Trace Id is missing
July 04, 2024

Cdiscount: Azure OpenAI Service and GitHub Copilot join forces for e-commerce

Household appliances, cell phones, second-hand goods, ticketing services... With no less than 70 million products, 9 million customers, and some 14,000 sellers, the Cdiscount marketplace has been a key player in French e-commerce for over 20 years.
 

The company's success is due to its ability to reinvent itself in line with market trends and technological opportunities. When generative AI made its spectacular entry into the digital world, Cdiscount indeed immediately understood its power and decided to infuse it fully at every level of the company. Isabelle Serot, Data Manager at Cdiscount, and Nicolas Thibaut, Director of Software Factory and Devex at Peaksys, Cdiscount's technology subsidiary, share the first use cases.

Cdiscount

Riding the wave of disruptive technologies

"Generative AI represents a real technological breakthrough that we need to seize," asserts Nicolas Thibaut, pointing out that within Cdiscount, the potential of this technology is unanimously supported, first and foremost, by management. In fact, the French e-commerce leader has already begun its deployment across all business lines: “We want all our employees to have access to generative AI, and to be able to use it to develop business and professional ideas.”

This high ambition begins with the provision of GitHub Copilot to all Peaksys Techs, i.e., 500 people. For the group, the aim is to test the solution's capabilities through a "limitless" deployment. "This demonstrates our keen interest in generative AI," emphasizes Nicolas Thibaut.

Making GitHub Copilot an ally for developers

A fortnight later, the tool has already been adopted by around 100 developers, who use it to generate algorithms, frameworks, and tests in JavaScript, React, Python, and .NET Core, so that they can concentrate solely on high value-added tasks. "Our developers do the same code as before, but more efficiently," says Nicolas Thibaut. “Thanks to GitHub Copilot, they save an average of half a working day a week.” More concretely, the pull request rate—enabling a developer to inform his team members that the newly developed functionality is available for revision and merge in the main branch—is up 10-15% for senior profiles and 20% for junior profiles.

Soon, the application should also become part of the Data Science ecosystem: "Some high stakes around data require Python skills, and the population of Machine Learning Engineers who accompany Data Scientists is benefiting more and more from GitHub Copilot. This enables them to gain efficiency and allows Data Scientists to concentrate on their core business: identifying the best models to respond to business issues," says Nicolas Thibaut.

Team commitment, the cornerstone of innovation

While these initial experiments have been conclusive, the adoption of generative AI cannot be achieved without the full involvement of the teams: "We have carried out a major effort to raise awareness of the tool, increasing the number of users from 100 to 290," Nicolas Thibaut says, pleased. Prompt engineering training courses have also been set up, for people with technical profiles wishing to become more autonomous in the field of generative AI. “Our employees are keen to learn and perfect their skills, which is why we invest in training.”

This dynamic of skills development and acculturation is also expressed by hosting a community dedicated to GitHub Copilot. The aim is to encourage feedback, the sharing of best practices, and collective emulation.

Creating a new-generation customer experience

With its 17 million unique visitors per month and 70 million product sheets, Cdiscount has an unrivaled playground for testing Large Language Models (LLMs) on its business processes. On the customer experience side, the company is seizing the potential of generative AI and combining its existing, tried-and-tested in-house models with Azure OpenAI Service to boost the performance of its online references. "If a product sheet is not correctly written, does not contain the right pictures or the right properties, then it will not be categorized in the right section and will lose effectiveness," explains Isabelle Serot, Data Manager. To avoid this pitfall, the e-commerce giant set about recategorizing these sheets, optimizing the power of its algorithms by combining its own AI models with the LLMs available in Azure. As a result, after ten months, 23 million of the 66 million files already submitted to Azure OpenAI Service have been recategorized, and now boast a 30% conversion rate. "With Azure, the LLMs are trained in Europe, and Microsoft already enjoyed the trust of our CISO, DPO, and legal teams," explains Isabelle Serot.

At the same time, the technical teams have developed a pre-sales advice chatbot based on product sheets. This chatbot is also based on Azure OpenAI Service and deployed by partner iAdvize. Since its deployment, this chatbot has achieved a 70% customer satisfaction rate, the same as that generally observed for an exchange with a human.

This promising progress encourages the group to continue its exploratory work: "We plan to continue improving our various algorithms, while exploiting the opportunities that exist in conversational technologies," concludes Isabelle Serot.

“With Azure, the LLMs are trained in Europe, and Microsoft already enjoyed the trust of our CISO, DPO, and legal teams.”

Isabelle Serot, Data Manager, Cdiscount

Take the next step

Fuel innovation with Microsoft

Talk to an expert about custom solutions

Let us help you create customized solutions and achieve your unique business goals.

Drive results with proven solutions

Achieve more with the products and solutions that helped our customers reach their goals.

Follow Microsoft