​Consumers file between 30,000 and 40,000 complaints with the French government's consumer protection agency, DGCCRF, each year. Processing all of those complaints is complex and time consuming: Staff have to read each complaint carefully and take into account the latest changes in any applicable laws or regulations before responding. The agency turned to the CEA for a solution to streamline this task.

CEA-List image and language processing experts tackled the challenge, coming up with a tool capable of selecting boilerplate paragraphs from a repository. A learning-based statistical classification algorithm breaks down incoming data—whether it's in an email, attachment, or form—into "bags of words." The complaints can then be matched with boilerplate paragraphs associated with the type of complaint. The tool also assigns a confidence score to the paragraphs selected. The staff processing the complaints simply select the paragraphs they would like to use in their replies; they no longer need to retype the entire letter. The tool also has a rules manager so that users can manually add new rules based on news about a particular company or changes in laws or regulations in just minutes.

It won second prize in the jury's favorite category of the BERCY INNOV tech innovation competition held by France's Ministry of the Economy. This time-saving tool helps ensure greater consistency and service quality.