La Centrale Revolutionizes Car Ads with Agentic AI

La Centrale bets on AI to transform car listings

La Centrale, well-known for its car listings, is looking to take a new step by integrating artificial intelligence into its processes. The goal is to automate more and more tasks related to the sale and purchase of vehicles. To better understand this transformation, we spoke with Thomas Berger, the company's Chief Technology Officer, who explained how these tools are already changing the game for professionals and could soon impact individuals.

Thomas Berger has been the CTO at La Centrale since 2022.

The role of AI at La Centrale

In a recent press release, La Centrale mentioned its "agentic automotive platform." Currently, the company's AI agents are transforming data into concrete actions. Historically, La Centrale provided numerous indicators to professionals through its back offices, such as market trends, price levels, vehicle turnover rates, and lead volumes. However, a car dealer is not a data analyst.

The idea behind the agentic approach is to move from dashboards to actionable recommendations: should a price be adjusted, should a vehicle be better highlighted, or should stock be readjusted? For now, the agent does not act alone on behalf of the professional. The next step will be to enable these agents to execute more actions for sellers once the right level of trust is established.

Data sources

Where exactly do your proprietary data come from?

Our data is divided into three main blocks:

• The first is the vehicle reference. We have built an extremely detailed database in-house, which includes engine types and trims, covering about 200,000 configurations in the French market. This database is enriched with external sources and calculated data.

• The second block concerns listing data, which refers to the offer: pricing positioning, vehicle turnover, options, and services offered.

• The third block is demand: this involves user behavior, both at a macro level to understand market trends and at a micro level to personalize the experience.

By cross-referencing supply and demand, we obtain a near real-time view of the market. This view allows us to feed our pricing, sourcing, and recommendation tools. We also cross-reference it with tests and reviews from Caradisiac to better assist individuals.

Use of AI models

Do you use your own AI models or those available on the market?

La Centrale does not aim to develop its own language models. The approach is to use the most suitable model for each use case. For targeted tasks, such as SEO optimization of certain pages, models like Nova Lite are used for their efficiency and low cost. For richer uses, such as the Pilot AI tool designed to advise professionals, more advanced models like Nova Pro or Claude Haiku are preferred for their performance and ability to manage context.

Pilot Video: a major innovation

The product generating the most interest is Pilot Video. What is it exactly?

Pilot Video addresses a simple problem: publishing a listing still takes between five and ten minutes, sometimes more, with highly variable quality. The goal is to reduce this time to 60 seconds, which is the time it takes to walk around a car with a smartphone.

From a video, our agents extract the best photos, identify the vehicle using the license plate and visual analysis, retrieve its characteristics from our reference database, generate a promotional text, and then propose a price consistent with the market. We can also produce formats suitable for different channels, including social media, with videos designed for their respective codes. The idea is to transform raw footage captured on a phone into a usable listing, with maximum automation.

Deployment and service control

When will this service be available? And will the professional retain control?

The deployment will occur in phases over the next 12 to 18 months. Some features already exist, such as generating the "seller's note" or pricing tools. The extraction of photos from a video is expected to arrive this summer. The visual transformation of photos is planned for the second half of the year, and the generation of videos suitable for social platforms will follow.

Regarding control, there is no question of taking the reins away from the seller. The professional will be able to review, correct, and arbitrate. La Centrale will propose a price but will never impose one. The company's role is to provide a recommendation justified by data, not to replace the commercial decision.

Competition in the market

Is there currently an equivalent service on the market?

To my knowledge, not in this complete form. There are scattered components here and there, but not the entire integrated chain. However, some technologies that we use or adapt now come from OLX, which has already begun deploying certain functions, such as video-to-photo transformation, on other automotive marketplaces within the group. OLX's contribution is precisely this ability to industrialize faster, test on a larger scale, and reinject innovations into local markets.

Integration of WhatsApp

You have also integrated WhatsApp for professionals. Why use a third-party platform instead of keeping the conversation internal?

Because our job is not to confine exchanges but to streamline the connection between buyers and sellers. WhatsApp is already massively installed on smartphones, both among individuals and professionals. It is a tool that everyone knows, allowing for the exchange of text, photos, videos, and which integrates naturally into the daily lives of sellers.

For us, the real challenge is not to keep the conversation in a proprietary messaging system but to maintain lead tracking and service quality. The logic is the same as for the phone, where we use call tracking.

Fighting fraud

With AI advancing everywhere, how do you combat fraud?

Fraud has been a longstanding issue for us, particularly in the consumer-to-consumer (C2C) segment, and we have a dedicated team that has been working on it for years. The first layer is bot protection: we rely on market solutions like Datadome to detect non-human interactions, block bots, and also prevent data scraping.

Next, we have a whole layer of machine learning and generative AI to analyze conversations over time, as fraud does not occur from a single message but rather through a progressive scenario.

Finally, we use what is called a social graph: we cross-reference different signals (IP, email, phone number, vehicle, location, etc.) to track down a fraudster even if they change part of their apparent identity. La Centrale is also beginning to leverage OLX's expertise, which is highly exposed to these issues on its large C2C platforms.

Positioning of La Centrale

Many users still imagine La Centrale as a transactional trusted third party. Is that really your role?

No, not really. La Centrale connects a buyer and a seller but does not sell vehicles itself. In the vast majority of cases, the final transaction occurs offline, often directly at the dealership. This is especially true since most of our business is conducted with professionals. We assist with discovery, comparison, connection, and part of the follow-up, but we are not currently positioned as a transactional trusted third party in the sense understood on other second-hand platforms.