Saitama City's AI Matching Algorithm Cuts Nursery-Place Allocation From Days to Seconds

Matching thousands of children to licensed nursery places each year, while respecting families' ranked preferences, priority scores, and complex sibling rules (such as requiring siblings to attend the same facility or start on the same date), had previously taken Saitama City's staff several days of manual work each admissions cycle.

In 2017, Fujitsu Laboratories and Kyushu University's Institute of Mathematics for Industry developed a game-theory-based matching algorithm — in the same family as the deferred-acceptance mechanisms used for US medical residency placement — to automate the process, and Saitama City announced its adoption for the 2019 admissions cycle.

In its first production run, the system matched 7,959 children to 311 nursery facilities across the city, honouring eight sibling-preference rules.

The vendor's own evaluation on anonymised historical data completed the computation in a matter of seconds, cutting roughly a week off the time it took to notify parents of results. Other municipalities followed: Minato Ward in Tokyo later reported cutting a reported 500 person-hours of manual matching work down to about five minutes of computation, and Japan's Ministry of Internal Affairs and Communications listed more than 30 municipalities trialling similar systems by fiscal year 2018.

The headline speed figures come from a vendor evaluation conducted before deployment, not an independent post-hoc audit, and no published study has examined whether the resulting matches are fair or resistant to strategic manipulation by applicants. The system is scoped to licensed childcare (early-childhood) admissions rather than K-12 school administration, and it relies on combinatorial optimisation rather than a learned/trained model.