Nearmap builds AI moat from 20 years of aerial data

Nearmap is rolling its HyperCamera 3 system across its aerial capture operations while using AI to extract property intelligence from long-running imagery archives, building its strategy around proprietary imaging, historical data and large-scale processing.

"We have very novel cameras. We've started flying our HyperCamera 3 system in the last year or two, and we're swapping our whole fleet to that," said Michael Bewley, Senior Vice President, AI & Computer Vision, Nearmap.

The technology supports Nearmap's broader shift from supplying aerial imagery to providing information about individual properties. The company operates solely in the B2B market, serving sectors including property insurance, construction, architecture, local government and solar installation.

Customers range from one- or two-person businesses to large organisations. Some begin with a proof of concept before expanding their use of the platform, accessing Nearmap imagery, AI-derived data or both.

Property focus

"We've focused on property intelligence, which means most of our customers don't actually care about latitudes and longitudes and things. They say, 'I've got this address. Tell me what's going on there. Tell me what's been going on there, and tell me what might happen in the future.' That's a GIS problem for us, but not for our customers," added Bewley.

That property-level focus shapes where Nearmap captures imagery and how often it returns to the same locations. The company concentrates primarily on urban residential and commercial properties rather than trying to cover land mass uniformly.

Nearmap runs a fixed capture programme across the four countries in which it operates. In Australia, some areas are flown up to six times a year, giving customers repeated views of how cities and individual properties are changing.

The economics become less attractive when aircraft must travel to farms, mine sites or other low-density locations for a relatively small number of customers. Bewley compared the approach with mobile network coverage, where reaching a high proportion of the population differs from covering the same proportion of a country's land mass.

Applications include property inspection and maintenance, solar installation planning, construction and landscaping. Imagery can also reduce how often workers need to climb onto roofs, while owners of large building portfolios can use the data to monitor assets.

"We've got a pretty long tail of customers, particularly in Australia. It's all businesses, so it's B2B. But some of those are pretty small, one- or two-person operations, and some of them start small with a POC and then grow. Then we've got some of the biggest. A lot of the names you would expect to be using this kind of stuff in Australia are using this kind of stuff, whether it's imagery or AI or both. So, a very broad base here," said Bewley.

Camera efficiency

Nearmap continues to rely on crewed survey aircraft for image capture. It does not own the planes, instead working through long-term partnerships with survey aircraft operators, while developing its imaging hardware separately.

The equipment is mounted through an opening in the floor of a light aircraft and must compensate for vibration, temperature changes and movement while capturing detailed imagery. Nearmap's system is bespoke, extending beyond the sensor to the components required to maintain image quality under those conditions.

"It doesn't look like a camera anymore. If you can imagine a little light plane, you cut a hole in the floor about that big, and then you're bolting this whole apparatus with spinning bits and mirrors and lenses and things going on in it. If you've ever looked out a light plane window, it's vibrating, it's cold and it's windy, and you're trying to capture those really sharp pixels. So it's not just about a good sensor. It's about how you deal with motion and vibration and temperature variation. Every part of our camera system is bespoke," added Bewley.

The aim is to combine image quality with capture efficiency. Flying higher while retaining sufficient detail can reduce the number of passes an aircraft must make over an area, lowering the cost of collecting imagery at scale.

Nearmap has considered other aerial platforms, including gliders, balloons and blimps. Its current model remains centred on conventional light aircraft because they allow crews to follow prescribed routes while accounting for weather and air traffic restrictions.

"Turns out for us, the sheer efficiency of getting pilots and little white planes is by far the simplest thing. We don't own the planes. We have long-term partnerships with those survey planes. If there's other technology that comes up, we'd assess it," said Bewley.

Historical AI

Nearmap's AI work centres on custom deep-learning models that process imagery and generate property attributes at scale. That allows the company to apply AI not only to newly collected images but also to historical datasets.

It has processed historical US imagery for a roof product and large volumes of Australian imagery. Nearmap has about 20 years of imagery in Australia, adding a time dimension to individual property observations.

"We've got very custom deep-learning models that can produce all of those attributes in a single pass at a cost ratio that makes it viable. We can run full history. So we ran full history in the U.S. for our roof product. We've run a whole lot of massive history in Australia, and that allows you to access time. The moat becomes complexity with AI because you're not looking at one image. You're saying, 'I've got 20 years of imagery in Australia, and I want to know not just the building, but the things nearby.' Maybe I start to weave in third-party data sources, and that forms this rich tapestry of owned data that we have, weaving in third-party data," added Bewley.

Nearmap has begun incorporating more third-party information as it moves further into property intelligence, particularly for property insurance. It previously concentrated more heavily on its own datasets but is adding external sources in response to customer demand for richer property information.

Its aerial imagery and AI remain the core of the offering. The combination gives Nearmap control over both the original capture process and the software used to derive information from the imagery, while direct customer relationships help shape the attributes it develops.

Subscription scale

Nearmap's capture programme is tied to a subscription model that differs from traditional commissioned aerial surveys. Rather than waiting for an individual customer to fund a flight over a specific area, Nearmap sets a capture plan, repeatedly flies those locations and gives subscribers access to the resulting data.

Camera efficiency is central to that model because the cost of collecting imagery can be spread across customers interested in the same urban areas. Subscribers also gain access to repeated captures rather than receiving a one-off set of files after a commissioned flight.

"There's been aerial imagery providers for a long time. The big change that Nearmap precipitated was because of the camera efficiency. We could fly subscription. Everyone else was taking some local government saying, 'Can you fly my area once this year and I'll pay you this much? Can you give me the files on a hard drive?' We said, 'No, we're going to fly a plan, and you're going to subscribe to that plan.' That was a big shift in how the aerial imaging business works," said Bewley.

The aerial imagery market remains fragmented, with numerous smaller operators. Nearmap continues to build camera systems for its own programme as it replaces existing equipment with HyperCamera 3, but Bewley did not outline plans to acquire other providers.

"I won't speculate on future strategy. That's not my role. But we're still growing very rapidly," added Bewley.