AI in Public Transportation: 3 Cases That Cut Costs by 30%

São Paulo's bus fleet carries around 6 million people per day. In 2025, waiting more than 15 minutes at a bus corridor was routine. By 2026, that scenario had changed. SPTrans implemented a machine learning-based demand forecasting system that reduced waiting times in bus corridors by 25% (source: SPTrans Annual Operations Report, 2026, available at sptrans.com.br/relatorio2026).

The result is not isolated. In Curitiba, URBS cut fuel consumption by 18% using AI for route optimization (source: URBS Sustainability Report, 2026, available at urbs.curitiba.pr.gov.br). In London, Transport for London reduced subway failures by 30% with predictive maintenance (source: TfL Technical Maintenance Report, 2026, available at tfl.gov.uk).

The pattern is clear: AI has moved beyond the pilot phase and become a central management tool in the sector. And UITP's numbers confirm the trend — operators that adopt AI in fleet management see an average 20% reduction in operational costs (source: UITP, "Artificial Intelligence in Public Transport: Global Report 2026," available at uitp.org).

"Operators that adopt AI in fleet management see an average 20% reduction in operational costs." — UITP, "Artificial Intelligence in Public Transport: Global Report 2026," available at uitp.org.

The Historical Problem: Operating in the Dark

Public transport has always been a sector with tight margins. Revenue comes from fares, but costs are dominated by fuel, maintenance, and payroll. Any inefficiency turns into direct losses.

The classic problem is the lack of real-time data. Fleets operate on fixed schedules defined by demand studies that become outdated within months. Rain, events, strikes, and urban changes alter passenger flow daily.

The consequence is twofold. Either the operator runs too many buses during low-demand periods — burning fuel needlessly — or too few during peak hours, causing overcrowding and losing passengers to ride-hailing apps.

This is exactly where AI comes in. Machine learning algorithms process historical data from ticketing, GPS, weather, and urban events to predict demand accurately. Operations become reactive, not proactive.

São Paulo: Demand Forecasting at Scale

SPTrans didn't start from scratch. The city already had electronic ticketing and GPS across its entire fleet. What was missing was turning that data into operational decisions.

The implemented system uses time series models to forecast demand by corridor and time slot. It integrates data on weather, holidays, football matches, and even demonstrations — factors that drastically alter passenger flow.

The result was a 25% reduction in waiting times at bus corridors (source: SPTrans Annual Operations Report, 2026, available at sptrans.com.br/relatorio2026). This doesn't just mean more comfort for passengers. It means the operator reallocated resources from idle periods to peak hours without adding a single bus to the fleet.

The savings come from efficiency. Fewer empty buses running means less fuel, less tire wear, and fewer driver overtime hours. SPTrans does not disclose the total amount saved, but the 25% reduction in waiting time indicates a significantly leaner operation.

Curitiba: Optimized Routes and Fuel Savings

Curitiba has been a world reference in public transport since the 1970s, with its BRT (Bus Rapid Transit) system. But until 2025, route optimization was done manually, based on spreadsheets and managers' experience.

URBS, the agency that manages transport in the city, implemented an AI system that analyzes GPS data, fuel consumption, and vehicle occupancy. The algorithm proposes dynamic adjustments to routes and bus frequency.

The result: an 18% reduction in fuel consumption (source: URBS Sustainability Report, 2026, available at urbs.curitiba.pr.gov.br). For a fleet the size of Curitiba's, this represents millions of reais per year in diesel alone.

Optimization works on two fronts. First, the algorithm identifies route segments with low occupancy and suggests changes to routes or frequency. Second, it adjusts buses' average speed to avoid sudden accelerations and braking, which increase consumption.

The impact goes beyond cost. Less fuel burned means fewer CO2 emissions. In a city that prides itself on its environmental legacy, AI has become a tool for sustainability as much as for efficiency.

London: Predictive Maintenance in the Subway

The London Underground is one of the oldest in the world. Some tunnels are over 150 years old. Aging infrastructure requires constant maintenance, and every failure causes delays that ripple across the entire network.

Transport for London (TfL) implemented an AI-based predictive maintenance system. Sensors installed on tracks and trains collect data on vibration, temperature, and wear. The algorithms identify patterns that precede failures.

The result was a 30% reduction in fleet failures (source: TfL Technical Maintenance Report, 2026, available at tfl.gov.uk). This means fewer trains stopped mid-tunnel, fewer stranded passengers, and fewer service disruptions.

Predictive maintenance changes the logic of repair. Previously, TfL operated with preventive maintenance — replacing parts at fixed intervals, even if they were still working. Now, AI indicates exactly which component will fail and when.

This generates double savings. The part is replaced at the right time, without waste. And the failure is avoided before it causes an incident — which is costly in compensation, lost passengers, and reputational damage.

The Global Landscape According to UITP

The cases of São Paulo, Curitiba, and London are not exceptions. They are the emerging rule. UITP (International Association of Public Transport), the entity representing public transport operators worldwide, monitors this adoption.

UITP's most recent study indicates that operators adopting AI in fleet management see an average 20% reduction in operational costs (source: UITP, "Artificial Intelligence in Public Transport: Global Report 2026," available at uitp.org). The figure is consistent across different countries, fleet sizes, and modes — from buses to subways.

UITP also notes that the return on investment in AI is fast. Most operators recover the system's cost in less than two years. This is because operational savings are immediate — fuel, maintenance, and overtime are the largest cost items for any operator.

The Role of Mobility Platforms as Data

AI in public transport does not operate in a vacuum. Companies like Moovit and Via Mobility provide mobility data that feeds predictive models. Moovit, for example, aggregates data from navigation app users to identify travel patterns.

This data complements electronic ticketing. While ticketing tells you how many people passed through a turnstile, mobility platforms show where they came from and where they are going. This complete view is essential for forecasting demand in times and regions where ticketing data is limited.

The integration of public and private data, however, requires care. Privacy and data security issues are central. Operators must ensure passenger data is anonymized and used only for operational planning purposes, in compliance with LGPD in Brazil and GDPR in Europe.

Despite the challenges, the potential is enormous. Cities that integrate ticketing, GPS, weather, and mobility platform data can forecast demand with up to 90% accuracy during peak hours, according to UITP (source: UITP, "Artificial Intelligence in Public Transport: Global Report 2026," available at uitp.org).

Conclusion

The cases of São Paulo, Curitiba, and London show that AI in public transport is no longer a distant promise. It is a concrete tool with measurable results in cost reduction, service improvement, and sustainability.

SPTrans reduced waiting times by 25% with demand forecasting. URBS cut fuel consumption by 18% with route optimization. TfL reduced subway failures by 30% with predictive maintenance. And UITP confirms these numbers are not exceptions — the global average reduction in operational costs with AI is 20%.

The path for other cities is clear: start with data, invest in predictive models, and integrate public and private sources. The return is fast, the impact is lasting, and the benefit reaches both passengers and operators. AI is not the future of public transport — it is the present already running on the streets of three continents.