L’Oréal aimed to reinvent beauty advice with a personalized AI experience—available 24/7, secure, and scalable to millions of users.
With Azure OpenAI Service, L’Oréal launched Beauty Genius, combining generative AI, cloud performance, and privacy for a seamless, personalized beauty experience.
The result: a high-performing, ethical AI assistant available 24/7, delivering a smooth experience with latency under 5 seconds, at global scale
As a leading force in the cosmetics industry, L’Oréal Groupe has been creating beauty that moves the world for over a century. Thanks to the power of Beauty Tech, L'Oréal Groupe offers personalized, inclusive, and responsible beauty experiences for each. Always striving to develop the best products, the Group is also shaping the future of beauty by leveraging the latest advances in cosmetics… and new technologies.
From this ambitious approach comes “Beauty Genius” by L'Oréal Paris a personal beauty assistant powered by generative AI, available 24/7 in consumers' pockets, offering a seamless and personalized customer experience, along with technical prowess. Moutia Khatiri, Chief Technology Officer of Online & Omnisales, together with William Lenoir, Beauty Genius Solution Architect at L’Oréal Groupe, give us an exclusive look behind the scenes.
How can we fully exploit the opportunities of the virtual world while ensuring customers receive the same level of expertise and support as they would from a real advisor in a beauty store? This strategic challenge for all industry players is the starting point for Beauty Genius, driven by L’Oréal Groupe’s Beauty Tech—the department dedicated to innovation—and based on Microsoft Azure OpenAI Service technologies.
The idea? To develop a Generative AI-powered assistant that directly interfaces with customers, capable of interacting simultaneously with each user. To achieve this, the infrastructure hosting Beauty Genius’ GenAI components must meet several technical requirements: ensure stability and consistent latency—both essential in cases of high traffic—be available and reliable, with the platform relying on a fully managed service and benefiting from the elasticity and high availability of the Cloud. Not to mention the security and confidentiality of customer data, which are essential criteria for the Group. “Being the first company to release a GenAI-based beauty assistant was a challenge, but also an exciting journey alongside the Microsoft teams. We worked together to continuously optimize the accuracy and the performance of the tool to make the most out of Azure OpenAI service,” explains Khatiri.
“Being the first company to release a GenAI-based beauty assistant was a challenge, but also an exciting journey alongside the Microsoft teams. We worked together to continuously optimize the accuracy and the performance of the tool to make the most out of Azure Open AI service.”
Moutia Khatiri, Chief Technology Officer of Online & Omnisales, L'Oréal Groupe
“We didn’t want to sacrifice the performance, neither waste provisioned, non-used resources, so this hybrid architecture came out as the best compromise in alignment with our eco-design principles," Khatiri continues. Faced with these requirements, the Azure OpenAI Service environment was chosen by the L’Oréal Groupe’s tech teams. Interactions generated by the conversational agent remain strictly private, in a dedicated L’Oréal environment, precisely meeting the Group’s commitments to data protection and ethics.
Equipped with powerful LLMs (Large Language Models) such as GPT-4o, which is particularly suited for advanced chat completions, and Ada—which enables the implementation of a RAG (Retrieval Augmented Generation) system specific to L’Oréal Groupe—Azure OpenAI Service offers two deployment modes: a classic pay-as-you-go model and a Provisioned Throughput Units (PTUs) mode.
While the pay-as-you-go model was initially chosen to launch Beauty Genius, the increase in traffic due to the application’s success quickly led the teams to consider PTUs, in order to benefit from more predictable performance, with guarantees on maximum latency and throughput.
“While we have mastered web performance where milliseconds matter, Generative AI presented a new latency frontier. The challenge was to balance the speed users demand with the high-quality answers they deserve. Azure OpenAI Service PTU provided a crucial element to solve this equation, ensuring a seamless and performant experience,” mentions Lenoir.
More specifically, the group devised a hybrid architecture, combining a base of fixed PTUs and automatic switching to the pay-as-you-go model, when the PTU limits are reached. The goal? To ensure that 100% of requests receive a response, while guaranteeing optimal performance for a given volume of requests per minute via PTUs. This approach was also more efficient in terms of cost management and resource allocation, avoiding the need to size the infrastructure based on a theoretical maximum traffic peak.
Next, a series of load tests was conducted to determine the robustness of the architecture. Initially, the system was deployed in a temporary environment dedicated to performance testing, configured with a minimum of 15 PTUs. A simulation of 100 simultaneous users was carried out. The load gradually increased up to 2,000 simultaneous users, and the number of PTUs was adjusted as needed to support this scaling.
Switching from a pay-as-you-go model to a hybrid architecture played a key role in significantly improving performance. PTUs made it possible to keep maximum latency below 5 seconds, while offering greater stability in response time variance. Since PTUs are also independent of the used model type, they could be reassigned to different models as needed. This flexibility allowed L’Oréal Beauty Tech teams to gradually migrate certain workloads and further optimize performance results. The result: Beauty Genius can truly scale up without compromising user experience.
By relying on Azure OpenAI Service, L’Oréal Groupe demonstrates its ability to integrate the most advanced technologies into its Beauty Tech ecosystem, with a strategy resolutely focused on performance, ethics, and scalability.
Beauty Genius is much more than an AI assistant. It is a foundational building block in the digital transformation of beauty consulting, capable of reproducing, on a global scale, the personalized experience of an expert.
Follow Microsoft