How Mercado Libre Turned Push Campaigns into an Autonomous AI Growth Engine

About The Company

MercadoLibre is the largest online commerce and payments ecosystem in Latin America.  Through a suite of technology solutions including Mercado Pago, Mercado Ads, Mercado Envios, and Mercado Crédito they enable customers in 18 countries to carry out their commerce, offering solutions across the entire value chain.

Executive Summary

MercadoLibre’s Engagement & Growth teams were managing a complex ecosystem of recurring promotional push campaigns across markets. As the number of campaigns, value propositions, and user segments grew, so did operational complexity.

MercadoLibre partnered with Muttdata to:

• Unify and automate campaign orchestration
• Introduce robust monitoring and governance
• Productize and scale a Reinforcement Learning personalization model
• Transform static lifecycle campaigns into an autonomous, self-learning system

What began as operational simplification evolved into a fully AI-driven growth engine.

The Challenge

The Engagement & Growth team managed more than 80 recurring push campaigns, each tied to specific value propositions, coupon strategies, and lifecycle-based segments. Campaign execution required coordination across multiple internal systems, with:

• Managing several tools simultaneously without systematic cross-validation between them
• Manual configuration steps across different systems for each campaign
• Fixed discount amounts
• Fixed send times and frequency
• Manual monthly budget validation
• Static user segmentation logic
• Limited user-level traceability
  
  

While each tool worked as intended, the lack of unified orchestration and automated validation introduced operational complexity. Errors could occur across systems and were not always immediately detectable.

As campaign volume and experimentation increased, this approach created friction and limited the ability to scale personalization and testing efficiently.

The opportunity was clear:

• Reduce operational complexity
• Increase experimentation velocity
• Move from segment-level generalization to user-level personalization
• Leverage AI to optimize value delivery in real time

The Solution

To transform push marketing into a scalable, intelligent system, Mercado Libre and Muttdata approached the challenge in two strategic phases: operational transformation and AI-driven personalization.

1. Operational Orchestration at Scale

Before introducing AI, the campaign ecosystem needed to be unified and automated.

Muttdata began with a comprehensive discovery phase, mapping the full landscape of pre-existing internal tools involved in campaign creation, segmentation, coupon management, and push delivery. These systems had evolved and were deeply embedded in MercadoLibre’s technical ecosystem.

Through this process, we not only analyzed each tool individually, but also uncovered how they interacted with one another, something that had not been fully mapped across Marketing, Marketing Science, and Engagement & Growth. This discovery created immediate value by clarifying dependencies, workflows, and points of friction within the ecosystem.

Based on these findings, Muttdata designed and implemented an orchestration layer that integrated all relevant systems while fully adapting to MercadoLibre’s existing infrastructure.

This orchestration layer:

• Centralized campaign configuration into a Single Source of Truth
• Connected previously fragmented internal systems
• Automated validations before activation
• Introduced budget control checks
• Simplified audience logic and coupon management
• Added robust monitoring and observability
• Enabled user-level traceability for every push notification

Rather than replacing existing tools, the solution unified them, respecting the company’s technical architecture while eliminating operational bottlenecks.

This foundation significantly reduced operational friction, enabling the Engagement & Growth team to shift its focus from manual execution to strategic experimentation.

It also made something critical possible: scalable, reliable learning in production.

2. Reinforcement Learning Personalization

With the infrastructure stabilized, the next evolution was intelligence. Together with MercadoLibre’s data science team, Muttdata put into production a Reinforcement Learning system based on a contextual bandit framework, moving from segment-level decisioning to user-level optimization.

Instead of manually defining:

• Coupon values
• Wording
• Send time
• Send day
• Frequency
  

The system continuously learns from real interactions. It balances:

• Exploration — testing new combinations
• Exploitation — prioritizing high-performing ones

For each user, the model dynamically determines the optimal value proposition based on behavioral signals and prior responses.

To support this, Muttdata built a robust MLOps backbone, including:

• Preprocessing pipelines
• Automated training and retraining
• Daily inference generation
• Country-level model specialization
• Real-time monitoring via Slack
• Full traceability of coupon, wording, timing, and performance

Every notification becomes feedback for the next decision. The result was no longer campaign optimization. It became continuous learning.

Outcome

The transformation delivered measurable impact across both operational efficiency and revenue performance.

Revenue & Performance Impact

The transformation delivered measurable impact across operational efficiency, revenue performance, and scalable AI-driven optimization.

Beyond revenue growth, the initiative significantly improved marketing efficiency and governance:

• 50% reduction in audience administration effort
• 11 hours per month saved in monitoring
• 18 hours per month saved in audience creation and updates
• Simplified management of 80+ campaigns
• Automated budget validation before activation
• Preventive coupon validation before sending
• Improved targeting of eligible users

The operational transformation enabled faster experimentation, improved campaign control, and more strategic allocation of marketing resources. The introduction of automated orchestration and reinforcement learning had a significant impact on performance. This means:

• Higher open rate on push sent under the Bandit model
• Even higher incremental lift vs. business-as-usual (BAU) in Average Treatment Effect under the reinforcement learning (bandit) model. 

This means that not only did notification open rates increase, but the users targeted by the model also showed higher incrementality.

In earlier automation phases, even before full AI personalization was deployed, campaign management improvements allowed for easier experimentation which also contributed to increased revenue.

These results demonstrate both immediate performance gains and the long-term compounding potential of continuous learning.

The reinforcement learning model demonstrated statistically significant performance improvements while continuing to explore new opportunities.

This created not just a short-term lift, but a system capable of compounding performance over time.

‍

Impact

‍

The team moved from manual orchestration to strategic control.

Mexico operations scaled from pilot (5–20%) to full 100% deployment, with Brazil already partially migrated and additional channels planned for expansion, including other communications and Mercado Pago.

Wrap up

For CMOs operating at enterprise scale, the lesson is clear:

‍

By first unifying and automating campaign operations, we created the foundation necessary for reinforcement learning to thrive in production.

The result is a push marketing system that:

• Learns continuously
• Personalizes at the user level
• Optimizes value delivery in real time
• Scales across markets
• Generates measurable incremental revenue

What began as operational simplification evolved into an autonomous AI-driven growth engine, and because the system is built on continuous exploration and exploitation, its performance compounds. That is the power of combining orchestration, governance, and reinforcement learning at scale.

Do you want similar results for your e-commerce platform? Book a call with us!