Predictive Logistics AI in LATAM, Fernando Angulo

Constraint → AI Capability

Informal Address Grids

Favelas · Comunas · Colonias

Up to 40% of LATAM addresses are unstructured: "casa amarilla, frente a la panadería." GPS coordinates fail without context.

→ Forces Geocoding ML on landmarks + crowd-sourced waypoints

Volatile Weather Windows

Tropical storms · Andean rains

Bogotá hailstorms, São Paulo flash floods, and Caribbean hurricanes can collapse a delivery route in 30 minutes.

→ Forces Real-time route reoptimization with weather-feed RL

Megacity Traffic Density

São Paulo · CDMX · Bogotá

São Paulo loses 86 hours/year per driver to congestion. CDMX averages 15 km/h in rush hour. Static ETAs are useless.

→ Forces Time-of-day predictive ETA models per micro-corridor

Restricted Vehicle Zones

Pico y Placa · Hoy No Circula · Rodízio

Bogotá, Mexico City, and São Paulo rotate plate restrictions by day and license number, fleets juggle compliance hourly.

→ Forces Plate-aware fleet assignment + multi-modal handoff

Cash-on-Delivery Economy

~40% of LATAM e-commerce

Unbanked buyers pay drivers in cash. Returns spike, fraud risk rises, and routing must factor change-making and security.

→ Forces Demand forecasting + driver-cash optimization + fraud scoring at the doorstep

Operationalized by Mercado Libre · Rappi

In LATAM, the constraint is the capability. Fragmented infrastructure didn't slow logistics AI, it forced a more sophisticated stack than anything Silicon Valley needed to build.

Fernando Angulo · Senior Market Research Manager at Semrush, an Adobe company