L'Oréal advances conversational analytics with Claude

L'Oréal is the world's largest cosmetics and beauty company, operating in over 150 countries with a portfolio of 37+ international brands across skincare, haircare, makeup, and fragrance. The company is positioning itself as a Beauty Tech leader, combining cutting-edge research and innovation with AI, data, and digital capabilities.

With Claude, L'Oréal has:

• Achieved 99.9% accuracy on conversational analytics applications, compared to 90% with previous GenAI approaches
• Built a multi-agent system that transforms natural language questions into data insights and visualizations
• Enabled employees to query data directly rather than building custom dashboards for each question
• Deployed Claude across its internal AI platform, serving 15,000 daily users and 44,000 monthly users

The problem

After years of developing and deploying traditional AI applications, L'Oréal began its generative AI journey in 2023. The company needed to democratize AI capabilities across its global workforce while maintaining security, governance, and business value. A key challenge was enabling employees to access and analyze data without requiring technical expertise or custom dashboard development for each inquiry.

L'Oréal required sophisticated reasoning capabilities for complex analytical and financial tasks involving math, coding, and SQL generation—particularly for conversational analytics where employees could ask questions in natural language and receive accurate insights. The company also sought to boost productivity by eliminating repetitive tasks and allowing employees to focus on higher-value work, while accelerating time-to-market, gaining market share, and speeding formulation research.

Claude orchestrates multiple specialized agents for complex workflows

L'Oréal selected Claude through rigorous testing across their use cases.. "Our auto evaluation capabilities, such as LLM as a judge, have proven many times the superiority of Claude models for a various number of use cases, particularly for the most complex ones," said Thomas Menard, Head of Agentic Platform and LAB at L'Oréal.

For conversational analytics specifically, L'Oréal conducted extensive hands-on testing to validate Claude's performance. Claude's advanced reasoning capabilities proved essential for the company's most demanding applications, especially in conversational analytics where accuracy and reliability are critical for building user trust.

L'Oréal deploys Claude as an intelligent orchestrator

L'Oréal uses Claude in two primary ways: within AI companions that require high-quality reasoning capabilities, and across their software development lifecycle through tools like GitHub Copilot and Claude Code.

For conversational analytics, Claude serves as the main orchestrator of multiple specialized agents that work together to transform user questions into insights and visualizations. When an employee asks a question in natural language, Claude coordinates with semantic API agents, data retrieval systems, and specialized agents for calculations, product master data, and geography master data.

This orchestration approach allows Claude to determine which specialized agents to engage based on the user's question, coordinate their work, and synthesize results into clear answers with supporting data visualizations. Rather than relying on a single model for all tasks, L'Oréal built specialized agents for specific functions. For calculations, Claude routes to a dedicated calculator agent. For product dimensions, Claude delegates to a product master data agent. Geography queries are handled by a geography master data agent.

The architecture helps mitigate potential accuracy issues. "Risk of hallucination has been tracked by the team and mitigated," Menard explained. By routing specific types of queries to purpose-built agents, the system maintains high accuracy across diverse analytical tasks while Claude manages the overall workflow and synthesizes results. The system queries L'Oréal's Beauty Tech Data Platform in natural language while managing user identity and access controls.

Testing has validated the architecture's effectiveness. "Through our tests, it's been proven that Claude 3.7, then 4 and now 4.5 Sonnet are by far the most capable and advanced models when it comes to drive sub-agents to fulfill the user needs," said Menard.

The outcome

The impact on accuracy has been substantial. L'Oréal now reaches 99% accuracy scores on conversational analytics applications, compared to 90% with previous GenAI approaches. This improvement is building trust and confidence among employees to adopt these new capabilities at scale.

The shift in how employees can work with data is equally significant. Rather than building custom dashboards for each ad-hoc question, employees can now simply ask their questions in natural language. This represents a fundamental change in how L'Oréal's workforce can access and analyze information.

L'Oréal's internal AI platform has achieved broad adoption, with 15,000 daily unique users, 33,000 weekly unique users, and 44,000 monthly unique users generating 2.5 million messages per month across multiple models. Early testing of Claude Code with 30 to 50 developers is also showing promising results for engineering productivity.

Looking ahead, L'Oréal sees agentic AI as still in early stages within the company. "The rise of Agentic is starting to be built and we count on Anthropic models to put us at the forefront of Beauty Tech," said Menard. The company plans to expand its agentic capabilities and deepen its partnership with Anthropic to maintain its position as a Beauty Tech leader in the competitive cosmetics industry.