How Carrefour leveraged AI to drive design delivery

Since 2022, the French multinational Carrefour has been moving rapidly toward becoming a digital retail company.

This ambition is centered on four strategic pillars: an omnichannel, agent-driven customer experience; the digitization of stores; the modernization of headquarters and support functions; all of which are driven by data and AI.

Since then, the digital team has doubled its staff of designers and CX experts, distributed across several Digital Factories organized by business lines working with the 4-in-the-box methodology, which combines several complementary areas of expertise: business, product, design, tech, and data.

Structuring AI usage to enable collective acceleration

As in many companies, practices emerge spontaneously: product managers are beginning to adopt vibe coding tools to speed up product and update rollouts. However, these initiatives did not account for the entire production chain or design thinking, so they quickly revealed their limitations and brought governance questions to light, leading to a first in-house generative AI tool: Carrefour AI.

In May 2025, the decision was made to collectively step up efforts to standardize AI usage practices within the Design & CX department. An internal survey reveals that nearly 50% of its members use Carrefour AI.

The transformation is already underway; the next step is to organize it. Carrefour AI is moving towards adopting Google’s Gemini solution, and the team plans to spend the next six months structuring the use of AI and taking it to the next level.

Setting the stage

Fostering an AI-driven culture and standardizing AI usage

Survey responses reveal that Design and CX teams are on the lookout for ways to boost their productivity, especially in the discovery phase and in their day-to-day work. The company is implementing a transformation plan to support this shift, dedicating a quarter to AI onboarding. This includes training on prompting and tools like Figma Make and Gemini, as well as naming ambassadors to promote AI adoption within the department.

At the same time, the department strives for consistency in practices to avoid a fragmented adoption. A library of tested and validated prompts is made available to designers, while a dedicated channel allows for sharing concrete use cases. The most relevant initiatives are then presented during office hours, and the entire system is supported by regular workshops where practices can be tested, adjusted, and refined based on feedback.

Carrefour has also created a reference document where teams can submit their ideas for Gemini agents (GEMS), including their names, functions, expected outcomes, and a link to the GEM. The GEM is then tested, validated, and officially rolled out to the teams. Designers can help improve it by suggesting optimizations. There are now more than 50 GEMS, ranging from presentation assistance and project scoping to content generation and UX research reporting.

Getting the Design System ready for AI

To make its Design System Marcel work with generative tools, Carrefour simplified and restructured it. The rise of AI has highlighted a common limitation of Design Systems: when they’re too complex or not well-documented, generative models produce inconsistent interfaces, or ones which are impossible to integrate.

To address this, the team directly connected the AI to its Design System. The first step was to extract the context (components, tokens, icons, typography, glossary) via the Figma MCP Console, to obtain queryable context in the form of JSON files feeding 16 tools and 5 system prompts called in real time by the LLM. In practice, users simply need to mention @Marcel in the Figma Make chat to start working with the assistant. Finally, the goal is to plug the entire team into Figma Make by creating a connector. Thus, the MCP acts as the guardian of the Design System and a comprehensive assistant able to challenge design decisions, suggest data visualization components, and refine UX writing.

Thanks to this process, every prototype generated with Figma Make directly draws on the Design System’s components and variables, ensuring that the generated interfaces automatically adhere to the colours, spacing and typography defined by the organisation, and thus to its experience and accessibility standards. Furthermore, this method allows designers to focus on the UX, rather than using AI credits to adjust the UI.

Making AI a standard tool for discovery

At Carrefour, the most significant impact of AI for the design and product teams is felt during the discovery phase, where designers play a central role in shaping the user experience strategy. The teams use tools such as Gemini and NotebookLM to analyse user verbatim transcripts, summarise research, reframe problems or structure workflows. In this way, raw data (interviews, workshop notes, reports) is transformed into actionable insights for design. Discovery thus becomes faster, more structured and more systematic, whilst remaining driven by human expertise.

Prototyping and testing optimisation

AI has also transformed the way interfaces are prototyped and tested. The design team now uses Figma Make as a rapid exploration tool to generate and compare different patterns (such as card, timeline or bento layouts).

Prototypes can directly incorporate interactions and animations, enabling the creation of demos that are much closer to the final product, boosting productivity by up to 30% during the back-and-forth between design and development, and speeding up validation phases. Prototyping is becoming a space for experimentation that is faster, more autonomous and better connected to user testing and real-world production constraints.

AI also opens up new possibilities for designers, who can now create their own internal tools, such as token explorers or icon managers, without relying on the technical team.

Moving towards an AI-native approach

This shift also strengthens collaboration between designers and product managers, who can now participate more actively in the exploration phase using generative tools, whilst designers provide expertise in context, experience consistency and the foundations of the Design System. Prototyping becomes a space for co-creation where ideas can be tested more quickly, whilst remaining guided by design expertise.

Alongside this increasingly widespread automation, human validation remains a core principle. While generative tools accelerate the production and structuring of ideas, the final decision still rests with the product and design teams. AI enhances design capabilities without replacing expertise. This balance between automation and human oversight is what enables Carrefour to deploy AI at scale while maintaining the quality and consistency of its product experience.

So what’s next? The teams are preparing to integrate insights from Discovery and the product knowledge base to further refine our prototypes. They are also continuing to develop their tool use cases and AI agents to facilitate project scoping and improve the customer experience.

Carrefour’s Roadmap: How to get started with AI

• Quickly define use cases

Establish from the outset where and how AI should be used to ensure product consistency and safe use for the company. Identify areas of the technical stack where AI can be easily integrated and implemented.

• Train and onboard teams

Invest in skill development (prompts, tools) through dedicated training and internal champions to accelerate adoption.

• Standardize and share best practices

Establish a common language (prompt library, shared methods, discussion forums) to build a collaborative system.

• Ensure interface consistency

Connect your Design System to AI so that your prototypes comply with your standards.

• Target the most relevant use cases and measure impact

Start by using AI for discovery, then expand its use to prototypes. Track concrete metrics to demonstrate benefits and establish long-term adoption.