en

Malakoff Humanis: Improving Customer Relations With the Power of NLP

To address their growing challenges in keeping up with customer demands and providing quality customer service, Malakoff Humanis turned to Dataiku’s Deep Belief program and collaborated with Dataiku’s data scientists on two advanced natural language processing (NLP) projects.
 

Malakoff Humanis has a dedicated data science and analytics department led by a Chief Data Officer. The data department is comprised of four main branches, each in charge of:

  • Data Science and Analytics
  • Data Governance 
  • Data Architecture and Cloud
  • AI and Data Visualization

To address their growing challenges in keeping up with customer demands and providing quality customer service, Malakoff Humanis turned to Dataiku’s Deep Belief program, which provides consulting services to tackle ambitious AI projects. Through this program, Malakoff Humanis collaborated with Dataiku’s data scientists on two advanced natural language processing (NLP) projects.

Natural Language Processing for Classifying Customer Claims

Initially, Malakoff Humanis started working with Dataiku on an AI-based solution that helps understand the topic of online claims through NLP classification algorithms and automatically dispatch the claim to the appropriate customer service team. 

The developed model served as a foundation for building and implementing another solution for improving telephone customer assistance through NLP, which today is fully operationalized and widely used across the customer service department. The initial project helped prove the benefits of using a centralized Enterprise AI and data science platform for end-to-end AI, and more specifically NLP-related projects, as well as the value of the reuse and capitalization on data projects. 

Speech Analytics and Sentiment Analysis for Improved Telephone Customer Service

The purpose of the second AI project that Malakoff Humanis developed using Dataiku’s Deep Belief program was to analyze the content of customer calls (themes and tone) in order to identify areas for improvement of telephone assistance. The main goals of the project were:

  • Improved management of telephone assistance thanks to a deeper understanding of the callers’ motivations, pain points, and satisfaction levels
  • Shorter calls and fewer re-calls
  • Less pressure on customer support teams
  • Improved customer satisfaction
Watch Video

The solution is composed of two main modules which answer two main questions:

  1. Topic classification: What are the calls about? The goal is to find out why there is a surplus of calls on certain topics, in order to have more precise staffing forecasts.
  2. Sentiment analysis: What is the level of satisfaction of calls? The objective is to build a model that allows to have new information on the global tone of the calls and to know on which topics and problems customers tend to be most dissatisfied in order to react promptly. Furthermore, this would allow to assess the levels of customer satisfaction across different teams, and compare the effectiveness of internal versus outsourced customer support teams. 

Even though the object of classification, or the input data, in this second project was different than the first one (telephone call voice recordings as opposed to written online claims), the similarities in terms of topic categories and the NLP classification techniques used allowed for the reuse and repurposing of the classification algorithm built for the first project. This allowed for a significant reduction in the time required to put the model into production. 

Thanks to Dataiku, we were able to take an already existing model and seamlessly repurpose and reuse it on a new type of data for a new use case. Gauthier Lalande Lead Manager AI, Malakoff Humanis

Malakoff Humanis NLP project made in Dataiku DSS

The sentiment analysis NLP model built to assess the tone of telephone calls generates predictions for the overall tone, the tone of separate sentences in the conversation, and the sentiment at the beginning and the end of the conversation (20% of the first and last words). In the absence of labeled transcripts for the tone, the predictions were verified empirically.

Finally, a dynamic dashboard was built to present the results of the predictions in real-time and inform decision-making across the data and customer assistance teams.

The project allowed us to deploy a new method of analyzing phone calls thanks to a mixture of AI and business rules that intelligently complement each other. This provided the opportunity for the customer relations team to understand the contributions of AI and for data scientists to incorporate the business area expertise into the model. Gauthier Lalande Lead Manager AI, Malakoff Humanis

The Results

  1. The project made it possible to create a processing chain which takes transcriptions of telephone calls as input and analyses and classifies the tone as well as the topic of conversations. 
  2. Despite the relatively low number of labeled transcriptions, the information obtained allows to analyze and monitor the overall content and sentiment of calls, and classify them accurately into the main call categories.
  3. Dataiku’s Deep Belief program allowed to identify and operationalize a new advanced NLP use case for the Malakoff Humanis AI team in a secure and scalable way that empowers users to be autonomous, continue monitoring the models in production, and potentially reuse it for other text classification problems. The Deep Belief approach facilitated knowledge transfer and allowed for a smooth project handover between Dataiku and Malakoff Humanis.
  4. The guidance and direction provided by Deep Belief helped to go beyond a purely “algorithmic” approach and focus the project efforts not only on the technical aspects, but also the concrete business objectives, by actively collaborating with the business side and providing them with actionable insights in the form of a dynamic dashboard. Dataiku’s collaboration features were considered a major advantage and were appreciated both by executives and users. 

Aviva: Powering Insurance With Data, Machine Learning, & AI

Aviva has been using Dataiku for many years to build a range of data analytics and machine learning use cases, ultimately forwarding their customer-first mission.

Read more

Santéclair: Detecting Fraudulent Claims More Effectively

Santéclair uses Dataiku to enable fraud detection teams to target actual fraud cases 3x more effectively, saving money for both the company and its customers.

Learn More

Aviva: Bringing Insurance Into the Age of AI

Aviva’s Customer Data Science Team is 5x more efficient in developing data projects from beginning (building a model) to end (pushing into production) with Dataiku.

Learn More

Frende Forsikring: Using NLP to Automate Claims Reporting

An initial BERT model trained on 10,000 emails in Dataiku is now being used to distribute all emails in the claim center.

Learn More

Heraeus: Improving the Sales Lead Pipeline With LLMs

Each of Heraeus’s 20 operating companies has its own lead identification and qualification process. See how Heraeus uses Large Language Models (LLMs) in Dataiku to support these processes, ultimately saving time and increasing sales conversion.

Learn More