When was the last time warranty claim issues made it to the top of your operational to-do list? Perhaps it happened after the marketing department complained about a poor review in a consumer magazine. Or worse, when the legal department called to let you know about a lawsuit.
For too many companies, warranty claims are not a high priority unless undertaken to avoid something -- angry customers, alienated dealers or a public relations debacle. This holds true despite the fact that warranty claims eat up 2% of U.S. company revenue.
Two percent is likely a conservative figure. Think of the financial, legal and image agony tire manufacturers went through several years ago, then consider:
- How much could your company save if it could mine warranty detail quickly enough to isolate defects and stop newly defective merchandise from leaving the factory?
- How much could your company save if it decreased or eliminated the 10% of warranty claims that are fraudulent?
- Does the 2% of revenue figure include the PR and legal costs for companies that are slow to respond to a serious warranty problem?
Reducing revenue loss starts with analyzing warranty claims. But software designed to do that isn't created alike. Most companies are swimming in data -- much of it ends up being useless.
Text Mining Is The First Step To Solving The Problem
For several years text mining has been touted as a solution to warranty issues, particularly the issue of quickly identifying problem areas and fixing them. Instead of technicians trying to select from hundreds of warranty categories or clerks trying to guess which of hundreds of warranty categories a technician's written assessment belongs in, text mining software is supposed to read written assessments and devise a list of the top warranty areas.
There have been two problems with this approach. First, most programs don't look for changes over time. Often, the software simply spits out a list of the Top 20 warranty claims. What happens, though, when a type of freezer defect jumps from a ranking of 200 one week to 25 the next? It's still off the Top 10 list and more importantly its sudden and dramatic jump in occurrence goes unnoticed.
The second problem has been the software's tendency to throw out too many red flags. A simple reporting system answers this question: Which plant/model combination has the highest failure score? The more useful question would be; which plant/model combination is significantly worse?
View article on one page