Warranty Repairs Reimagined through Machine Learning and Statistical Pattern Recognition (Part 1)

Abstract

The automotive industry has been funding warranty repair work for many decades. The most common vehicle warranty is 3 years or 36,000 miles [1]. Original equipment manufacturers (OEM) in North America have dealers record all the work completed and submit claims for the work that qualifies for warranty reimbursement [2]. The OEM reviews the request and pays dealers for the work performed. In addition to payments, the database is also used to complete quality analysis for the vehicles. Often the software being used by dealerships is old and not designed for quality analysis. Reviewing all the warranty work done can be an arduous task. OEMs can receive 100,000 or more claims each day. To speed up the analysis process the OEMs will divide the repair work into sections based on the segment of the vehicle requiring work. This categorization allows the OEMs to spread the work across many experts in the company. But what does the OEMs do when the problem cannot be located at the dealership? The dealer still requires payment for the time they spent trying to find the issue. This is often categorized as TNF (trouble not found). This type of work without a resolution can account for a sizable percentage of warranty costs! It can be as high as 38% depending on the manufacturer. It can also affect customer satisfaction and customer loyalty. In fact, one of the most common complaints submitted to NHTSA (National Highway Traffic Safety Administration) is “dealer unable to locate problem” [3]. To make matters worse, most customers end up returning to the dealer multiple times before the issue is, if ever, located, and fixed [4]. So how can we find a better way to analyze these warranty claims and improve the customer experience while decreasing the cost for the OEM? This problem can be improved through machine learning and statistical pattern analysis.