Analytics by Proekspert predicted customer re-claims with 80% accuracy.

Predictive insights establish future performance and measure potential customer reclaims, avoiding critical losses and increased business profitability.

Customer Claim Prediction

Customer reclaims stemming from shipping out items is a significant factor for all manufacturers. Therefore, minimizing the probability of customer reclaims is of major importance.

Based on test and reclaim data, we created a machine learning model that was able to predict future customer reclaim with approximately 80% accuracy.

Analysis

The analysis addresses the following questions:

  • Measurements, diagnostics, and repair of signal processing and radio frequency modules are currently time consuming and thus also expensive. Is it possible to cut costs by building a decision aid for diagnostic technicians that will shorten repair times?
  • What are the patterns in diagnostic measurement results that provide predictive insight into potential faults and future claim returns from customers?

The methods used for the extensive analysis were the following:

1) Exploratory Data Analysis
Tools Used:

  • Microsoft R Server for .csv files, data.table
  • SpectX for log/blob parsing

2) Modelling
Tools Used:

  • R with H2O, xgboost, randomForest

3) Methods:

  • Parallelized Random Forest
  • Feedforward Deep Neural Networks
  • Extreme Gradient Tree Boosted Classifier
  • Intuition

Our predictive models were suspiciously powerful with nearly 80% precision. A reverse causality case was uncovered: received claim cases undergo different measurement procedures. By removing the measurements that were done after a claim date, the picture became more realistic.

There are no obvious predictors of claims, rather hundreds of weak ones that accumulate into one strong predictor.

Results

Based on the existing data and metadata we were able to:

  • Build a classifier that, according to measurement results, predicts the future customer claim on the unit.
  • Provide some business recommendations for improving the process.
    Some units go through a proportionally high number of test cycles and still experience customer return in later stages. Pending on BoM cost, “how much testing and repair is enough?”
  • Optimal sequencing of measurements could cut down the time required for measurements.

Want to hear more?

Get in Touch.

Peeter
Meos

Data Science Lead and Partner
peeter.meos@proekspert.com
Phone: +372 5663 0316

Getting started with Proekspert

Free *

  • Assessment of organization 
data maturity
  • General recommendations of
    improvement potential

* Limited number of free consultations available.

From 5000 €

  • Assessment of organization data maturity
  • General recommendations of
    improvement potential
  • Roadmap development

From 15000 €

  • Assessment of organization
 data maturity
  • General recommendations of
    improvement potential
  • Roadmap development
  • Prototype development for one identified business question
  • Potential to operationalize the prototype as a repeatable solution
  • Clear definition of implementation plan

Go smarter with Proekspert.

Please fill in the contact form below and we'll get back to you as soon as possible.

Thank You!

Your message has been sent. Our team will get back to you as soon as possible.

Close this window