Smile launches Smile.AI, a conversational AI agent based on a RAG architecture. Discover how Smile is reinventing the digital experience.
Paris, May 21, 2026 – Smile, the European leader in open digital services, announces the launch of Smile.AI, a conversational AI agent integrated directly into its website smile.eu. Designed and developed in-house in less than two months, it embodies the group's conviction: to support its customers in adopting AI, Smile starts with itself.
“To advise our clients on their AI transformation, we must be the first to have experienced it ourselves. Smile.AI is not a technological demonstration: it is our operational reality, built with the same requirements that we apply to our clients.”
Marc Palazon, President of the Smile group.
Smile.AI goes far beyond a search engine: it detects the user's intent, maps the context (sector, urgency, solutions considered), and formulates personalized responses based on a RAG architecture fueled by group content and continuously enriched by past conversations. When the exchange is mature, the agent collects the contact's details and automatically forwards them to the sales teams with a conversation summary and a suggested approach.
Under the hood: the reality of an AI project in production
Far from theoretical discussions, Smile shares the key stages of this experiment conducted in less than two months by a small technical team:
- Reliability of responses : The agent uses the RAG (Retrieval-Augmented Generation) architecture deployed on Google Cloud Platform (Vertex AI), chosen for its robustness and its ability to address the group's multilingual challenges. It does not generate text from general knowledge, but relies on two complementary data sources: the site's dynamic content and a corpus of rigorously selected and validated internal documents.
- Data governance : meticulous work was required to define what information can be publicly exposed, ensure its freshness, and purge obsolete data.
- Security and monitoring : Leveraging the robustness of Google Cloud Platform, the Secure by Design architecture is based on three key principles: native security filtering of each request before it reaches the generative engine; a perimeter strictly locked to the Smile ecosystem; and an absolute prohibition on rendering unvalidated URLs. Since its launch, three types of attacks have been detected and neutralized: prompt injection attempts aimed at bypassing the agent's basic instructions, identity spoofing attempts seeking to make it reveal confidential information, and function hijacking aimed at making it generate out-of-scope computer code.
“ The quality of an AI agent is built first and foremost on data. We dedicated a significant part of the project to governance: sorting, validating, structuring. It is this often invisible work that guarantees reliable and secure responses. Smile.AI is living proof of this. ”
Frédéric Pastore, Senior Data & AI Architect, Smile.
Initial results and developments
Jean-Charles Bordes, Chief Operating Officer of the Smile Group and the driving force behind the Smile.AI project, identifies two major benefits: “ The chatbot changes two things in concrete terms. First, the visitor's experience with Smile from the very first interaction: they express their intention through an intelligent conversation, and the agent provides initial answers. Second, a structured summary, built from the questions about the need asked by the agent throughout the exchange, is sent to our experts to allow them to better prepare for the first appointment. This saves time for both our clients and our teams.”
Initial findings from the analysis of the conversations:
- Query volume: 5,500 questions asked with an average of 3 questions per conversation and an average duration of 8 minutes per session.
- Most frequently asked questions: technical expertise (42%), recruitment (25%), customer needs and projects (22%).
- User feedback: a very low bounce rate (18%) demonstrates strong user interest and the relevance of the information provided.
This first iteration is just one step. The architecture, currently based on Google Cloud for its rapid deployment, is already designed to evolve towards a more sovereign architecture.
This project is part of a broader internal SmilAI program, aiming to make AI an accelerator of the group's transformation. Automated analysis of calls for tenders, skills matching, contractual compliance, technical capitalization: the work is only just beginning.