How data mining helped GM limit a recall to 4 cars

Consider the following scenario.

The owner of a Chevrolet brings the car to a dealership, saying its brakes have been locking up. An engineer discovers that a brake valve in the car has a hole in the wrong spot because it was loaded into the wrong machine in manufacturing and cannot properly relieve the brake fluid pressure when the driver slams on the brakes.

In the past, an automaker might have recalled the model and inconvenienced tens of thousands of owners, despite thinking just a few cars had the defect. Or it might have let the defect slide, putting a few customers at risk of a crash.

But when that exact scenario unfolded this year, with a 2012 Chevrolet Volt brought in for a warranty repair in Europe, General Motors had a third, better option.

On May 30, the automaker assigned an engineer to dig into a database that tracks the parts used in its cars, and it collected manufacturing records from the supplier -- in this case, TRW Automotive. Within a month, GM had identified all the cars on U.S. soil with the faulty valve, called their owners and sent a formal notice to the U.S. government.

That is a quick turnaround, but what was more striking was this: The total number of cars touched by the U.S. recall was four. Not 4 million. Not 4,000. Four.

Maureen Foley-Gardner, director of field performance evaluation at GM, said the automaker has used its database for 20 percent of field actions this year, up from 5 percent in 2012, and watched as the average action has shrunk by 40 percent.

"It has become crystal clear to me that it's having an impact," Foley-Gardner said.

Not all of the actions necessitate a safety recall.

In one case this year, GM found that a defective transmission clutch plate was causing an annoying rattling in the Chevrolet Camaro. The company found the 18 affected cars quickly enough that 10 were still on dealership lots.

Tiny recalls of that kind are growing across the industry, experts say, as automakers, like drug companies and food manufacturers, build sophisticated data-mining operations to guard against costly and reputation-crippling recalls.

The auto industry spends $45 billion to $50 billion a year on recalls and warranty claims, according to data analytics company Teradata Corp. Teradata tells automotive clients that its track-and-trace software can cut warranty costs by 10 to 35 percent, administration costs by 25 to 45 percent, and compliance costs by 30 percent.

The main tools in the toolbox are bar codes and radio frequency tags.

The average car has about 15,000 components. Many automakers already trace hundreds of those by scanning parts on the assembly line.

GM started with big, expensive components such as engines and transmissions and is working its way down to smaller, less complicated parts. There are limits.

"It would be really costly to scan every nut and bolt and screw to the vehicle identification number at every assembly plant every day," Foley-Gardner said.

But the number of parts being traced is rising by the year. Another reason is that it helps in dealings with regulators.

David Strickland, the administrator of the National Highway Traffic Safety Administration, gushed about the track-and-trace effort in a recent interview, saying car companies are solving puzzles that previously might never have been solved.

He recalled a head-scratcher of a problem that one automaker -- he wouldn't say which one -- managed to contain with data mining. "It was just amazing," Strickland said. "It was a particular mixture of particular chemicals for a particular component at a particular factory at a particular time of day at a particular change of shift. But they nailed it down.

"The bracketing of this problem, I don't see how in the world they could have figured that out. But they did."

As automakers and suppliers have gathered more data points on their cars, they have had to beef up their software to handle the glut of information.

Nissan, for instance, added a feature called "bread crumbs" about 18 months ago to make the most of track-and-trace data, said Jim Blenkarn, senior manager of field quality investigations at Nissan North America.

The name of the feature is a reference to the fairy tale "Hansel and Gretel," in which the heroes leave a trail of crumbs to keep from getting lost in the forest. The new feature, similarly, is an attempt to help Nissan's engineers stay on track.

Blenkarn said that at Nissan, each component in a car is assigned to an engineer. That engineer is supposed to keep an eye on the warranty costs, customer complaints and reports of defects for his or her assigned parts.

That used to mean lots of customized searches on Nissan's database. But now, they can program "bread crumbs" to update automatically, giving them real-time data.

"The ideal circumstance for us is to never have to fix a car," Blenkarn said. "But the fewer cars that are impacted, and the quicker you can act on it, the less impact there is on customer satisfaction."