Sales Forecasting

Challenge

Previously, our client conducted business development forecasting with significant manual effort. By leveraging machine learning, the client aimed to enhance forecasting accuracy and establish an automated process to improve the planning capabilities of sales as a decision support tool. The goal was to predict rated production—comprising new business, canceled contracts, and inventory—in the insurance sector. The primary challenge was to consider forecasts across different product and distribution hierarchies to enable precise sales management. The model needed to be integrated into the client's IT infrastructure.

Approach

To support sales, we developed a forecasting engine that uses time series models to predict the rated production at the end of a month or year. The model was trained using a pooled approach, meaning multiple time series were used in parallel for training, allowing the model to better learn the interactions between the time series. Since this was the client's first data science use case, we chose a platform-independent, container-based approach for implementing the model.

Result

Based on the forecasts, we helped our client proactively shape their sales activities. The container-based integration of the model into the IT systems and the automation of training implemented the first use case into the client's infrastructure. Our client can now apply this implementation method to additional data science use cases. The automation of training significantly reduces the manual effort compared to the previous method.