#1454 2013 · Seoul Metropolitan Government (Owl Bus) · Urban night transit
Seoul designed its night bus routes from 3 billion phone call records
the problem
Night buses lose money on guesses; surveys cannot see who moves after midnight
background
Late-night transit is the hardest planning problem in public transport: demand is thin, dispersed and invisible. Traditional origin-destination surveys sample daytime travel and cost months; between 1 and 5 a.m. the city had essentially no data on where people actually were, so night routes were guesses — and empty buses reinforced the guess that the demand didn't exist.
Seoul's data and statistics division solved the measurement problem with a public-private partnership: Korea Telecom shared anonymized call records — over 3 billion logs from a single month — plus over 5 million taxi journeys captured through the T-money fare system. The city divided Seoul into 1,250 cell units of one-kilometer radius and mapped the late-night floating population, using mobile locations as a proxy for where night travelers started and billing addresses as the destination end.
what everyone would do
Commission an origin-destination survey and pilot a route or two — months of sampling daytime behavior that misses the midnight population entirely, then empty buses that 'prove' the demand was never there.
what they saw
Nobody could survey the midnight city, but every traveler carried a phone that reported in. Treat the call network as a census of the night: the demand map already existed in the telecom's logs, waiting to be read.
the move
The analysis replaced survey and intuition with an observed demand map: routes connected the cells with the highest volumes of late-night presence, optimized through a GIS design system rather than the conventional process. The first two Owl Bus routes launched in 2013; a satisfaction survey scoring 82, with 88.4 percent of 1,000 respondents asking for more, justified expansion to nine routes from September 2014.
why it works
Call records passively observe what surveys cannot: presence at 1 a.m., sampled at city scale and full population rather than questionnaire panels. The one-kilometer cell grid matches the resolution at which bus stops actually matter, and taxi flows triangulate the trips people pay to avoid. Because the data covers the whole city at once, weak cells are as visible as strong ones, so routes connect observed demand peaks instead of loud constituencies — and the fare-savings and safety outcomes then justify expansion on evidence, not advocacy.
the payoff
By December 2016: 7,900 passengers a night, ~2.3 million fewer car trips a year, ~$13M in taxi-fare savings, +11% women's nighttime activity
where it breaks
Call records are a proxy, not a ticket survey: they show where phones sleep and wake, not precise destinations, and smartphone-era data skews away from the elderly and poor who remain invisible. Privacy compliance is a hard gate — Seoul stripped personal identifiers and needed the carrier's partnership. And the method designs the network once; it doesn't fund it, so thin night routes still bleed money if fares and frequency aren't tuned to the observed densities.
what came after
The Owl Bus became the canonical case for telecom-data-driven transit planning, studied and copied by cities worldwide as the proof that exhaust data can replace surveys entirely.
references
- [1]Using Big Data to Design Night Bus RoutesAsian Development Bank (Development Asia), 2020development.asia
- [2]Mobile Phone Data for Optimizing Bus Routes in KoreaData Collaboratives (The GovLab, NYU), 2018datacollaboratives.org