CNP Assurances tackles the very time-consuming treatment of professional emails

Automation of low value-added tasks is a very topical issue for businesses, regardless of business or industry. Among the recurring tasks that can benefit from innovations related to automation, professional email processing is important. Sorting, tracking and responding can be extremely time-consuming, even though quality is essential in a professional's relationship with his client. This is the reason why the start-up Recital was born in 2017. One of its solutions, named Quieto, currently in production at CNP Assurances, proposes both to automate recurring tasks (classification and monitoring of emails, process triggering, document retrieval, responses) and eliminate errors (checking outgoing emails, preventing data leaks). "We identified Recital in the spring of 2018 and put in production the first tests in July on the individual insurance part, explains Vladimir Micetic, Head of Production Angers and Customer Relations. This was quick because we knew our needs and we were already in project mode ". This is set up in three months, with the integration of 200 emails to train the machine sorting in 13 major categories defined. A response library is created.

Concretely, the software is based on the Automatic Language Processing (TAL), with the objective of developing machines able to understand human language, here in writing. Unlike other actors who focus on the semantic analysis of words, the solution, driven on massive data and models that reproduce human language word by word, focuses on morphological analysis and context to guide the mail to a category and generate a response. "This allows us to understand an unknown word, which happens when a term exists only in the jargon of a company," says Gilles Moyse, co-founder and CEO of Recital.

Analysis of feelings (urgency, satisfaction), attachments, assignment of the email to one or more category (s), anonymization of the text, extraction of information. He can then formulate an answer, the object, or add a signature. Machine learning makes it possible to constantly improve the elements of automation and personalization.