MercadoLibre automates vendor image moderation
with AI

Latin America’s largest eCommerce ecosystem uses in their vendor onboarding processes to assess images faster and more accurately.

The big picture


reduction in rollback rates across 3 months


reduction in the number of calls coming from sellers


reduction in false-positive detection in text moderation

88% of shoppers think product images are crucial for purchasing. Marketplaces and brands understand this and have invested significant resources in ensuring product images have clarity and create internal guidelines to secure quality. For marketplaces, a big part of the challenge in maintaining this quality comes from the vendor onboarding process. Every single image that a marketplace onboards needs checking to ensure guideline adherence and product credibility.

MercadoLibre is Latin America’s largest eCommerce ecosystem. They operate in over 18 countries and serve more than 550 million people online. Their marketplace is composed of a mix of small sellers and big brands. They have roughly 12M sellers in their database and a little more than 642 million visits per month from 18 countries across Latin America.

MercadoLibre started using’s automated image moderation solution to help improve the image QA process.’s AI image recognition technology automates image and content moderation. It can help ensure images are compliant with business guidelines - and can do this in half the time and in a way that would require minimal human intervention. AI also removes the prospect of human error - leading to faster and more accurate vendor image onboarding.

The need

MercadoLibre routinely onboards and sells hundreds of thousands of products from various vendors. And every single image needs to be checked to ensure quality control. A manual process was in place to check for image quality. Manual image moderation was inadequate for the scale of operations they were dealing with. It was also resource-intensive and prone to errors that slowed down the time taken to monetize a product.

To solve these issues, MercadoLibre worked with to:

Automate the image quality checking process

Work across all the guidelines required by them

Reduce the effort required by their team to do QA

Be trained through the feedback and QA loop

Streamline the entire process end to end, and save time

MercadoLibre chose since the solutions offered would help them reduce vendor onboarding time, reduce dependence on manual QA, and ensure that image quality guidelines were adhered to on their site.

How it works: Image and text moderation's image moderation tool automatically checks the quality of images sellers onboard into marketplaces.
Incoming images that don't meet the guidelines are flagged and returned to the seller. Vendors receive a notification saying they are losing exposure on the listing.

MercadoLibre uses the tool to check:

The presence of any watermarks

If any text on the image is not a part of the product

If there are borders that impact product image usability

Whether there are multiple products within a single image

Whether the background complies with the specified guidelines


Reduced rollback rates through better understanding of
customer guidelines

Images coming into marketplaces are sometimes inaccurately marked and sent back to the vendor. The seller/vendor in such a situation can contact the customer team to call out the error and apply for a rollback.

  • is being used across the vendor's site to check product imagery coming in from sellers for text, borders, and background compliance.
  •'s image moderation tool understands customers' guidelines and has the ability to re-train with a constant feedback loop to mark images as accepted or rejected effectively.
  • MercadoLibre saw a 70% reduction in rollback rates across three months with a 45% average Month-on-Month reduction in rollback rates.

Reduced contact rate from robust automated moderation’s image moderation tool helped by accurately marking images as ‘Accepts’ or ‘Rejects’ lowered the number of contact tickets raised by sellers to MercadoLibre.

  • During QA, when images are marked inaccurately, the seller can contact the marketplace or raise a ticket to highlight an error. The number of tickets raised or contacts made represents how accurately the tool has accepted or rejected an image.
  • The constant retraining and feedback loop built into’s image moderation tool ensures images are marked according to the latest guidelines, and the system learns from errors made.
  • MercadoLibre saw the number of contacts from vendors reduce 30% on the average month on month across three months. The reduction across three months was 51%.

Key takeaways

  • MercadoLibre works as the main point of contact between vendors and their customers. Detecting text and watermarks that might contain information about vendors ensures customers don’t bypass the site and contact vendors directly.
  • There is an observed reduction in false positives. Lower false-positive rates imply fewer images that are flagged as requiring further review by the vendor. This ensures greater efficiency of human moderators where necessary and more cost savings for MercadoLibre and the vendors.
  • AI-powered image moderation systems save both time and money by automating an otherwise manual process. Vendors receive feedback in real-time on the images uploaded, resulting in higher productivity and money held. Automated feedback loops also ensure mistakes are not repeated.