Rappi | Dynamic Geotargeting

Muttdata designed and built Megatron, as part of the Rappi Martech initiative: a production-grade paid media automation platform deployed on AWS and fully integrated with Rappi’s existing business workflows.

Automated creative management 

• Built all required code resources, database models, and processing services to support massive creative uploads.
  
  
• Deployed the full infrastructure on AWS, orchestrating workflows with Airflow and leveraging managed storage and persistence layers.
  
  
• Standardized cross-media campaign and ad group taxonomies to ensure deterministic relationships between creatives and targeting resources.
  
  
• Enabled automated activation and deactivation of creatives at the campaign and ad group level.
  
  
• Implemented execution status reporting and notifications, allowing non-technical users to understand outcomes and resolve issues without engineering support.
  

Business logic codification

A significant part of the work involved translating Rappi’s implicit operational knowledge into explicit, governed rules. Muttdata worked closely with Rappi’s teams to:

• Unify platform-specific execution logic
  
  
• Formalize decision rules that had never been documented
  
  
• Encode those rules into a repeatable, auditable system

This reduced dependency on individual operators and created a scalable foundation for future growth.

Dynamic geotargeting through optimization

A central differentiator of Megatron is its ability to operationalize Rappi’s highly dynamic and business-driven geographic definitions, something that traditional ad platform workflows were not designed to support.

Advertising platforms, however, impose strict constraints:

• Geotargeting is typically defined using points and radius-based circles, not polygons.
  
  
• There are hard limits on the number of geographic definitions that can be applied per campaign or ad group. These limits are significantly more restrictive when configurations are managed through platform user interfaces than through APIs (for example, dozens of locations via UI versus hundreds via API). In addition, geotargeting models differ across platforms: while Meta and Google Ads rely on point-and-radius logic, TikTok requires targeting based on predefined platform locations. Addressing this required standardizing both automated and manual input sources across all three platforms, ensuring consistent geotargeting behavior despite fundamentally different platform constraints.

• Manual configuration through platform interfaces makes frequent updates impractical and error-prone.
  

To bridge this gap, Muttdata developed a geotargeting optimization engine within Megatron that translates Rappi’s business-defined geographies into platform-compatible targeting configurations.

Starting from polygon based inputs, the system computes a finite and optimized set of circles that best approximates the intended region while respecting platform constraints such as:

• Minimum and maximum allowed radius
  
  
• Maximum number of circles per campaign or ad group
  
  

The optimization minimizes a formally defined error function, balancing two competing factors: over-coverage, where ads are shown in areas that should not be included, and under-coverage, where relevant areas are not reached. In addition, the algorithm favors solutions that achieve comparable accuracy using fewer circles when additional complexity does not produce a meaningful reduction in error, ensuring efficient and platform-compliant targeting configurations.

This approach enables precise control over how closely the platform-level targeting matches Rappi’s real operational footprint.

Crucially, these geographic definitions are fully dynamic and managed through business inputs. Marketing teams can activate or deactivate cities, adjust priority regions, or update custom segments without manual reconfiguration in each ad platform. Changes made in a Google Sheet propagate automatically through Megatron to the corresponding campaigns and ad groups.

As a result, Megatron enabled Rappi to execute geotargeting strategies that were previously impossible to deploy with accuracy, while also making frequent updates operationally viable. What was once a rigid, manual process became a flexible, scalable capability aligned with real business needs and local market dynamics.

What makes this capability particularly impactful is that it does not merely optimize an existing workflow, it enables an entirely new class of geotargeting strategies. Before Megatron, some scenarios were operationally infeasible to execute with accuracy. The complexity of these definitions, combined with platform limitations and manual configuration effort, meant they were either approximated or not executed at all. By shifting geotargeting from a manual, interface-driven task to an automated optimization problem, Megatron turned geography into a flexible, business-controlled variable, making frequent updates, high granularity, and cross-city segmentation not only possible but practical at scale.