Rather than selling warranty analysis software, PolyVista sells analysis software that can look at warranty claims as easily as it can look at airline flight data or oil and gas trading records.
When we first checked in with PolyVista Inc. two years ago, the company was basking in the glow of the World Class Solutions Award it had just won from business intelligence magazine DM Review for a warranty analytics system it installed at the Personal Systems Group headquarters of Hewlett-Packard Co. in Houston.
It wasn't really a warranty analytics package, though. Rather, it was a software package that could dig into a huge volume of data and return a list of previously unnoticed patterns and anomalies that would be useful to production and engineering types who were looking to reduce product defects, boost quality, and reduce the amount of warranty claims. A year later, in fact, we dubbed PolyVista to be one of several accidental warranty software vendors who started out analyzing something else but ended up analyzing warranty claims.
Warranty Analytics Software
For each of those annual profiles, we talked with Bob Ford, PolyVista's vice president of technical services. This summer, we chatted with Shahbaz Anwar, founder and CEO of PolyVista. The company continues to be focused squarely on business intelligence software development, he said. And it continues to market its technology as a warranty and quality analytics solution. But in the years since HP signed on, PolyVista has had more luck selling its analysis package to decidedly non-manufacturing types for non-warranty applications.
Ten years ago, Houston-based PolyVista began developing a package that could analyze the contents of a supermarket basket and turn that data into meaningful insights into buying habits and behavior. From 1995 until late 2000, PolyVista was more or less in development mode. "It took us that long to get the software right," Anwar said. Sales and marketing efforts began in earnest in early 2001.
"We thought that retail was where we wanted to go," he said. "And those were also the Internet years. So we were also looking at Internet-related technologies such as the Web and e-commerce. So a lot of our focus was transactional or commerce-related -- retail over the Internet."
But then -- and here's the accidental angle -- somebody knew somebody who knew somebody at the also-Houston-based Compaq Computer Corp. There was no harm taking a meeting to discuss Compaq's need for a dashboard-style console for executives who wanted an instant and comprehensive view of conditions.
That meeting quickly turned into a discussion of what Compaq really needed: a package that could help them not only write reports about their call center, but also perform analyses of the calls. They didn't need software to help them merely count or time the calls. What they needed was software that could help them make sense of what customers were telling the agents during the calls, about the products they had bought.
PolyVista took away some call center records, trying to make some sense of it so that Compaq could improve its customer satisfaction levels. And of course call center data has a high correlation with warranty claims data, so PolyVista started looking at that database too. They quickly realized that customer satisfaction levels and warranty costs have an inverse relationship: the more was spent on warranty, the higher the satisfaction ratings. The more cost-cutting on warranty, the lower were the satisfaction levels.
"We truly believed that if we could merge their call center data with their warranty parts data, that we could get a head start on identifying areas that are beginning to see problems, and relate that back to both service and part requests, and track how particular models were doing," Bob Ford said at the time. "So we pitched to them was an analytical solution more so than a visual dashboard solution."
The rest, as they say, is history. Compaq ditched the dashboard and bought PolyVista's solution for its laptop division. Compaq merged with HP, and the Houston facility became the base for HP's Personal Systems Group. PolyVista's solution was picked up by HP's desktop line of business, as well as its printer group.
Anwar said PolyVista still counts Hewlett-Packard Co. as a flagship client, and still believes high-tech manufacturing companies of all shapes and sizes can make use of PolyVista's business intelligence solution. But the next few major customers have come from outside that industry. Indeed, they aren't manufacturers at all. One is British Petroleum plc, the global energy conglomerate. The other is Southwest Airlines, the Dallas-based no frills airline.
Needle in a Haystack
As with HP, PolyVista sold BP and Southwest a solution that can examine a large volume of data and return a meaningful list of patterns and anomalies, like finding a needle in a haystack. Tom Landry, a staff fellow at HP, has been quoted as saying that the package can do the work of a hundred analysts.
"The example that I like to use is finding the needle in the haystack," Anwar said, "when you don't know what the needle looks like, but the needle has a lot of value."
For HP, he said, the needle takes the shape of a warranty or a quality issue that manifests itself through a new pattern or anomaly observed in the call center or warranty claims data. "So it's a new relationship that they've never seen before. And since they've never seen it before, they don't know what it looks like. But they need to identify it very quickly and then do something about it."
For BP, Anwar said the PolyVista system helps energy traders within the company comply with government regulations and company policies in a post-Enron world. "They want to monitor those transactions, making sure that nobody is trying to do something that is not kosher," he said. PolyVista's software also helps BP's own management keep tabs on the risks they face, so that one day they don't wake up to find out they lost a billion dollars on a trade that went bad somewhere in the Far East. In that case, the needle would be a trade that poses some significant unseen risk, camouflaged among thousands of more routine trades executed by hundreds of traders.
Southwest, meanwhile, uses PolyVista to analyze safety reports made by pilots. The Federal Aviation Administration has pilots on a kind of honor system, where if they promptly report minor errors or deviations they're forgiven. For instance, a pilot might be told to cruise at 30,000 feet, but for whatever reason he ends up cruising at 31,000 feet. When he lands, he's supposed to write an entry in an aviation safety reporting system.
"So all the anomalies -- all the little things that can go wrong -- Southwest wants to report them to the FAA and it wants to capture them in its own database," Anwar said. "Southwest feels that all these anomalies are safety-related, and it needs to do something in order to avoid an accident down the road."
The reports include both structured data -- flight number, location, type of equipment, etc. -- and unstructured data in the format of a pilot's own description of what happened. Therefore, any system that aims to analyze those reports need to process the mix of structured and unstructured data much like is done with warranty claims or energy trades. Southwest needs to be able to look for trends at a certain airport, in certain weather conditions, or perhaps with a certain pilot. "For whatever reason, they want to be proactive; to get an early warning of a relationship that's brewing, that they need to do something about," he said.
Early Warning System
Anwar doesn't formally call it an Early Warning System, but that's exactly what it does: provide a manufacturer with an early warning of production problems. And he doesn't call it Text Mining, though the system does prowl through free-form text looking for evidence of patterns and anomalies.
"We want to integrate free-form text and structured data -- bring it together and then present it to the user," he said. "We feel that when the text is embellished with the structured data, it becomes more meaningful." For instance, assume that a disk drive is used in four different models, but is causing problems in just one model. Such a pattern would be found faster by looking at the free-form text for descriptions of the failure in combination with the structured data that lists the model number.
Early Warning Systems and Text Mining Systems have made their earliest and deepest inroads in the automotive industry, Anwar suggests, because many of the manufacturers in that industry are still using old mainframe-based systems that put a premium on disk space. Because disk space is (or was) so expensive, the systems typically allocate less room per field.
For instance, remember the old mainframe-based email systems that insisted on everybody's name being eight characters or less? That was driven by the economics of disk space. And for those of us old enough to remember Y2K, the whole faux panic was driven by programmers' use of only two digits for the year field, again in an effort to save disk space.
Translated into a legacy warranty claims processing system, the effort to save space typically left little room for free-form descriptions of problems or failures, forcing technicians to use abbreviations and shorthand to express themselves in 128 characters or less. Even when forms are populated on a PC, they're going to be processed by a mainframe, so those field length restrictions remain in effect across the automotive industry.
Automotive manufacturers, therefore, are typically looking at heavily abbreviated descriptions of problems whose very brevity obscures and prevents the discovery of interesting patterns and anomalies. Throw a text mining solution against a mound of data that's in poor shape and it's sure to discover some interesting patterns. So it's no surprise that automotive manufacturers have emerged as enthusiastic early adopters of the technology.
In contrast, computer manufacturers are highly unlikely to be using 20- or 30-year-old hosts for data processing. They're less likely to artificially limit text fields to 128 characters or less. So computer repair technicians are free to write as much as they want to describe a problem.
"I think it's a matter of technology maturity," Anwar said. "But I think that as automotive folks redesign and redo their systems, I think this constraint will go away." Then the automotive folks also will need to find a way to extract trends from lengthy descriptions of failures. As the data gets better, so will the tools.
For better or for worse, most business intelligence system vendors tend to concentrate on one industry, where they sign one client and then go after all its competitors. In contrast, PolyVista has gone after a problem -- finding the needle in a haystack -- and has adopted the system for three very different needles at HP, BP, and Southwest.
Anwar said he's talking with all the other major computer manufacturers and energy companies, and would be more than happy to sign some up as clients. He'd also look at other airlines if they weren't in such precarious financial health. But he's also looking at completely different industries that have a need to find a "needle in a haystack," such as pharmaceuticals, insurance, and financial companies.
In the near future, Anwar said he wants to look for applications for PolyVista's technology in pharmaceutical research. In that industry, the "needle" would be a molecule or a compound that shows promise as a drug, and the "haystack" would be the millions of compounds that have little or no use as drugs.
In the insurance industry, he said he also sees possible applications, where the PolyVista system could be used to search for fraudulent claims. And what PolyVista now does for BP to monitor energy trading could be adapted into a more general purpose financial trading watchdog system. It also might find uses in the credit card industry, where it could look for unusual purchasing patterns that suggest credit card fraud. He's already expanding PolyVista's sales presence in New York and other financial and insurance centers in the Northeast.
Airline Industry Bankruptcies
Anwar said he's more or less written off further expansion within the airline industry, because so many of the carriers besides Southwest are in such dire financial straits. Pricing, even for an entry-level system, starts at $100,000 or so, and that's just too rich for most of the major airlines at this point in time. But the system PolyVista installed for Southwest is likely to take on additional applications outside of pilot safety reports. Who knows what interesting patterns and anomalies it will turn up if it's pointed at corporate or customer data?
Meanwhile, to go deeper into high-tech manufacturing, Anwar said, he'd probably have to think about translating PolyVista's software into some of the east Asian languages, so it will appeal to the actual manufacturers of the circuit boards and other components going into consumer electronics and computers. So much manufacturing has moved offshore that it's becoming increasingly difficult to sell manufacturing software to the American-based importers, assemblers, and sales organizations.
PolyVista is already expanding abroad, but more for software development than for sales. When Warranty Week spoke with Anwar for this article, he had just returned from a business trip where he visited the company's development centers in Bulgaria and elsewhere in Eastern Europe. Just as many American manufacturers are moving production offshore, so too are American software developers.
"I'm looking to expand in Eastern Europe," Anwar said. "They seem to have a lot more talent. I'm surprised by the amount of educated people in Eastern Europe, for example, in Bulgaria. It's really interesting to see the number of college-degreed people they have. It has to be one of the highest ratios in the world."