Skip to content
Data Science & AI

RAPP AG — Automating Customer Request Processing with AI-powered Email Classification

In a nutshell

Our AI-based email classification system intelligently categorizes customer requests and automatically directs them to the respective business department, thus eliminating manual sorting, significantly reducing processing time and improving customer satisfaction as well as employees’ efficiency.

Rapp AG
Business Challenge

A high volume of incoming customer inquiries leading to a substantial workload due to manual sorting of customer emails, resulting in delayed responses and impacting customer satisfaction.

Project overview


09/2023 – 12/2023

Team setup

1 Lead Data Scientist, 1 Data Scientist (Parkside), Business Experts (Rapp AG)


Python, HuggingFace, Azure, GitHub


AI & Data Science

About Rapp AG

Rapp AG is a Swiss engineering and construction company with a strong focus on innovation and sustainability. Leading in the field of integral design, the company provides a wide range of services (e.g., infrastructure, architecture, energy, mobility, and logistics solutions) while focusing on their customers’ needs and developing sustainable concepts for the forward-looking design of living spaces.

Challenges & objectives

A common challenge in developing AI-based email classification algorithms is the need for a large amount of labeled data. The large distribution of different emails in particular can present a model with tricky challenges. In close cooperation with Rapp, suitable classes were first defined and a corresponding number of emails collected. A significant problem in this process was posed by different class sizes, as it was not possible to collect the same number of emails per class. To overcome this challenge, we needed to use transfer learning, a technique in machine learning (ML) that reuses a pre-trained model on a new problem, thus enabling deep neural networks to be trained with comparatively little data.

The overall objective was to automate the email categorization and forwarding process, reducing the employees’ manual workload and enhancing their overall operational efficiency.

Our solution

To make up for the limited data we had, we employed transfer learning and utilized a pre-trained transformer-based natural language model from HuggingFace. We then fine-tuned this model on Rapp AG’s customer emails, allowing it to effectively learn the patterns and characteristics of customer requests. Our model, trained on just 30 labeled examples per department, achieved an impressive accuracy of over 98%, ensuring that customer requests and inquiries get accurately classified and forwarded to the appropriate department.

End results & future plans

Our AI-powered email classification algorithm proved to be an excellent solution for Rapp AG’s problem of automatically routing different types of emails to the right department without having to filter them manually. The use of such a solution shortens response times and ensures that customer inquiries are processed promptly and efficiently. It also allows employees to focus on higher value tasks, increasing their overall productivity and efficiency.

In the future, the accuracy of the model can be further improved by providing more edge cases of different classes. We have also developed a concept for deploying our solution via Azure, which enables seamless integration into Rapp AG’s existing infrastructure. This end-to-end approach facilitates troubleshooting and maintenance in the continuous use of this machine learning application.