Fujitsu, Tokyu Construction and Kitano Construction have started a field trial of an artificial intelligence system designed to identify omissions and schedule risks in construction projects before they become costly problems.
The trial runs from August 3 to December 25 at Fujitsu Technology Park in Kawasaki, Japan, according to a joint announcement. The companies are testing whether generative AI can analyze project records and flag potential delays or missing tasks one to two months in advance.
The system combines fragmented project records
Construction teams typically manage information across schedules, daily reports, work procedures, inspection records, applications and regulatory documents. The project aims to bring those records together so an AI system can compare what is planned with what has been completed and what rules still apply.
According to the companies, the technology will look for inconsistencies and omissions across those sources. It is also intended to present the basis for an alert, giving project managers evidence they can review rather than only producing an unexplained risk score.
The trial is testing how the system performs when information arrives in different document formats. That is a practical issue for construction technology because contractors, subcontractors and project owners often use different templates and levels of digitization, even within the same development.
Fujitsu is providing the AI and data-processing technology, while Tokyu Construction and Kitano Construction are contributing construction-management knowledge and project data. The three companies said the work would assess both the accuracy of the findings and whether the alerts arrive early enough to support decisions on site.
Earlier warnings could reduce rework
A missed approval, inspection or material dependency can affect multiple stages of a construction schedule. Detecting such issues before they reach the site could help managers adjust sequencing, allocate labor or obtain missing documentation with less disruption.
The partners are specifically targeting risks that may emerge one or two months ahead. That horizon is intended to move the tool beyond daily reporting and toward proactive planning. The announcement does not disclose accuracy targets, project savings or a timetable for commercial availability, and those points will need evidence from the completed trial.
The initiative reflects a broader push to apply AI to the operational data generated by physical projects. Construction has been a difficult environment for general-purpose software because conditions change frequently and much of the relevant information is held in drawings, images, free-form reports and specialist systems.
Future work may add richer site data
The companies said they plan to explore the use of building information modeling data, photographs and three-dimensional point clouds. Those sources could give the system a more direct view of physical progress, although connecting them reliably with schedules and contract records remains a technical and organizational challenge.
Any production deployment would also need clear rules for data access, responsibility and human review. An AI warning may help a project manager investigate a risk, but it does not replace engineering judgment or the contractual processes used to approve construction work.
For Fujitsu, the trial is a concrete enterprise AI deployment in its home market. For the two builders, it offers a way to test whether generative AI can improve project controls without requiring every partner to adopt an identical document format from the outset.
The partners expect to evaluate the results after the December conclusion.
Featured image: taro ohtani on Unsplash
Hong Kong’s Varadise secures $3M in Series A funding round led by Betatron