Meet the insurtech: Tensorflight

Typically, rating the quality of building roofs, whether residential or commercial, has relied on information about their age and materials, or at best a top-down view captured aerially or by a satellite.

This may not always yield the most accurate result for roof replacement or other building replacement costs that insurers have to cover. Tensorflight, founded in 2016, started out using drone technology to assess the value of agricultural land and crops for insurers. The company then found a way to get more accurate roofing information that led it to pivot its entire business in 2017.

Natalie Cass

“There’s a certain subset of data you can get out of a top-down view,” says Jacob Grob, chief revenue officer of Tensorflight. “That’s why everyone does roof condition scores and defensible space scores, working in those types of attributes. … We’re different because we’re looking at the building from the side. We can start extracting the construction type.”

The New York-based company has a London office and engineering operations in Poland. Tensorflight made the switch from drone technology to satellite imagery and changed the subject of its data collection, according to Robert Kozikowski, CEO. “We needed to focus on the issue that seemed the most promising, the most valuable demographic and the most exciting to solve,” he says.

Tensorflight uses views of buildings taken from the side (as demonstrated on the company’s homepage), applying computer vision techniques to recognize and detect objects and scenes, and AI to draw conclusions based on computer vision data combined with building data. This gives underwriters more data points to work with, including roof condition, roof geometry, the type of building facade, building footprint, year of construction, building height, occupancy information and type of construction, according to Grob.

“The end goal is an accurate replacement cost for buildings,” he says. “With just an address, if we have the imagery from that side view, and from the top view, we can extract all the data necessary to feed into our replacement cost calculator and give an accurate replacement cost.”

Tensorflight began its services with images and data about commercial buildings, but has recently added residential buildings to its platform, to address homeowners insurance, according to Kozikowski. For residential, he says, “We highlight attributes that can be extracted from the facade, like replacement cost, construction type and square footage.”

Residential properties are simpler to capture and process than commercial ones, Grob observes. 

“Now we’re applying everything we learned to the simpler problem,” he says. “We’re taking that same technology and extracting data out of single family residences and individual properties.”