ML-based residual value prediction of vehicles

The project at a glance

• Prediction of residual values through machine learning: In our solution, a machine learning (ML) model is trained based on training data. This ML model is used to predict the residual values of vehicles in the future and to determine the book value of the current leasing portfolio of a leading car manufacturer.
• Auditability by auditors: As the solution provides figures that are relevant to the financial statements, it is audited by auditors. Various components were implemented for this purpose, including audit logs, approval workflows and comprehensive doc-as-code-based architecture, process and test documentation.
• Multi-market: The application was rolled out for nine markets worldwide. Market-specific requirements were taken into account both in the ML model and in data aggregation.
• Cloud-native microservice architecture: The system behind our solution has a modular structure. Individual, independent services can be flexibly scaled, updated and maintained without affecting the entire application. Kubernetes and AWS are used as a scalable platform.

Initial situation

The car manufacturer was faced with the challenge of developing a standardized solution for predicting the residual values of its vehicles. Previous approaches were market-specific and highly fragmented, leading to inefficiencies and limited accuracy of predictions. In some markets, residual values were purchased externally, resulting in additional costs. The objective was to introduce global software that could still be customized to each market’s needs and meet the auditing requirements.

Solution

Together with the car manufacturer, we developed a scalable platform for predicting vehicle residual values and evaluating the leasing portfolio. Using machine learning models, our solution delivers precise forecasts, takes market-specific requirements into account and automates end-to-end processes from data aggregation to approval.

The system processes data from a variety of market-specific interfaces, prepares it and integrates it seamlessly into the platform. The results of the forecasts and valuations are transferred directly to the contract management system.

A particular highlight is the complex front end, which offers a user-friendly and powerful interface. It has been specially tailored to the needs of users who are used to working with tables and Excel. By supporting table-like input, the platform enables intuitive, flexible and efficient data capture without changing users' usual workflows. The platform is used by various user groups, including the market CFO, and therefore fulfils the highest requirements in terms of user-friendliness and transparency.

To further optimize and use the data, a BI interface was integrated, which makes the data available in real time for reporting and dashboards. In addition, our solution was integrated into the car manufacturer's internal API management system, ensuring standardized and secure API management and allowing the platform to be seamlessly integrated into existing internal systems.

The technical basis includes Spring Boot for microservices, Spring Batch for ETL data processing, Angular for the front end, Celery for machine learning workloads as well as Kubernetes and AWS for flexible scaling. GitHub Actions enable automated tests and the provision of services in the DevOps approach.

With audit logs, approval workflows and automated tests, the application fulfils the requirements of auditing. The platform improves operational efficiency, enables informed decisions through accurate ML forecasting and provides a scalable, future-proof solution for forecasting and managing residual value data.

Outcome

The development of this solution resulted in a standardized and globally scalable platform that now covers more than 50 percent of the car manufacturer's global leasing portfolio.

By providing benchmark residual values and comprehensive risk assessments, the platform offers a reliable basis for well-founded decisions worldwide.

End-to-end processes, precise forecasts and the seamless integration of market-specific requirements ensure efficiency and flexibility. At the same time, standardized processes, audit logs and comprehensive test documentation ensure a high level of auditability and transparency.