delivery expectations. Having data makes these kinds of conversations easier (and more constructive) because you can demonstrate what is occurring.
You also need to be open to sharing data that shows where you may be underperforming as well, and then work with the customer to find the root cause of the problem to develop a viable solution. Do not be afraid of what your data is potentially telling you, and do not be afraid to have those discussions with your client — highlighting what you are doing that is great, as well as where you can do better.
Just how much data do you need to make accurate predictions? A reasonable amount of data is two to five years worth depending on what you are attempting to evaluate. For example, if you are looking at equipment data and your tractors have a seven-year usable life, there may not be a need to look further due to consistent advances in technology. However, if you operate in a metro area where road conditions and construction change on a regular basis, looking at a shorter timeline for traffic metrics may be sensible.
If, on average, a vehicle, a driver, a customer or a resource falls into some regular pattern or trend but a small amount of data falls outside that pattern, you probably are looking at an anomaly. However, it can be worth it to dig into that anomaly.
For example, if your driver is making deliveries on-time 80% of the time, dig deeper into that 20%. You may discover that even within the 20%, much of the time the driver is close to being on-time but there may be other times when they are hours late.
Is this a driver issue? Is this a delivery point issue? Is it an origin/load ready issue? Is it a technology issue, such as a wrong address or geofence in your system?
Those outliers need to be removed from your analysis in order to get a better “game of averages,” or a more accurate picture of what happens on a recurring and common basis — but they should not be completely ignored.
Another important point to consider when it comes to data is its source. Is the source fully automated, like data from an ELD or telematics device? Or is the source a handwritten manifest from the driver? Where can I get the most accurate and reliable data?
If you use an enterprise asset maintenance management system your data should be fairly trustworthy, although there still may be some small percentage of errors or anomalies. If, however, you go into the shop and see handwritten repair orders, and handwritten pre- and post-trip inspection forms, you may want to be a little cautious about the accuracy and timeliness of the data you are compiling.
The goal of predictive analytics is to better serve the customer (and your own operations) by using data to support actionable decisions. On the logistics side this could mean routing more efficiently, delivering in a different sequence or dispatching the driver at a different time. On the maintenance side it manifests as saving the cost of a potential road call and managing expectations regarding equipment availability for future service.
If the idea of predictive analytics seems daunting, don't feel like you have to jump into the data ocean. Start with one area and find one pattern, one metric, one thing that you can act upon. Swimming in the ocean can be overwhelming, especially if you don't know what you are looking for. It's okay to dip your toe in and not get soaked, but you will eventually need to get wet.
Tom Poduch is Senior Director Solutions & Technology at Transervice Logistics. Poduch has been a logistics industry professional for more than 23 years serving in a wide range of roles from fleet maintenance, warehousing, field-based technology deployment logistics, engineering and consulting. His passion is to leverage technology by integrating with cross-functional systems, eliminating redundancy and maximizing data visibility.
