Google reroutes under 2% of trips and cuts city-wide congestion

Google Research ran a six-month experiment across 10 major US cities testing whether coordinated, network-aware routing in Google Maps could ease traffic system-wide rather than just optimizing individual trips. In each city, researchers identified roughly 100 chronically congested road segments and, on alternating “treatment” days, nudged any trip heading through those bottlenecks onto alternative routes of comparable travel time and road type. The design was a city-wide switchback that flipped between the modified and unaltered algorithms day to day, and it touched fewer than 2% of observed trips.

The measured gains were small but statistically significant. Analyzed with a hierarchical Bayesian model that pooled effects across cities and hours, targeted segments saw driving speeds rise about 2% and fuel burn drop 0.5–1.0%. Across the broader set of roads affected — including peripheral streets absorbing diverted vehicles — median speeds still improved around 0.35%, rising to 0.5% during morning and evening peaks. Crucially, dispersing traffic away from major bottlenecks let those secondary roads keep moving even under heavier load, and the researchers estimate the intervention could avoid thousands of tons of CO2e per city each year.

The significance is less the raw percentages than the proof of concept: benefits accrued to all drivers, not only Maps users, showing that a navigation platform can act as the “control tower” ground transportation has always lacked. Published in Nature Cities, the work establishes a rigorous experimental framework for moving from individual-trip optimization toward cooperative routing, and the authors frame it as a template for future smart-city efforts like dynamic signal control and real-time network optimization.