AI forecasting seen lifting taxi earnings 20%

The DASH-HKUST system points drivers to demand hotspots, lifting trips and cutting idle time.

An AI-powered demand forecasting system developed by transport company DASH and the Hong Kong University of Science and Technology (HKUST) could increase taxi drivers' earnings by almost 20% and boost completed trips by more than 30%, according to interim results from the project.

The system, called StreetSights, is designed to help drivers decide where to go for their next fare by predicting where passenger demand and empty taxis are likely to be across more than 350 areas in Hong Kong every 15 minutes.

The model was built using data from about 2,000 active taxis and more than 6.45 million recorded trips. DASH and HKUST said it can predict passenger demand and taxi supply with more than 90% accuracy.

To test the system, the partners ran a simulation using historical trip data where they created virtual "ghost" drivers that started from the same locations and worked the same hours as 100 real drivers but followed StreetSights' recommendations.

The simulation found that the AI-guided drivers completed 31.2% more trips and earned 15.8% more in daily fares than the real drivers.

Waiting time between trips fell by 30%, whilst passenger matching improved by 14.7%, according to DASH and HKUST.

A live field test is planned for the fourth quarter of 2026 before any wider rollout.

The project comes as Hong Kong prepares a licensing framework for ride-hailing platforms and taxis. DASH and HKUST said the collaboration shows how local transport data can be used to improve the efficiency of the city's street-hail taxi network.

Street-hail taxis account for about 60% of taxi demand in Hong Kong, but the segment has been harder to measure and improve than app-based ride-hailing services, the organisations said.

In its first year, DASH said its platform handled about 6.1 million trips covering 92 million kilometres across Hong Kong.

Jason Ma, co-founder and managing partner of DASH, said the project combines university research with real operating data to give drivers better information without changing how they work.

"As Hong Kong charts the next chapter of its transport future, the question is not how many vehicles are on the road but how well every one of them serves the city," Ma said.

Professor Hong K. Lo, dean of engineering at HKUST, said the partnership shows how local data can be turned into practical tools for the transport sector.

"Our role at HKUST was to turn that data into a predictive model rigorous enough to withstand academic scrutiny and practical enough to make a difference on the road," he said.