AMERICA’S deadliest building fire for more than a decade struck Oakland, California, on December 2nd 2016, killing 36 people attending a dance party in a warehouse that had become a cluttered artist collective. The disaster highlights an open secret: many cities lack resources to inspect for fire risk all the structures that they should. Even though the Oakland building had no fire sprinklers and at least ten people lived there illegally, no inspector had visited in about 30 years. How might cities make better use of the inspectors they do have?
A handful of American cities have begun to seek help from a new type of analytics software. By crunching diverse data collected by government bodies and utilities, the software works out which buildings are most likely to catch fire and should therefore be inspected first. Plenty of factors play a role. Older, wooden buildings, unsurprisingly, pose more risk, as do those close to past fires and leaks of gas or oil. Poverty also pushes up fire risk, especially if lots of children, who may be attracted to mischief, live nearby. More telling are unpaid taxes, foreclosure proceedings and recorded complaints of mould, rats, crumbling plaster, accumulating rubbish, and domestic fights, all of which hint at property neglect. A building’s fire risk also increases the further it is from its owner’s residence.
As for how predictive such data can be, consider Firebird, a model developed at Carnegie Mellon University for Pittsburgh’s fire department. Of the city’s 20,636 non-residential structures, Firebird ranked 103 as highest risk. In the first five weeks after Pittsburgh began using the software earlier this year, fire incidents had occurred in 58% of them. In the seven years since New York City switched on a predictive model called FireCast, which was developed for about $1m with help from IBM, inspectors have quintupled the number of “pyrogenic” conditions identified, says Mike Flowers, New York’s chief analytics officer at the time. Each year since then the number of fires has fallen.
But many cities are unlikely to benefit from such software. Firefighter unions have put up fierce resistance, notes Mr Flowers, as the software changes firefighters’ duties to include tasks that were not allowed for in the established collective-bargaining framework. Mr Flowers also notes that successful software will reduce job opportunities for firefighters. Also, astonishingly, some cities prohibit prioritising certain inspections as this requires de-prioritising others, a common bureaucratic no-no, says Matthew Hinds-Aldrich of the National Fire Protection Association. Some privacy advocates fear future models will ingest lifestyle data and disproportionately flag buildings with, say, smokers and hoarders. Predictive software designed at Harvard that Portland, Oregon, will soon begin using will not do that just yet. Perhaps more importantly, the city’s fire chief noticed that buildings marked as being the biggest risks are clustered in areas lacking good schools, public transport, health care and food options. Healthier, happier people start fewer fires, he concluded. He now lobbies officials to reduce Portland’s pockets of blight.