Taking the Uncertainty of Cash Flow Fluctuations Away Reducing Wild Fluctuations in Cash Flow Due to Customer Claims

For any product, warranty claims can give major hiccups to the financial controller of the company. These claims are hard to predict and thus it is not possible to allocate budget from quarter to quarter. Often, such products are out in the market for less than a year. With a limited amount of data, it becomes even more difficult to use many standard data analytics techniques.

In one such case, we received data for machinery that was out for just over a year. The company had recorded multiple parameters. However, the significance of those to any of the field failures was not known. In addition, these parameter names were cryptic and thus finding their engineering relevance was not possible. The customer did not have enough time to explain all the details. They just gave the training dataset. This set spanned over two quarters and thus the dataset size was limited. The customer was interested in knowing the top five failures, as per the severity, out of about 20 possible ones.

We spent most of our time in finding the most important parameters through data analysis. Further, we developed mathematical equations based on the trends. As always, we did not use the nth order polynomial fit. We believe that often such models do not help in accurate predictions. Our equations were unique and exhibited behavior similar to the data. The mean squared error was also within the acceptable bounds.

With this model we predicted failures for the subsequent two quarters. Our top three predictions matched the test data.

Our expertise lies in getting deep insights from the limited data. We developed a methodology that gives reasonably good predictions, thereby bringing some control to almost random fluctuations in the required cash flow.