Improving Margins for Used Machinery with Machine Learning

Challenges

A consulting agency specializing in agriculture machinery wanted intelligent, ML-driven price predictions offered as a suite across inventory and trade-in offerings.
The existing rule-based valuation system did not utilize data — missing detail and important market dynamics. Data was highly dimensional with many optional add-ons and missing values to infer. Some information would also be unknown from physical inspection at inference time.
To build trust, it was essential to explain price predictions in a simple way for dealers and buyers. Offered as cloud SaaS, the solution required strong scalability and generalizability.

Engineered dataset from multiple sources — extensive exploration, cleaning and mapping.
Tested multiple model families for best generalizable performance, resulting in multiple models in production.
Feature-engineered important predictors with subject-matter experts to align with real-world use.
Architected and developed inference API to serve predictions.
Developed “explainability” and “confidence” metrics, stripped of statistics and tailored to buyers.
Built front-end dashboards and web app on top of the prediction backend, integrating supplementary analytics.
Deployed to Azure, isolated to dedicated VNet, architected for SaaS scalability.
Implemented monitoring for production performance, model and data distribution drift.

Two packages focused on getting the optimal price across the business — increased margins, lower depreciation risk.
Inventory valuation predicting the optimal sales price on SKU-level and which period to obtain the best price, accounting for seasonality.
Trade-in solution predicting a fair trade-in value in physical dealerships.
Modularized and scalable as a cloud SaaS-solution, utilizing state-of-the-art machine learning methods.

Technologies & methods

Predictive Modelling Machine Learning XGBoost K-Nearest Neighbours Regression Python SQL Azure API Streamlit