Scaling Used Car Pricing Intelligence with Vertex AI
EdgeRed diagnosed and refined a used car price prediction model for a leading Australian automotive marketplace – replacing inconsistent, experience-dependent pricing with a fast, reliable model that any team member can use.
The Challenge
A leading Australian automotive marketplace was relying on the judgement of seasoned buyers to price second-hand vehicles accurately. The process was inconsistent, hard to scale, and created a knowledge dependency that left the business exposed whenever experienced staff weren’t available. They had a pricing model in place – but it was underperforming for high-volume brands, and nobody was sure why.
Technology Used
- Snowflake — data processing and model training infrastructure
- Google Vertex AI — cloud-native ML platform for model training, refinement, and scalable inference
- Tableau — pricing dashboards surfacing model outputs and insights for the buying team
- Streamlit — interactive application for real-time pricing exploration and decision support
What We Built
EdgeRed investigated the model’s behaviour from the data up – focusing on what the training data was doing before touching the model itself.
-
Data quality & diagnostic analysis
Identified mapping discrepancies between two source databases that were introducing pricing inaccuracies - fixing the data before adjusting the model. -
Feature engineering & resampling
Model year identified as the most influential pricing feature. Resampling and feature engineering applied to improve accuracy across high-volume brands where the model had previously struggled. -
Snowflake & Vertex AI
Model training and data processing run on Snowflake and Vertex AI - giving the business a scalable, cloud-native foundation for ongoing model refinement as new vehicle data flows in. -
Pricing dashboards
Interactive dashboards built to surface model outputs and pricing insights in real time - giving the buying team a clear, consistent basis for pricing decisions without needing to interpret raw model outputs.
The Outcome
Pricing is now consistent, fast and accessible to the whole team – not just experienced buyers. The model processes large volumes of vehicle transaction data across hundreds of makes and models, delivering reliable price predictions at scale. New staff can price confidently from day one, and the buying team has a clear, data-driven basis for every decision – no experience required, no inconsistency, no guesswork.
