Text as Data

CHAPTER 6Classification and Prediction

USE CASE SCENARIO

Here we consider repairs under warranty for a global truck manufacturer. In this situation, the manufacturer is trying to get a more accurate warranty repair report, as better accuracy drives better manufacturing processes in the future. The current situation relies on relatively short, cryptic texts left by mechanics. Manual spot checks have demonstrated that the text that is entered carries an inaccurate warranty code about 50 percent of the time. The main goal is to associate mechanics notes with actual repairs undertaken. This can lead to better ways of assigning warranty codes that more accurately reflect actual repairs. We developed a text analytics-based prediction and deployment application for the truck manufacturer. This work was documented in two successive SAS User Conferences presentations.ⁱ

Currently, the warranty claim process at this vehicle manufacturer assigns a warranty type/cause (WTC) code to warranty claims (this is an eight-digit code). This system, carried out by human operators, had been in production for 15 years and has been audited and verified to produce an approximately 50 percent accuracy rate to claim type.

This system and the claim type/cause assignment also form the basis for the quality targets that are set for the vehicle manufacturing division.

The main goal of the analysis was to conduct a specific proof of concept and financial impact analysis to determine whether there is enough ...