Reshaping retail media for Mercado Libre with GenAI

Introduction

Mercado Libre, AWS and us joined forces to tackle the challenge of producing high-performing banner ads with generative artificial intelligence (Gen AI), for dozens of thousands of smaller sellers. Using Amazon Bedrock, Amazon S3, and Amazon DynamoDB, we developed GenAds, an innovative solution that automates the creation of compelling product images, streamlining the process for sellers and enhancing their market presence with image-generating AI. Through GenAds, Mercado Libre helped small sellers through image generation for advertising, enabling sellers of all sizes to attract customers, increase click-through rates (CTR) by an average of 25%, and boost sales effectively.

The Company

Mercado Libre is the largest online commerce and payments ecosystem in Latin America, operating in 18 countries and offering comprehensive solutions across the entire e-commerce value chain.

The Challenge

In e-commerce, the visual impact of product images is often the deciding factor in a successful sale. High-quality visuals significantly shape a potential customer’s experience, yet creating compelling display ads remains a challenge for many vendors. It’s a process that demands time, expertise, and resources.

This includes understanding the buyer persona to designing and producing multiple versions of promotional images. For small and medium-sized businesses (SMBs) limited budgets and resources often prevent them from accessing professional agencies that could elevate their product promotions. Many sellers also underestimate the power of well-crafted banners in driving visibility and sales.

Considering that 80% of Mercado Libre’s sellers are SMBs, they were looking for a solution tailored to empower SMBs and accelerate their growth on the platform through its ad products. This innovation comes at a time when digital advertising is booming, projected to grow by 6.5% in 2024 to $442.6 billion, representing 58.8% of global ad spend. Retail media is at the forefront of this growth, with an impressive 17.2% Compound annual growth rate (CAGR) over three years.

Retail media spending alone is expected to climb by 10.2% this year and 10.5% next year, reaching $141.7 billion and accounting for 13.6% of total ad spend (Source: Dentsu and Mobile Marketing Magazine). In this thriving landscape, solutions that enhance ad performance and metrics aren’t just relevant— they’re essential for sellers aiming to stay competitive and succeed .

Digital advertising is set to grow by 6% in 2024, reaching $442 billion, with retail media growing at an impressive 17% CAGR over the next three years.

Solution | GenAds: a new AI-driven product imaging service

We worked closely with Mercado Libre to develop a cutting-edge solution to streamline the creation of high-quality product images for its sellers. Leveraging AWS’s robust infrastructure and AI tools, the companies collaborated to create GenAds, an innovative, AI-powered tool that empowers small and long-tail sellers to quickly go to market with visually impactful product listings.

Here are a few images created with GenAds.

How GenAds Works

Mercado Libre kicks off the process by pre-selecting sellers and their products using performance and popularity metrics. From this initial group, products are further curated for advertising purposes. By leveraging Stability AI’s image-generation model Stable Diffusion within Amazon Bedrock, the tool produces contextualized images that place products against visually appealing backgrounds. It also fine-tunes key details like reflections and shadows to achieve a professional-grade finish.

Only the most visually captivating and technically precise images pass the final review, meeting strict advertising standards such as high resolution and proper cropping. This automated process not only lightens the load for sellers but also speeds up the creation of impactful ads, helping Mercado Libre’s sellers capture customer attention and succeed in the competitive e-commerce arena.

GenAds streamlines ad creation for sellers on the marketplace. Sellers can list items and opt for display ads featuring automatically generated banners. Each week, category-based backgrounds are generated using Stable Diffusion via Amazon Bedrock, stored in Amazon Simple Storage Service (Amazon S3), and indexed in Amazon DynamoDB. Daily, a selection mechanism identifies popular and visually appealing items, and Anthropic’s Claude 3 Sonnet model creates prompts to blend these items with the backgrounds, forming banners that are stored similarly. Finally, an API enables the front end to retrieve and display these banners for ad campaign proposals.

Results

If you want to learn more about this case and how we crafted this solution with our expert #DataNerds, book a call with us and we'll be happy to tell you more.