When truck-maker Navistar debuted its OnCommand Connection in 2013, its open architecture promised to make the benefits of big data analytics available to fleet operators, regardless of which truck brand or telematics provider they used. Making it a free service in 2015 on its 2007 and newer trucks made its base telematics capabilities all but standard, and only ensured that it would grow significantly.
It triggered a tsunami of data.
“Our telematics on approximately 200,000 vehicles provides data on miles, engine hours, idle time, and fuel consumption, with a frequency as fast as every 5-10 seconds,” explained Andy Minteer, Director, IoT Analytics and Machine Learning. “We can add to that data from manufacturing, warranties, even weather data from NOAA.”
“Now we have this data coming in, what do we do with it?”
Fortunately, the company has been working on those answers for a while already, having chosen from the beginning to forgo outsourcing it, and instead building an internal capability.
“Combining the power of external resources with the knowledge we have internally lets us move faster,” said Dan Pikelny, VP, Analytics, who leads a team that reports to Navistar’s head of strategy, and then serves its internal clients. Minteer is the lead analytics evangelist for OnCommand Connection.
The analytics team is working on at least four keys to harnessing its data tsunami, and putting it to work across Navistar’s business:
First, it clusters usage in hopes of identifying service issues, but finding actionable insights can be complicated. Minteer recounted a run that identified a component failure among certain high use fleets, yet there’d been few corollary reports from the field. It turned out that team drivers kept those trucks in action 24/7, so the relatively small set of vehicles were acting like canaries in the coal mine and predicting problems for other fleets.
Second, it also looks for product design insights, and was considering producing a vehicle to satisfy the needs of low-frequency users (it could be made of lighter duty components, and therefore cost less). Pikelny explained:
“When we looked more deeply at the use cases, the low-frequency users were high-intensity when they were driving, so we waved off plans to build a lighter truck,” Pikelny said.
In another instance, learning that long-haul trucks were being used in off-road applications, like driving into cornfields to pick up harvested corn, informed changes to product design.
Third, it looks to flag service issues before they appear in warranty claims, primarily to improve customer satisfaction. Minteer said that the company captures thousands of “fault codes,” along with other vehicle data, as a way to track vehicle health and generates a graphical display that flags predicted events. This represents a “paradigm shift” away from dealing with customer complaints, and toward precluding them.
Finally, the team strives to keep its head out of the weeds. “We can look at the data by product line and production runs,” Pikelny said. “But we need to be careful how low we go, because there needs to be a business case for whatever need or opportunity we discover.”
The next key will be building models for understanding marketing/sales expenditures, with an eye to forecasting return on investment the same way the team has proven it can predict mechanical issues.