Generative artificial intelligence is reshaping daily operations at L’Oréal, where more than 75% of the company’s 95,000 employees have completed training programs since 2023, Les Echos reported.
The French cosmetics giant, founded more than 115 years ago and operating in 150 countries, generated approximately 44 billion euros in revenue in 2025, marking a 1.3% increase.
The 95,000-employee enterprise now relies on custom internal tools to accelerate marketing campaigns, draft emails, and handle brainstorming sessions. Weekly active users of L’Oréal-GPT, the group’s proprietary assistant, have reached 40,000 employees.
Meanwhile, marketing teams use an in-house platform called CreAItech to produce images and videos that match the visual identity of the group’s 40 distinct brands.
“We have thus the same levels of requirement in terms of supervision and security in access management,” Marcelo Goes, chief information, data, and AI officer for the group, told Les Echos.
Underpinning these deployment efforts is the Beauty Tech Agentic Platform, known as BTAP. The agentic framework combines Google Cloud infrastructure for global operations with Alibaba Cloud for operations in China. BTAP operates as a multi-LLM system utilizing models from Anthropic, Google, Mistral, and OpenAI, alongside Qwen and DeepSeek models reserved exclusively for Chinese personnel, according to Les Echos.
BTAP works alongside the Beauty Tech Data Platform, which has run on Google Cloud infrastructure since 2022. Goes described the repository as the world’s largest beauty database, housing over 25 petabytes of information. The system ingests real-time data covering every stage of a cosmetic product’s lifecycle, from laboratory formulation and packaging to supply chain logistics and end-user performance.
To govern cloud spending and environmental impact, L’Oréal integrates FinOps financial controls and GreenOps environmental monitoring tools directly into its AI workflows. L’Oréal-GPT exemplifies this approach by routing tasks to language models based on their energy consumption and cost.
Employees select a specific task from an interface menu, such as writing an email or conducting a brainstorming session. The FinOps allocation system then assigns a suitable language model for that exact task while displaying the associated energy consumption to the user.