Warranty Master Class:
The day before the sixth annual Warranty Chain Management Conference opens next month in Los Angeles, several warranty experts will deliver three-hour workshops aimed at those who already have some familiarity with warranty analytics.
Usually, the day before a conference officially opens is one that gives traveling executives a chance to get some rest after landing and checking in. Usually, there are also a few tutorials and "Introduction to" courses aimed at those making their first visit to the show, and maybe a reception with some food and drinks.
However, at the WCM Conference this year, most of the pre-conference workshops seem less like introductory overviews aimed at beginners and more like master classes in warranty analysis. If these were college courses, they'd probably be offered only to juniors and seniors who've already declared warranty to be their major.
At the same time, there are no prerequisites, and no need for attendees to first become intimately familiar with the works of Waloddi Weibull or Vilfredo Pareto. In other words, experience is preferable but is not required.
Wrote the Book on Warranty
To begin the morning program on March 2, Wallace R. Blischke, a consultant in statistical analysis and professor emeritus in the Marshall School of Business at the University of Southern California, will present a tutorial on the "Collection and Analysis of Warranty Data."
"But this is not a pre-conference, get you up to speed thing," Blischke warned. "That's for sure."
Instead, it's a sort of master class aimed at the warranty professional who wants to increase their existing level of knowledge. In other words, if you work in the warranty department or are regularly asked to analyze warranty data, this course is for you.
Blischke, along with collaborators such as University of Queensland professor D.N. Prabhakar Murthy, literally wrote the book on warranty data analysis. In fact, they've written or co-written six books on the subject, and about to release their seventh. Among the titles still in print are Warranty Cost Analysis (ISBN 0824789113), Product Warranty Handbook (ISBN 0824789555), and Warranty Management and Product Manufacture (ISBN 1852339330).
Blischke said he expects the latest title in the series, Warranty Data Collection and Analysis, to be published by Springer later this year.
"We've been working on it for a couple of years," he said, with he, Murthy and Rezaul Karim from the University of Rajshahi in Bangladesh listed as co-authors. A fourth co-author, Kazuyuki Suzuki from Tokyo's University of Electro-Communications, had to drop out for personal reasons, Blischke added.
None of those books are required reading before one can attend this workshop, however. In addition, Blischke said the material won't be too heavy on the mathematics, and also won't be too detailed about the way the warranty data is analyzed. "But we talk about the methodologies, and we talk about the data structures. So just about anybody who's looked at warranty data can relate to that," he said.
Some people analyze warranty data to predict cost, he said. Others look at warranty data to predict reliability. Blischke said he hopes there's enough content in the workshop to make it relevant to both camps.
"I would say this is aimed at the practitioner: someone who is analyzing data," Blischke suggested. "It may or may not be warranty data, but [it's someone who] at least has experience in analyzing data."
Cleaning Dirty Data
Warranty is at the intersection of statistics, probability, and psychology, Blischke said. There are literally tons of data available for analysis, but the data is rarely "clean" enough to allow immediate data mining. And then there is the task of predicting the future, using data from the past. "So it gets a bit beyond just statistical analysis," he said.
Blischke, now largely retired except for his book publishing activities, said he got into the warranty business rather unintentionally almost 40 years ago. After doing a bit of consulting and teaching in the statistical realm, he was asked to take on an aerospace statistics project for a Boeing supplier.
"I was approached by a company that was producing 747 windshields, among many other things," he said. Boeing wanted the supplier to pick up the cost of all warranty claims by itself. So the supplier wanted to know how much to set aside to fund future warranty claims, and how much to raise their prices to Boeing.
Blischke said he started simple: with performance and failure data. Then he began to realize how complex it would be to develop a model of the future cost of failures and replacements. His first paper on the subject was published in 1973, and that led to another project, and another, and another.
He met Murthy a few years later, after Murthy had turned his personal pique over a denied warranty claim for his transmission into the basis for an equally unplanned warranty-related career in Australia. Soon they were co-authoring warranty-related books, and recently they celebrated a quarter-century of collaboration.
Service Contract Workshop
The other workshop scheduled for the morning of March 2 is entitled, "Designing a Successful Service Contract Program."
It will be jointly presented by Terry Hawkins, the senior director of Assurant Solutions' Warranty Management Group, and Michael Yates, Assurant Solutions' vice president of risk management and underwriting.
Hawkins said he's ready to deliver a presentation either filled with details, or aimed at the beginner, depending on who shows up.
"I've done a few of these workshops before," Hawkins said, "and I like doing them, because it provides an opportunity to interact with the audience and to direct the content towards them. I like to look at who's coming, and then to shape my presentation to anticipate what I think they're going to want to hear."
So far, he said, most of the signups seem to be coming from the marketing side rather than the legal or insurance side. "I'm going to talk a little about the regulatory side -- but not much," Hawkins said, because there are several presenters, including Michael Bardell, senior counsel for Assurant Solutions, scheduled for the main conference that will probe that topic in detail.
Bursting at the Seams
Still, he wants to be prepared. Hawkins said Michael Yates, his co-presenter, asked him why the appendix in the handouts was so big. "So I said, 'Well, you never know. Someone might ask a question, and I'll want to have the backups to be able to respond.'"
Hawkins said he also plans to talk about the design of marketing materials, and to compare and contrast the different tax treatments available for the self-insured versus a fully-insured program. Yates will delve into the intricacies of actuarial analysis, and how the statistics can be misleading. But they're ready to expand or collapse that list of topics based on who shows up.
Hawkins said he's been in the service contract business for more than 25 years. Back in the mid-1980s, he was an attorney in private practice who had a client that needed help with the state compliance issues surrounding service contracts. Later on, General Electric became a client, and that evolved into a management position with GE.
"I've seen all the things you can do wrong. And I've seen all the things you can do right," he said. And he's eager to share some of those stories at the WCM workshop. At last year's WCM conference, for instance, he spoke about what a company can do to "right a ship" that's gone "under water" because claims costs are out of control.
Overall, the business isn't as simple as it was 25 years ago, Hawkins said. Over time, the major players have learned to use their own data to analyze and predict the performance of their service contract programs.
"One of the things that Mike and I are going to do is to look at some of the intricacies of the business," Hawkins said. That should help people who need to drill down and understand why a given program isn't performing as well as it could. "What you see at first glance might not be the cause."
During the afternoon session on March 2, attendees will have a choice between two Warranty Chain Management workshops. "Fraud Detection and Quality Enhancement Via Text Mining and Data Mining," is a joint presentation of Sergei Ananyan, CEO of Megaputer Intelligence Inc., and Vadim Kozyrkov, president of Aculocity, which is built around a case study of a text mining and data mining project that fixed mistakes and reduced labor coding errors for a Midwestern truck manufacturer.
The second choice of the afternoon is "From Stumbling Block to Stepping Stone �- Accelerated Warranty Transformation Utilizing Warranty Value Engineering," a joint presentation by consultants Victoria Christensen and Stefan Georgi from PCubed, a project management company based in London and New York. It is built around findings first summarized three years ago at the WCM Conference in Tampa, and published last year in a research bulletin entitled "Developing Warranty Excellence."
Last year, in the February 5, 2009 newsletter, Warranty Week tried to calculate the probability that Kozyrkov and Ananyan would deliver back-to-back workshops on new statistical approaches to warranty data. This year, they're in the same workshop.
Graham Muller, a project manager with Aculocity, said Ananyan will talk about the use of text analytics to fill in some of the blanks and correct some of the errors in structured data that could otherwise mislead an analyst. Then Kozyrkov will talk about how to use structured data, such as the warranty cost per unit data, to create an accurate view of how a product line's reliability is changing.
Fixing Faulty Data First
"For example, the analysis of labor codes would not be sufficient in warranty situations," Ananyan suggested, because they're frequently incorrect. Sometimes the repair shops get a bit careless about which code they choose from a long list. Other times, they're actively looking for a bigger payment, so they choose a "better" code that allows more labor time. Either way, they're providing information which could be misleading to the manufacturer.
"Frequently, the analysis of structured data alone is not sufficient," Ananyan said. "Companies need to understand the mixture of structured data and text associated with both warranty claims and repair notes. The benefits that companies can derive from this analysis relate not only to warranties, but are intimately tied to product quality."
Ananyan said Megaputer and Aculocity jointly studied the warranty data coming from a Midwestern truck manufacturer's dealers, and found that the clerks that were inputting the data and choosing the codes were introducing errors into the process. Whether they were being fraudulent or just careless, they were misinterpreting the notes written by the mechanics, he said.
"If there would be means to correct this picture, manufacturers could do a much better job on the analysis and fixing the problem," he said. The solution Aculocity selected involved bringing Megaputer in to examine the mechanics' notes and to detect any divergences between them and the codes selected by the data entry clerks.
Speaking Their Language
"We have to go to the actual notes," Ananyan said. But, he added, "they are extremely cryptic," filled with acronyms, misspellings and abbreviations that only a truck dealer would know and love.
Megaputer's Term Replacement Engine
CHRGING SYS NOT SHOWING CHRG ON GUAGE
CKD BATT, CONN OK.
CKD STRTR AND ALT OK.
REMVD CLUSTER AND FND VOLT GAUGE INOP,
REPLC GAUGE OK
CHARGING SYSTEM NOT SHOWING CHARGE ON GAUGE
CHECKED BATTERY, CONNECTION OK.
CHECKED STARTER AND ALTERNATOR OK.
REMOVED CLUSTER AND FOUND VOLTAGE GAUGE INOPERATIVE,
REPLACED GAUGE OK
Source: Megaputer Intelligence
Once all those abbreviations have been fixed, Megaputer looked for anomalies between the text and the actual repair codes chosen. To accomplish that, Megaputer runs a "basket analysis" on the notes, which looks at hundreds of thousands of claims and picks out those that diverge significantly from the norm.
"It's affinity grouping," Ananyan said. "Then we would run the analysis with other machine learning algorithms, to check what are normal and abnormal repair patterns. So the system learns what's normal, and then finds deviations, based on undistorted data."
Fraud or Flub?
Could it be fraud? Or could it have been a careless mistake? "In the area of fraud, nothing is black and white," Ananyan said. "It's all different shades of gray. It might not be fraud. It might be a case where the repair shop has very poor technician coding, or they actually don't care. They're just putting down the closest code they can find."
However, Ananyan said, he's also noticed that once the clerks and technicians find out that someone is watching them and what they submit, their behavior changes remarkably fast. "So it's more a prevention rather than going after them and demanding the money," he said.
Next, Megaputer is going to point its technology towards a prominent computer manufacturer's call center data, and then at a Midwestern medical device manufacturer's warranty data. And there's a high probability that we'll be hearing about those case studies at WCM 2011.