Boston needed a team of 10 working overtime for two months to build bus schedules each year — and the result was still inefficient

Before MIT researchers intervened, Boston Public Schools (BPS) required a team of 10 transportation planners working overtime for approximately two months every summer to construct school bus schedules for the coming year. Even using the best commercially available bus routing software, the output was invariably a set of inefficient routes with unnecessary costs and suboptimal student ride times. The district operated around 650 buses. No one on the team could tell the superintendent what would happen if a school's start time shifted by 30 minutes — the system was too complex for scenario modeling. The inability to model scenarios is what makes this problem so damaging. School start times affect academic performance (adolescents perform better with later starts), after-school childcare needs, teacher schedules, athletic programs, and parent work commutes. The American Academy of Pediatrics has recommended that middle and high schools start no earlier than 8:30 AM since 2014. But districts cannot shift bell times without understanding how it affects bus routing, and bus routing is an NP-hard combinatorial optimization problem that commercial software handles poorly at scale. The result is policy paralysis: districts know they should change start times but cannot model the transportation consequences, so nothing changes for decades. Boston had not reformed its school start times in 30 years before MIT's work. MIT's biobjective routing decomposition (BiRD) algorithm ultimately took 50 buses off the road and saved BPS $5 million annually without increasing average student ride times. The Boston School Committee unanimously approved the first start-time reform in 30 years in December 2017. But this required a multi-year collaboration with MIT's Operations Research Center — resources that 99% of school districts do not have. The commercial routing software market (Transfinder, Edulog, BusPlanner) has not incorporated this level of optimization. The structural barrier is that routing software vendors sell to individual districts and optimize for usability and customer support, not for cutting-edge combinatorial optimization research. The academic algorithms exist but remain trapped in papers and PhD theses, inaccessible to the transportation directors who need them.