How Amplitude Empowers Mercado Libre to Be AI-Native: Scaling Intelligence Across Latin America's Largest E-Commerce Platform

Mercado Libre isn’t just Latin America's largest ecommerce platform. It’s becoming one of the most AI-native companies in the world.

With operations spanning 18 countries, from North America to Patagonia, Mercado Libre (MeLi) collects billions of behavioral events from these operations every day. The real ambition, however ,was something more transformative: making every employee, in every function, capable of making real-time, data-driven decisions, without having to become an analytics expert.

This is the story of how MeLi is using Amplitude's AI capabilities—AI Agents, dashboards, and Model Context Protocol (MCP)—to compress the time from sensing to deciding to acting. It's about building an operating model where analytics becomes invisible infrastructure, powering decisions continuously rather than episodically.

The challenge: Analytics at continental scale

Running analytics for a company the size of MeLi presents challenges that most organizations never encounter. Consider the geography alone: delivery operations stretch across a continent and more, from bustling urban centers to remote rural communities. Each region has its own logistics partners, regulatory requirements, and customer expectations.

Before Amplitude, MeLi’s analytics teams faced a familiar enterprise problem at an unfamiliar scale. Knowledge was fragmented across functions. Teams discovered issues reactively. Analysis requests piled up in queues. And even when insights emerged, translating them into coordinated action across product, operations, and business teams required manual effort that couldn’t keep pace with the business.

MeLi has since already invested heavily in building an analytics culture. They grew from near-zero to over 10,000 employees with Amplitude access in roughly a year, with 2,126 monthly active users representing 37% year-over-year growth. Today MeLi is processing 782 billion events—a scale that would overwhelm most analytics implementations.

But scale created its own problems. With so many people using analytics, how do you ensure consistency? How do you help non-analysts access insights without overwhelming trained practitioners? And how do you move from dashboards that people check occasionally to intelligence that actively drives decisions?

The answer required rethinking analytics not as a tool people use, but as an AI decisioning that works for people.

The solution: A three-pillar AI transformation strategy

MeLi's approach to becoming AI-native rested on three integrated pillars, each building on the last:

The first was the behavioral data foundation as context. Amplitude Analytics, Session Replay, and Experimentation provided the trusted data plane where customer behavior is captured, understood, and acted upon. This behavioral context would capture the full customer journey, from first click to purchase to delivery.

The second was the AI execution plane. This is where Amplitude's agents would transform passive dashboards into active participants in the business. Agents continuously monitor funnels, identify anomalies, summarize findings, and distribute insights to stakeholders—all without requiring human initiation.

The third was the distribution plane, powered by Model Context Protocol. MCP enables MeLi's employees to access Amplitude's governed behavioral context through whatever AI tools they already use—Claude, Cursor, or their internal assistant called Verdi. With 63 MCP users generating 1,222 tool calls in the last 90 days, MeLi employees can ask questions against real Amplitude data without needing to learn an analytics interface.

This integrated approach means analytics isn't something employees do—it's effectively a decisioning engine for them to act on. The dashboard remains the canonical source of truth, but AI agents and MCP connections ensure that truth flows to wherever decisions are made.

Proof point: The Flex shipments use case

Flex is MeLi's flexible delivery program in Brazil, enabling sellers to choose couriers that can reliably meet delivery service-level agreements (SLAs). However, if sellers searched for an available courier and found none, friction occurred. Failed searches led to slower deliveries, frustrated sellers, and eventually churn.

Before Amplitude, identifying these coverage gaps required manual analysis. Teams would notice problems after they became severe. Prioritizing which regions to address first involved debates driven by anecdotes rather than data. And coordinating across product, business, and transport teams meant translating findings into multiple formats for different audiences.

MeLi's solution combined an Amplitude dashboard that made unmet demand visible at a granular geographic level with agents that operationalized that dashboard into automated, shared reporting.

The dashboard itself was the first breakthrough. By visualizing where sellers searched for couriers and found none—broken down by province and city—the team could quantify unmet demand and prioritize the highest-impact zones. Product, business, and transport teams used the same dashboard as a shared source of truth, eliminating the translation layer that slowed previous efforts.

But the real transformation came from pairing dashboards with agents. Agent-generated reports provided proactive, automated visibility into feature performance. Teams didn't need to request ad-hoc analysis to notice regressions or new pockets of unmet demand—the system told them. This created a closed-loop optimization cycle: identify unmet demand, onboard couriers with stronger SLA performance in priority zones, monitor the impact, and continuously refine.

The results speak clearly:

• Dropoff reduction: Searches with no available courier decreased 9.7%, from 31.8% to 22.1%, after targeting key areas of unmet demand identified in Amplitude.
• Feature adoption: 60% of Flex sellers in Brazil now complete shipments through the feature.
• SLA improvement: Feature-driven shipments reach 98% SLA, which is 2.5% higher than shipments outside the feature (95.5%).

This wasn't just a one-time project. With dashboards and agents acting as always-on instrumentation, teams can continuously tune coverage and courier onboarding as demand shifts. The operating model changed: from reactive discovery to continuous optimization.

The AI-native future

MeLi's transformation didn't happen in isolation. Amplitude's partnership with Minders, a martech agency specializing in analytics enablement, provided the operational arm for this rollout. Minders embedded in MeLi's cadence—executive business reviews, enablement sessions, pilot programs—helping translate business priorities into concrete Amplitude workstreams across Session Replay, Web Experimentation, and Marketing Analytics.

This partner model proved essential for change management at scale. When you're training hundreds of people simultaneously on new ways of working, you need more than software—you need people who can bridge the gap between capability and adoption.

Looking ahead, MeLi's AI-native trajectory continues to evolve. Their MCP implementation is moving toward deeper integration where internal agents can trigger Amplitude-powered actions—governance tickets, dynamic cohorts, annotations—directly from their workflow tools. The goal is "AI in the workflow, grounded in governed behavioral truth."

For MeLi, becoming AI-native isn't about novelty. It's about industrializing the time from insight to action until analytics becomes invisible—not because it doesn't matter, but because it's woven into how everyone works.