Building a crystal ball
NCR, which similarly collects information about the status of many of its products, including ATMs, self-checkout machines at grocery stores and movie theater ticket kiosks, is also using predictive analytics to get ahead of problems, says Mark Vigoroso, vice president of global services strategy and program management at NCR. The predictions indicate that a failure is likely to happen -- usually with a few days notice -- giving technicians time to get to the site with the right diagnostic and repair equipment before a failure happens, he says.
NCR has been doing this kind of prediction for several years, but Vigoroso says previously "it was a smaller operation with less precision, less accuracy and less coverage." That said, it is still the "early days of capturing the value of predictive services. Our effectiveness depends on how broad our predictive logic coverage is."
NCR has done some pilot programs where it marries data collected from its machines with other sources of data to draw different types of conclusions. For example, it has combined weather data with equipment performance data to look for patterns that might indicate that heat, humidity or cold are impacting equipment performance, Vigoroso says.
It has also started using cash management data, which it already supplies to customers of its ATMs, in new ways. NCR has long notified banks about nearby events like a major sporting game so that the bank can ensure an ATM will have enough cash to support users.
That same data, it turns out, is now helpful to NCR internally, because the company can use it to make predictions that help with machine maintenance. NCR knows how many card swipes the hardware can take before it begins to fail or how many receipts a printer will handle before it will have problems. Being able to factor in heavy usage related to events in advance allows NCR to more accurately predict when a component should be serviced -- before it fails.
"That's the part we're excited about. The new technologies that allow us to look across multiple data sets that allows us to crunch those numbers that we weren't able to do previously," Vigoroso says.
NCR is using Aster software from Teradata, a company that was spun off from NCR in 2007. Aster lets users create SQL-like queries to do complex analysis in a simple way, says Brian Valeyko, senior director of enterprise data warehouse and business intelligence for NCR. Analysts can make queries in an isolated environment without having to fear any negative effects on production apps, he says. NCR has built a unified data architecture that allows those queries to pull from Aster datasets as well as from other Teradata warehouses and Hadoop, he says.
The setup allows NCR to build new queries much quicker than it used to. In the past, it might take three to six months to build a new algorithm to do predictive analysis about a given component, Valeyko says. Plus, depending on the size of the data set, those algorithms might take days or weeks to produce results. With its current implementation, Valeyko figures the company can now run through that process in 20% of the time it used to.
That allows it to tackle new types of analysis, by correlating data, for example. Valeyko describes a scenario where NCR can now look at data about a printer component that's used in many different products. Rather than just knowing that the printer is having problems in all the products, analysts can discover, for instance, that it's actually only failing in products where it's combined with a certain kind of power supply.
For now, companies like Daikin and NCR have pieced together their sensor-analysis systems, using some off-the-shelf products plus plenty of their own development. Will it get easier? "Absolutely," says Avalon Consulting's Cagle. Once more work is done on easing the pain around unifying different kinds of data, putting together systems like what Daikin and NCR have won't be quite so challenging, he says.
This article, Predictive data, the real workhorse behind the Internet of Things, was originally published at Computerworld.com.