AI: operational excellence support for customer relations

It was the chatbot project that was finally selected. Its objective: to automate certain responses and point to online documentation of IRCEM sites.

The project was split into two parts: an artificial intelligence part and a more technical part, linked to the chatbot's GUI and its integration into the site. For each of these tasks, a team was formed comprising an IRCEM developer and an IBM Cloud Expert Labs developer. The team was headed by a project manager. This co-development organization was particularly well suited to the company: "Because our business is based on handling confidential data, IRCEM has a strong culture of internalization and control of our IS," confirms Ludovic Decarpigny, "so we wanted to acquire all the expertise on these developments so that we could continue to manage and grow these projects ourselves".

To enrich the knowledge base, the team began by exploiting a corpus of emails already received. The theme of social action (support for customers in difficult situations such as serious illness, disability, over-indebtedness), was chosen as a priority because it represents a large number of queries and has a lot of online documentation. IBM Cloud Expert Labs trained the IRCEM teams in the use of watsonx Assistant. "The interface provided by IBM Cloud makes it easy to get to grips with watsonx Assistant, configure it, test it and check that it's working properly. We can quickly identify how it interprets users' messages, correct it if necessary and improve understanding through machine learning. All this makes the use of AI very accessible, which is a real strength, as it allows us to concentrate on our application rather than on configuring services," confirms Alexis Flouw, Software and Development Engineer.

During the acceptance phase, business and customer relations experts were brought in to test the chatbot. Then, once the AI's understanding was deemed sufficient, the decision tree was moved into production for the public launch. "IBM's cloud-based feedback tools enabled us to track the chatbot's adoption right away, and identify any comprehension issues. We trained people to correct the learning process, and this was done over time," explains Ludovic Decarpigny. Once the first topic had been mastered, IRCEM extended the chatbot's knowledge to other areas, such as retirement and provident insurance. During this period of development, and thanks to the successful transfer of knowledge, IRCEM worked autonomously, calling on IBM only very occasionally.

The chatbot now accounts for almost 4,000 conversations a month, covering around a hundred different purposes. Specific updates have been implemented, such as Covid-related questions. Despite this success, the IRCEM teams realized that the number of emails was continuing to rise... The chatbot does indeed have its limits: some customers don't want to talk to what they consider to be a robot. Others prefer to write emails in order to be more precise.

A new Design Thinking session has enabled us to imagine another use for AI within an assisted messaging system. The principle: when the customer enters the content of his email in a dedicated window, the AI analyzes his texts to provide him, in real time, with general or personal information taken from his IRCEM data. In the end, if the customer does not obtain the satisfactory information at the time of entry, his message leaves to be processed in the traditional way by an advisor. "Since the launch of this service 9 months ago, we have observed that the assistant is activated 7 times out of 10. And following its intervention, between 5 and 10% of people finally decide that sending their message is no longer necessary," summarizes Ludovic Decarpigny.

Recruiting employees to manage customer relations is becoming increasingly complex. While AI can help reduce incoming flows of requests and therefore control workloads, it should also enable us to go further.

This is the medium-term ambition of IRCEM, which wants to "augment" the advisors thanks to AI, by making it easier for them to search for documents in the database and extract data.... This will free them from the most repetitive tasks, giving them more time for higher value-added tasks. For Ludovic Decarpigny, "it's a double opportunity to offer a more precise, rapid and efficient customer service, while offering our advisors a more attractive mission."

Groupe IRCEM is a non-profit, jointly-managed social protection group dedicated to the home and family employment sector. Founded in 1973, it protects its members from life's risks and supports the most vulnerable by placing its raison d'être: "to optimize the disability-free life expectancy of its members" at the heart of its 3 institutions: IRCEM Retraite, IRCEM Prévoyance and IRCEM Mutuelle. IRCEM's mission is to protect over 1.4 million childcare assistants and employees of individual employers, as well as 750,000 pensioners.

More information about ircem.eu