Google Maps' Tiny Reroute Experiment Cut Traffic and Emissions City-Wide

Google Research just dropped the first large-scale, real-world evidence that navigation apps can act as a distributed traffic control tower. In a study published in Nature Cities, they tweaked Google Maps to prefer alternative routes with similar travel times for a tiny fraction of trips—under 2%—and measured the impact across 10 major US cities over six months. The result: measurable, statistically significant improvements in network efficiency.

How the experiment worked

The team selected roughly 100 historically congested road segments per city. On treatment days, the routing algorithm was modified to steer trips away from those segments whenever a similarly-costing alternative existed. The design was a city-wide switchback (crossover) experiment—alternating between modified and unmodified routing on consecutive days. This avoided the noise of per-trip randomization and captured system-level effects.

Only trips that would have crossed the pre-selected bottlenecks were affected. That's the key insight: you don't need to reroute everyone. Just a small, targeted nudge can dissipate the demand that causes recurrent congestion.

Results that matter

Using a hierarchical Bayesian model to analyze outcomes, the study found:

• Median driving speeds on targeted segments increased by ~2%.
• Fuel consumption on those segments dropped 0.5–1.0%.
• Across all affected segments (including those that absorbed diverted traffic), speeds improved by ~0.35% on median, and ~0.5% during peak morning/afternoon hours.
• Extrapolated to city scale, this translates to thousands of tons of CO2e saved per city per year.

The peripheral roads didn't get clogged. Because the diverted traffic was spread across multiple alternative routes with similar travel times, the system as a whole ran smoother. This is the opposite of the classic Braess's paradox nightmare where adding capacity makes things worse—here, intelligent dispersion works.

Why this matters for engineers

This isn't a simulation or a toy model. It's a production-grade experiment on one of the world's largest navigation platforms. The paper provides a framework for moving from individual trip optimization (fastest route for you) toward cooperative routing that improves total network efficiency. For anyone building routing, logistics, or smart-city infrastructure, the implications are direct: even modest, system-aware interventions can produce outsized gains.

The full paper is available in Nature Cities. Google has also open-sourced their experimental design methodology—worth a read if you're doing large-scale A/B testing on infrastructure.