CASE STUDY

Machine Learning in Manufacturing

At a glance

Challenge

Help a heavy equipment manufacturer predict warranty claims that are supplier-at-fault to recover millions in lost claims.

Result

Pariveda developed a cloud-based solution using a Machine Learning model to predict which claims would be accepted by suppliers and to mark these claims for submission.

Impact

Pariveda’s model predicted that over $13M worth of claims that were not submitted to suppliers would have been accepted, with a potential recovery amount of over $7M.

Technologies used

Amazon SageMaker, Databricks Apache Spark™

One of the US’s largest corporations and one of the largest heavy equipment manufacturers in the world.

Heavy equipment manufacturers routinely deal with the assembly timeline and various suppliers while trying to keep customers satisfied. But, when there is a fault in the product, how do they collect from suppliers when their parts cause a warranty claim?

Leveraging machine learning to recover millions from the supplier-at-fault claims process for a leading heavy equipment manufacturer.

When our client gets a warranty claim, often a part supplier is at fault, and some contracts allow them to recover money from the supplier. Today, this manufacturer recovers money on only 3% of claims. It is estimated that 12-13% of claims could be recoverable, amounting to a reduction in warranty liability in the $10M range.

How Pariveda developed a cloud-based solution using a Machine Learning model:

The model was trained using the set of claims that had been submitted to suppliers based on the existing process using a fully automated platform to enable retraining, ensuring that the model will keep up with current trends.
The model’s execution is also fully automated, running daily over the claims that have been submitted to ensure the claims would be accepted by suppliers and marked for submission.
The Databricks Apache Spark™ platform was used for data engineering, with model training and execution built on Amazon SageMaker using the platform’s built in XGBoost algorithm.