Google Research: Rerouting Less Than 2% of Trips Reduces Urban Congestion and CO2 Emissions

Road congestion is not merely a frustration — it carries a measurable economic and environmental cost. Google Research set out to investigate whether a navigation app alone, without any infrastructure investment, could statistically reduce traffic bottlenecks. The answer they found, published in the prestigious journal Nature Cities, is suggestive: it can, and with far less intervention than intuition would suggest.

The research was conducted by software engineers Neha Arora and Aboudy Kreidieh from Google Research along with a group of collaborators including Alexandre Bayen and Andrew Tomkins. The methodological framework chosen for the study was a switchback experimental design — alternating treatment days and control days within the same urban areas, minimizing the influence of temporal and seasonal factors on the interpretation of results.

Six Months of Experiments in Ten Major US Cities

The experiment ran for six months and covered 10 major US cities. For each city, the researchers identified approximately 100 traffic segments with historically high congestion. They modified the Google Maps routing algorithm to deliberately avoid those segments — but only for a small fraction of trips passing through that area.

The key was the size of the intervention: less than 2% of observed trips per city participated in the treatment. Rather than massively rerouting users, the system selectively changed recommendations for a small fraction of trips, distributing traffic across peripheral road networks and reducing concentration on congested segments. The algorithm selected alternative routes of similar total travel time, so that users would not suffer significant efficiency losses.

Data analysis used hierarchical Bayesian modeling that accounted for variation at the level of individual cities and individual hours, isolating the true effects of the intervention from background traffic noise that would otherwise obscure the results.

How Much Traffic Needs to Be Rerouted to Reduce Urban Congestion?

The results surprised many: intervention on fewer than 2% of trips led to a measurable speed improvement on target segments. The median increase in driving speed on treated segments was +2%. Looking at the broader road network — all segments affected by traffic redistribution — the median speed increase was +0.35%, reaching +0.5% during peak hours.

The reduction in fuel consumption on targeted segments was a median 0.5 to 1.0%. These figures may appear modest in isolation, but they need to be viewed in context: this is an effect achieved without any physical intervention in traffic infrastructure, without road closures, and without forcing drivers onto longer routes.

What is particularly important is that the positive effects are not limited to Google Maps users. By dispersing traffic across a wider network, drivers who do not use the navigation app also benefit from freer passage. This is critical for assessing the true reach of the effect: when fewer than 2% of trips change their route, the traffic situation improves for all road users.

Potential Savings: Thousands of Tonnes of CO2e Per City Per Year

The researchers estimate that this approach can generate savings of thousands of tonnes of CO2e emissions per city per year — even at these intervention levels. Across the ten tested cities, the total climate benefit could be substantial, and all without additional cost to city authorities or app users.

The study openly addresses the limitations of the methodology: the switchback design precludes measuring long-term changes in driver behavior, and results may vary depending on traffic culture, road network density, and the specific characteristics of individual cities. But the strength of the research lies precisely in the fact that it was conducted under real conditions, with real drivers, in real urban traffic — not in a simulated or modeled environment.

The full title of the paper published in Nature Cities is “Urban congestion relief experiments through routing-app interventions.” For cities struggling with traffic congestion, Google Research’s findings point to an accessible, non-infrastructural tool that already exists in the pocket of every smartphone user.