Generative AI and E-commerce: The First Feedback from Vertbaudet

Vertbaudet is one of the pioneering French e-commerce players in the use of generative AI. Indeed, the online leader in the children’s market has initiated a series of proof of concepts (POCs) focused on e-commerce performance, in collaboration with Google and Converteo. Insights from Solenne Pasqualetti, the Marketing & E-commerce Director of the brand, and Charles Letaillieur, Senior Manager and lead for the generative AI offering at Converteo.

Why is a brand like Vertbaudet interested in generative AI?

Solenne Pasqualetti, Vertbaudet: While Vertbaudet is a company that was founded over 60 years ago—established in 1963—and historically associated with mail order and paper catalogs, it has also become a digital leader in the children’s market over time. This digital transformation began as early as 1996 with the launch of its first website (in the UK at the time) and continues to this day, including the development of its Second Hand site in 2022 and the transformation of Vertbaudet.fr into a marketplace in 2023.

Vertbaudet has fully embraced the digital and e-commerce shift, as this channel now represents 80% of our revenue, with 5 million orders processed annually. In this context, exploring the possibilities offered by generative AI to improve our e-commerce performance seemed like a natural step.

Charles Letaillieur, Converteo: More broadly, we have seen that 2023 marked the first year of generative AI, during which brands began to explore the topic to evaluate opportunities, as well as to understand the risks and limitations, particularly in terms of data privacy. In 2024, the ecosystem began to structure itself. There are increasingly more use cases scaling up. This is the case at Vertbaudet, which we are supporting in its reflections as part of our partnership with Google. We have worked together on two axes (the productivity and creativity of teams), with a very pragmatic vision: prioritize, prototype, and move quickly into production.

What are the initial use cases of generative AI at Vertbaudet?

Charles Letaillieur: One of the first challenges was to prioritize the use cases. After an ideation phase, we identified two themes where generative AI could be particularly relevant for Vertbaudet: improving e-commerce performance and optimizing creativity and productivity in styling, since Vertbaudet designs its own product ranges in-house. Given that e-commerce is the most mature topic, we directed our initial POCs (Proof of Concepts) in this direction.

Solenne Pasqualetti: Yes, initially we focused on a topic related to e-commerce performance: generating product presentation sheets for the e-commerce pages of the site, with a prototype assistant for content creation. We will return to the use cases in styling once our employees have been trained, which is currently underway, as we aim to have trained over 80% of our headquarters staff in generative AI by September.

In just eight weeks, we developed three tools: one to generate product sheets, another to enhance them using image recognition, and a third for translation into seven different languages for nine countries, with tones of voice adapted to each market. The models were trained on 10,000 examples of Vertbaudet product sheets, enabling us to produce a web title, an HTML product sheet, and a receipt title from a product description.

What is your assessment of these initial generative AI tools?

Solenne Pasqualetti: By automating the process of creating product sheets for e-commerce, our goal was to optimize the productivity of the e-commerce teams so they could focus their time on higher-value tasks for the company. This automation will help better manage peak activity periods, during which the e-commerce teams receive a significant number of new products to integrate into the site.

The initial generations of product sheets have been very successful, with instances where it was difficult to discern whether the content was generated by AI or humans when compared blindly. Regarding translations, all product sheets have proven correct, but some local formulations may need adjustments, and specific elements must be integrated, such as units of measurement, currencies, and mandatory legal mentions. Given these satisfactory initial results, we have moved into production for generating product sheets and translations, and we continue to refine the image recognition tool.

Charles Letaillieur: In terms of lessons learned, this project has clearly demonstrated the importance of having well-structured data in advance to obtain quality content. It is important to emphasize that, in all cases, there is a human validation of the generated content: the AI-generated product sheets are reviewed and completed by copywriters. Human input remains essential to ensure the final quality. The aim of the tool is to save time for the teams, allowing them to focus on optimization and adding value. Its deployment is part of a broader innovation approach within Vertbaudet, and it is accompanied by a change management framework.