Blog

A few months back I had the opportunity to help a client set up and run a few of our algorithms.  We had already created a number of work maps to perform many types of analyses on the structured data in their dataset.  The client wanted to create some new workmaps designed to enhance analysis of their freeform text.  After a few hours of setup, I had completed an entire set of workmaps and was ready to show them.

During a session with the client, I began showing what I found.  There were many typical types of analysis ranging from Summaries and Pareto charts of word clusters to trend line analysis of those clusters.  All of the results were very helpful to the client for reporting the current state of their business, but there was one algorithm in particular that stood out as a unique way to look at their data. The Discovery algorithm.

The Discovery algorithm is used to find outliers in data trends.  This method, when combined with text clusters, can provide some interesting and unexpected results.  In this case it delivered a result that didn’t seem to make much sense.   It kept returning a text cluster containing the word “Plates” combined with a specific product.  No analysis to this point had ever identified this combination of words from their freeform text and in this case it was still a very limited number of results.  The client informed me the problem with this result was that the product in question does not contain “Plates.”   After seeing the unusual result we decided to do some investigating to find out what might be causing this relationship in the clusters.  What we found was that that particular product was being received by their customers without name plates, a 50 cent part that most engineers don’t even think about.  These name plates have all of the information about a unit that a technician needs to install the product correctly. Instructions are also provided with the units, but they are typically discarded after the installation is completed.  The name plate is needed for any future maintenance or re-installations.  Because the name plates were missing, three units were later damaged during re-installation and returned for credit under warranty.

Weeks later the client and I reconnected to go over some new data and I asked if they had ever resolved the issue Discovery had found.  The answer was a very simple one.  The name plates for these units were shipped with the unit taped to the inside of the box.  More often than not the name plate was thrown away along with the packaging when the unit was assembled.  The fix was easy.  The name plates are now taped to the unit itself with some very brightly colored tape so the technician can’t miss them.  Customers who had received units in the past were also contacted and confirmed whether or not they had name plates on their units.  All in all, the cost to fix the issue was less than $20 but the three units that had to be scrapped cost more than $20,000.  Three units may not have seemed like a lot compared to the thousands of returns that are credited in a year, but the Discovery algorithm was able to point out an issue that was easy to fix and potentially saved a lot more money by reducing future returns.