May 11, 2004

Warranty Intelligence Software:

How does one launch a warranty software company that attracts John Deere and Whirlpool as early customers? And how does one attract the CIA as both an end user and an investor? Perhaps it takes equal parts skill, spunk, and luck.

The "Lean Quality" conference held in Livonia, Michigan on April 21 was by anyone's measure a great success for not only the warranty community and the automotive industry, but also for its co-sponsors.

In the space of one day, Carnegie Mellon University and Professor Sunder Kekre have established themselves as warranty experts. The Automotive Division of the American Society for Quality has set itself up as a warranty event organizer. And Attensity Corp. has set itself up as a major warranty software provider.

Attensity, founded in early 2000, can already name several major customers for its warranty claims analysis software, including Deere & Co. and Whirlpool Corp. And as the company's Web site explains, its investors also include In-Q-Tel, a private nonprofit venture group funded by the Central Intelligence Agency. Presumably, In-Q-Tel and/or the CIA is also using the company's software. So how exactly does a four-year-old company go about signing such famous clients?

Barton O'Brien, Attensity's vice president of business development, said it's all part of a plan. But it also involves a certain amount of being in the right place at the right time with the right product.

"What we do is called relational fact extraction," he said. Others call it text mining or text extraction. Like the artificial intelligence industry of 20 years ago, the field of text mining promises to free humans from the drudgery of having to actually read all the paperwork that passes before them. Text mining software will read it for them, and tell them when something important deserves a closer look. Ironically, the first few entities to pay the company and its concepts any notice were the overworked analysts of the CIA and the overworked analysts of the Deere and Whirlpool warranty departments.

Linguistics Research in Utah

Attensity is officially based in Mountain View, California, but thanks to the educational choices of co-founders Todd Wakefield and David Bean, it also has offices in Salt Lake City. Both executives have received degrees from the University of Utah, and Bean is still attending the school in pursuit of a PhD in computer science. It turns out that because of the religious practices of the Church of Jesus Christ of Latter-day Saints, Salt Lake City has become a world center for both linguistics and genealogy research. The linguistics research comes in handy when the church sends its young adults to all corners of the globe for two-year missions. The genealogy research is part of what they bring back, including birth and death records from virtually every government bureau and church rectory they can find.

It would be ever so convenient if some straight line could be drawn between the Mormons and the CIA, or between the educational choices of Attensity's founders and the marquee names of their first few warranty customers. But it doesn't appear that any links exist. So while the University of Utah is a magnet for linguists, it is not, apparently, also a haven for either spies or warranty managers.

"The CIA found us," O'Brien said, "because they were doing a nationwide search after 9/11, looking for new ways to more quickly assimilate facts and clues and things going on all over the world." By doing something as routine as scanning foreign newspapers, Attensity's software could pick up facts and patterns that would take a human analyst much longer to discern.

O'Brien agreed that any small and young company that starts off with the CIA, Deere and Whirlpool as customers is ever so fortunate, but he said it was more about being in the right place at the right time. Intelligence gathering and warranty analysis have little in common, except that analysts in both fields frequently are looking for non-obvious patterns in mountains of data.

"We came across warranty analysis because this is another area where the clues were hard to gather and pick up from current methods of structured information, and the unstructured technician comments contained a lot of good information that was not being used," he said. Attensity focused on large companies not only because that's where the money is, to paraphrase Willie Sutton, but also because that's where the complexity is.

At smaller companies, good warranty clerks can spot patterns and anomalies almost as soon as the paperwork crosses their desks. But at larger companies, there's simply too much paperwork for any one person to examine. So it turns out that Attensity's software really doesn't make sense for a company until its warranty clerks are drowning in a pool of paperwork, which means that there might not be more than a few dozen companies big enough to need this solution. Last year, Deere was America's 11th largest warranty provider with $321 million in claims, and Whirlpool was 14th with $248 million in claims.

"John Deere happened to give us a chance," O'Brien said. Part of the reason was the farm equipment company's reputation for giving new data processing concepts a shot, and part of it was Attensity's luck in making the acquaintance of just the right champion for its technology within the Deere's IT department.

"Whirlpool was even a funnier story," he said. An Attensity executive was watching a game on TV when he saw an advertisement for Whirlpool microwave ovens. But it wasn't telling people where to buy them; it was telling buyers where to bring them back for repairs. Figuring that the cost of a nationally advertised recall must be astronomical, the executive reasoned that Whirlpool must be having some massive manufacturing problems. "So we got in touch with Whirlpool's vice president of customer loyalty, and we started working with him."

Common Problems

Surprisingly, Attensity found that Whirlpool was having some of the same problems as Deere, namely that the standard codes on a warranty claim form weren't providing much guidance to those looking for patterns in the actual causes of product failures. It wasn't always a question of what parts were broken and needed replacement. Sometimes it was the way nearby parts were installed, and sometimes it was what an executive told O'Brien were the three L's: Leaks, Loose, and 'Lectrical.

"It's just that the codes are a very poor language," he said. "They don't provide much detail, and as a result, they also are very subject to judgment and error. The end result was that neither Deere nor Whirlpool were getting as much information as they really wanted about what was really going on. Therefore, they weren't reacting to problems as fast as they could."

Warranty software vendors are typically finding that sales cycles are very, very long, sometimes stretching for years. O'Brien said Attensity was able to speed things up a bit by finding a champion within the prospect company, but that doesn't happen every time. So now Attensity is embarking on a program of finding champions for its technology everywhere at once, through efforts such as the "Lean Quality" conference it co-sponsored last month with the Automotive Division of the American Society for Quality and Carnegie Mellon University.

"I came to the conclusion," O'Brien said, "that it was easier to sell the industry than to sell a single company. And the reason I say that is because the automotive industry is very internetworked. They all live together. They go to the same country clubs." And apparently, they also belong to some of the same industry associations and go to some of the same meetings.

Many of the major auto manufacturers also share a common approach to warranty claims analysis. The process usually ends up with all known problems ranked on a Pareto chart, with the most frequent problems to the left and the least common to the right. But there can be so many problems that the right side of a Pareto chart can have a very long "tail." Statistical theories hold that the most efficient approach would be to solve the left-most problems first. Following the so-called 80/20 rule, these theories hold that 80% of warranty costs are caused by the left-most 20% of the problems.

Lean Quality

In a white paper entitled "Lean Quality: The Coming Revolution in Automotive Quality," O'Brien argues that basically this rule is not true when it comes to warranty data. He argues that when there are a large number of problems, or at least a large number of ways of expressing the apparent cause of problems, patterns can go unnoticed because there are simply too many problems hidden in those long tails. If a manufacturer is continuously working from the left side, they may not get very far past problem number 50 or even number 25 in a given year. But there could be 700 or more problems on the chart, because the tail is so long.

Some of those low-volume problems towards the right-hand side could actually be the same problem expressed in dozens of different ways. In other words, the same root cause is manifesting itself through numerous failure codes, but because they're not all being grouped correctly as one cause, they never rise far enough in the Pareto chart to merit a closer look.

As an example, O'Brien described a real-world problem a customer had with plastic shavings in the fuel tank. The shavings, caused by the drilling and mounting done during installation, were in the tank because they weren't blown out correctly by the installer. But they typically manifested themselves through clogged fuel filters, fuel lines, or some other part failure. The codes on the warranty claim forms stated the problem was a clogged fuel filter, when actually it was debris in the fuel tank. And even if the technician described the root cause exactly and correctly in the "comments" field, nobody had enough time to read all these claims.

Another problem arose when a wiring harness got wet and shorted out the alternator. No amount of alternator replacements will correct that problem, because the failure of the part was not caused by a manufacturing defect within the part. Sometimes a perfectly good part is replaced by another perfectly good part, and the system still fails again because that part was only a symptom of the problem. "Some problems, such as electrical system problems," the white paper states, "are often spread over so many different part numbers that they appear as noise in the analysis."

The solution is right there in the data, but until Attensity's text mining software came along, problems weren't being characterized correctly according to their root cause. Instead, they were being described by the part numbers of the replaced or repaired components.

Because the descriptions of these problems were not machine readable, humans had to get involved. But because there were so many problems with apparently higher frequencies of recurrence, humans were never getting the chance to go far enough to the right down the Pareto curve.

In other words, warranty claims analysis isn't all that different from intelligence gathering. Analysts at Deere, Whirlpool and the CIA face the same problem: there aren't enough of them to read everything. And the solution isn't just a text reader. What they need is someone or something that can find a pattern or an anomaly in the data and can turn it into some useful preventive or corrective course of action.

The Company's Company

The CIA's interest in Attensity as both an investment and as a software supplier may at first seem puzzling, but after a bit of explanation it seems quite logical -- maybe even inevitable. The company that would become known as In-Q-Tel was set up in February 1999, after CIA officials paid a visit to MIT's Media Lab and came away impressed by the cutting edge research projects they saw. From the beginning, the company's mission was to fund companies engaged in the advanced development of information management products that one day could be used within the CIA itself.

In the past half-century, the CIA has on numerous occasions set up companies in the private sector and sponsored targeted research by the private sector. Air America is an example of the former, and the U-2 spy plane is an example of the latter. But In-Q-Tel is different. It operates more like a venture capital firm whose backers are intelligence officials rather than wealthy families. And unlike most venture capital firms, it funds the development of products its backers plan to use themselves.

An article written in early 2000 by Rick Yannuzzi in the Defense Intelligence Journal outlined the structure of In-Q-Tel. In-Q-Tel's founder, retired Lockheed Martin CEO Norm Augustine, established it as an independent Section 501(c)3 nonprofit corporation that follows the same rules and files the same forms with the IRS as any other nonprofit scientific organization. Its Board of Trustees, with ten members, provides guidance and oversight and sets the company's strategic direction and policies. The CEO, who was recruited by the board and reports to them, manages In-Q-Tel. The company has offices in Washington, DC and Menlo Park, CA, with a small professional staff and a group of business and technology consultants split between the two locations. Its first-year budget was reported at $28 million.

In-Q-Tel's current CEO, Gilman Louie, was most recently an executive at toy company Hasbro Inc., where he had no known links to the intelligence community but was nevertheless instrumental in importing Tetris from the Soviet Union 16 years ago. "Between Nintendo sales and PC sales, 70 or 80 million copies of that game sold," he told the New York Times magazine in 2002. "We even found out that Hillary Clinton loved playing Tetris on the Game Boy."

Commercial Crossovers

Examples of commercial applications that also support intelligence functions include data warehousing and mining, and statistical data analysis tools, which of course also are heavily used within the warranty industry for claims analysis. To date, In-Q-Tel has announced 27 separate investments, including Attensity. Under the heading of "Text Extraction for Threat Detection," In-Q-Tel describes Attensity as follows:

Attensity Corporation provides enterprise analytic software and services to government agencies and Fortune 500 companies. Attensity has developed breakthrough text extraction technology that transforms information captured in free-form text into structured, relational data. By powering link analysis, trending, exception reporting, and other advanced analytics and knowledge management applications, Attensity enables government agencies to dramatically expand their analytical capabilities in the area of threat detection.

Some of In-Q-Tel's investments also are backed by established Silicon Valley venture capital firms such as Kleiner Perkins Caufield & Byers and Draper Fisher Jurvetson. In-Q-Tel's other workflow- and text mining-related investments include Intelliseek Inc.; MetaCarta Inc.; Kofax Image Products; Stratify Inc.; Tacit Knowledge Systems Inc.; and Zaplet Inc., now part of MetricStream Inc.

Each of these companies counts both market researchers and intelligence agencies as customers, providing them with the tools they use to read, manage, and analyze large volumes of data. It turns out that both types of customers need to search unstructured data and filter the search results into a structured format. The same software that can help CIA analysts read foreign newspapers can help automotive engineers sift through tons of dealer correspondence. And in fact the Ford Venture Capital Group is an investor in at least one of these firms.

The In-Q-Tel Interface Center serves as the portal into and out of the CIA, and as the interface between employees with and without a security clearance. But day-to-day management is left up to In-Q-Tel, which operates at arm's length from the agency. In-Q-Tel is left to operate as it sees fit, establishing joint ventures, funding grants, sponsoring open competitions, and awarding sole source contracts. Business deals do not require agency approval and procurements do not need to follow government procedures such as the Federal Acquisition Regulations.

One would expect intelligence agencies to be among the most furtive of customers. A supercomputer salesman once said he used to write up orders for $3 million mainframes that were to be delivered to a parking lot in Laurel, Maryland, where the truck would be met by a driver who worked for No Such Agency. In-Q-Tel is a bit different, in that it doesn't arrange for deliveries to be made in empty parking lots.

Todd Wakefield, a co-founder of Attensity, told the San Jose Mercury News in a Nov. 2002 article that the In-Q-Tel team brought his Relational Extraction Server software to CIA engineers and let them play with it for six months. He said he had to take calls from people he knew nothing about, but he was surprised at how much they knew about him and his company. "They clearly understood more about our specialty than anyone else," he told the paper.

The Accidental Warranty Company

In the past, Warranty Week has written about the curious case of the accidental warranty software vendor. One company had a package that analyzed the basket of goods in a consumer's supermarket shopping cart. Suitably tweaked, it turned out to be a very good warranty claims analysis package whose first customer turned out to be one of the world's largest laptop computer manufacturers. Another company was helping an automotive company organize some faxes in the basement before time (and heat and light) rendered them blank. Along comes the TREAD Act and boom! They're a warranty claims reporting company.

Attensity is not an accidental warranty software vendor. But it is incredibly lucky to have such famous customers and investors after only four years in business. John Deere started making hay forks in 1825. The Upton brothers started what would become known as the Whirlpool Corp. in 1911. Next to these veterans, even the CIA looks young at 57 years old.

If there's a message here, it's not that it's who you know or even what you know. Sometimes it's who you meet, and who they know, and what they need. Some call it networking. Others call it serendipity. And in an industry as distributed and partitioned as warranty, a little bit of luck can mean all the difference.

Sign up for a free subscription to Warranty Week:
     subscribe     change of address     unsubscribe
Value Added Services
Assurant Solutions Shed The Light
Fulcrum Analytics