Mining the Data

IBM Corp. introduces technology to help auto makers and fleet owners better mine the myriad of data available in a bid to chip away at the $14 billion per year that warranty claims cost the industry annually in the U.S.

Warranty costs amount to $700 on an average vehicle and are robbing auto makers of 1%-3% of profits, says Jan Beauchamp, general manager-IBM Global Automotive Industry.

She notes the industry recalled more vehicles last year than it produced, and there is a ripple effect as OEMs pass some of the expense down the supply chain.

As vehicles become more complex, with more features, electronics and software, combined with pressure to launch more products faster, the situation likely will intensify, Beauchamp fears.

Cutting the time it takes to process a warranty claim from problem identification to resolution can result in significant savings. An IBM analysis suggests shaving 10 days from this cycle could reduce the number of overall claims by nearly 5%.

There is no shortage of data and information in the auto industry — everything from warranty records to Internet chat rooms. The trick is to manage it all as a resource to identify and fix problems quickly and ultimately prevent expensive failures from occurring.

Beauchamp says IBM has been tackling the issue of reducing warranty costs in phases. It started with a solution to block out irrelevant data, followed by ways to analyze and predict the performance of vehicles.

The idea is to recognize problems developing in the field, such as which parts are most likely to fail given the expected wear and tear on a particular model, in order to create alerts and react to this early warning system.

The third element, Beauchamp says, is the ability to use all forms of data, not just structured information such as diagnostics. The key is to be able to search unstructured data as well, such as partial reports or impressions on an Internet blog, and eventually feed the data back into the product-development process.

“Warranty claims, maintenance records, call-center logs, repair requests, online chat rooms and blogs present an untapped source of useful information for auto makers,” says Beauchamp.

IBM's new technology, dubbed its Quality Insight Solution, “provides the opportunity to harness that information to detect problems — and head them off — much sooner than we are able to do today,” she says.

IBM created technology to support software that can process text within documents that accounts for 90% of the data being processed. It pulls in text data that could not be accessed in the past, Beauchamp says. IBM used the recent Automotive Industry Action Group AutoTech conference in Detroit to announce Quality Insight Solution.

The information technology company spent 18 months developing and testing its Quality Insight Solution in a pilot project with International Truck and Engine Corp., says Larry Lieberman, automotive industry senior manager-IBM Research.

The big-truck maker provided more than 1 million records of historical warranty information that IBM automated to see how much faster International could analyze data, Lieberman says.

Terry Stewart, International's project manager, says compared with the old methods of poring over reports manually, the new system fingers problems 10%-30% faster, depending on the data.

International has gone from reacting to a system failure to a proactive system to ensure quality, Stewart says. The data provides a baseline to ward against a problem getting out of control. Now that the pilot program is finished, International is convinced it is the way to go, both for the company and the industry, Stewart says.

IBM is convinced the industry solution to reduce warranty costs lies with open computer standards, so that data in any form, from any source, can be put into a central repository and made accessible to all in the chain, including suppliers and dealers.

The open standards include a common language, with a dictionary of terms and list of synonyms to use the central resource in order to find like-problems that may have been described using similar, but different, words, Lieberman says.