RE-GrievanceAssist: Enhancing Customer Experience through ML-Powered Complaint Management

Overview

• This paper introduces an end-to-end pipeline called RE-GrievanceAssist for managing customer complaints in the real estate industry.
• The pipeline consists of three key components: a response/no-response ML model, a user type classifier, and an issue/sub-issue classifier.
• The system has been deployed as a batch job in Databricks, resulting in a 40% reduction in manual effort and a monthly cost savings of Rs 1,50,000 since August 2023.

Plain English Explanation

In recent years, digital platform companies have faced increasing challenges in managing customer complaints as more and more consumers adopt these platforms. To address this issue, the researchers have developed an end-to-end pipeline for real estate customer complaint management.

The pipeline has three main parts:

1. A machine learning model that can determine whether a customer complaint requires a response or not, using a technique called TF-IDF vectorization and an XGBoost classifier.
2. A classifier that can identify the type of user (e.g., tenant, landlord) who is making the complaint, using a FastText classifier.
3. A classifier that can determine the issue or sub-issue the complaint is about, again using TF-IDF vectorization and an XGBoost classifier.

By automating these key tasks, the pipeline has been able to reduce the overall manual effort required to manage customer complaints by 40%, saving the company Rs 1,50,000 per month since it was deployed in August 2023 on the Databricks platform.

Technical Explanation

The researchers developed the RE-GrievanceAssist pipeline to address the growing challenges that digital platform companies face in managing customer complaints. The pipeline consists of three main components:

1. Response/No-Response ML Model: This model uses TF-IDF vectorization to represent the text of the complaint and an XGBoost classifier to determine whether the complaint requires a response from the platform.
2. User Type Classifier: This component employs a FastText classifier to identify the type of user (e.g., tenant, landlord) who is making the complaint.
3. Issue/Sub-Issue Classifier: Similar to the response/no-response model, this classifier uses TF-IDF vectorization and an XGBoost classifier to determine the issue or sub-issue that the complaint is about.

The researchers deployed the RE-GrievanceAssist pipeline as a batch job in the Databricks platform, which resulted in a remarkable 40% reduction in the overall manual effort required to manage customer complaints, as well as a monthly cost savings of Rs 1,50,000 since August 2023.

Critical Analysis

The researchers have presented a comprehensive solution to the growing challenge of managing customer complaints for digital platform companies, particularly in the real estate industry. The use of machine learning models to automate key tasks, such as determining the need for a response and categorizing the type of complaint, is a promising approach that can significantly reduce the manual effort required.

However, the paper does not provide much detail on the specific datasets or training procedures used for the machine learning models. It would be helpful to understand the performance and generalizability of these models to different types of complaints or platforms. Additionally, the paper does not discuss any potential biases or limitations of the classifiers, which could be important considerations for real-world deployment.

Furthermore, the researchers could have explored the integration of the RE-GrievanceAssist pipeline with other customer service systems or language models to provide a more holistic solution for managing customer complaints. Leveraging knowledge graphs or comparative studies with other approaches could also have provided additional insights and perspectives.

Conclusion

The RE-GrievanceAssist pipeline introduced in this paper represents a significant step forward in automating the management of customer complaints for digital platform companies, particularly in the real estate industry. By leveraging machine learning models to streamline key tasks, the pipeline has demonstrated the potential to reduce manual effort and costs while improving the overall customer experience.

As digital platforms continue to grow and evolve, solutions like RE-GrievanceAssist will likely become increasingly important for managing the increasing volume and complexity of customer complaints. While the paper provides a promising foundation, further research and refinement could help to address any limitations and unlock even greater benefits for both platforms and their customers.