Nilfisk

Early warning system for back-orders

A gradient-boosting model predicts back-order risk across plants and materials — with ~86% accuracy — surfacing supply-chain problems before they hit.

Machine Learning & AIManufacturing & Industrials

Key result~86% accuracy predicting back-order risk

Challenge

As the first use case on its new data lake, Nilfisk wanted an early warning system combining many common sources to indicate the likelihood — and business impact — of back-orders in production.

Solution

A gradient boosting machine (GBM) chosen over random forests, neural nets and SVMs for superior performance, embedded in a live dashboard.

Algorithm. GBM classifier predicting back-order probability per plant/material combination.
Classification model. ~86% accuracy, tuned so genuinely at-risk combinations receive clearly elevated probabilities.
Validation. 8-fold, 4-times-repeated cross-validation, surfaced in a dashboard with stock, open orders and real-time status.

Results

Data-driven early warnings on supply-chain problems
Root-cause analysis & interaction overview
Prioritisation of high-impact focus areas

Topics

Machine Learning
Data Engineering

Challenge

Solution

Results

Topics

How we work, the Flowtale Way.