I imagine that being an insurance CIO and a venture capitalist have at least this in common: every third conversation ends up being about the opportunities and challenges of AI. I’m a venture capitalist and I can tell you there’s a lot of noise to pick through but also a lot of terrific signal. One of the areas of AI that is truly living up to its promise, in my opinion, is computer vision. Both the data sources and capabilities have grown in orders of magnitude over the last few years. This opens up huge opportunities for many businesses including insurance. If you add these to a more general view of geospatial capability, the possibilities seem endless.
Let’s start with sources of data: new satellite constellations have made imagery incredibly abundant and affordable. Planet Labs alone takes a picture of every piece of the Earth’s landmass every day, and there are new cubesat competitors launching every few months. In addition to cubesats, there are currently more than 150 larger Earth observation satellites in orbit. On top of that we have aerial imagery from both manned aircraft and drones and new imagery technologies being brought to market. For example, we made an investment in a company called Urban Sky. They build and launch autonomous high-altitude balloons that can photograph large areas with very high resolution, producing a quality on par with aerial imagery but on a bigger scale and at a fraction of the cost. And of course, there’s ground-level imagery from professional setups, to Google Street View, to the 3.7 billion smart-phone cameras that people take with them every day. In addition to these new sensors, the trends in increased bandwidth, the broad availability of affordable cloud computing and a slew of tools to deal with very large data sets also come together to make computer vision both possible and a critical competitive tool.
But as CIOs everywhere know, having a massive pile of data does not create value on its own. Companies need to be able to move, process, and clean the data. Imagery needs to be normalized and stitched together, then joined with first party data and non-sensor sources. All of this needs to integrate with or displace existing legacy systems and processes. The tools are still hard to use and skilled data engineers and scientists who understand them are a precious and extremely constrained resource. Only with all of these pieces and people in place can you start talking about useful applications that create business outcomes.
For many industries such as Insurance, harnessing these technologies isn’t just a nice to have but a requirement for remaining competitive
Despite these challenges, I would argue that the effort and investment is not only worth it but imperative. But you don’t have to do everything on your own. As startup investors, we look for companies that can serve new markets like insurance. They are the enablers of this transition and from a vc perspective, the picks and shovels of this particular gold rush.
Descartes Labs, for example, has built a platform to manage and process massive amounts of geospatial data and the models to glean useful insight from them. The company works with large insurance companies to simplify the development and management of property characteristics at scale and to perform modeling and risk assessment before and after natural disasters. By leveraging their customizable geospatial analytics platform, insurance CIOs can control their own data processing and make geospatial information a core competency.
These property characteristics can have all manner of applications in addition to risk assessment, including claims adjustment and application automation. Imagine pre-filling property attributes before a customer even begins an application. Roof and pool footprints, property conditions, obstructions such as trees or utility infrastructure and many more of these attributes can be measured without visiting a physical location.
Individual property attributes are one kind of useful computer-vision enabled data: floodplains, wildfire areas, changes to neighborhoods over time, foot traffic...everything matters. Time series data can lead to all kinds of important insights: is a neighborhood deferring more maintenance than before? Are factories running longer hours than average? Is climate fundamentally changing the flood or fire risk in a certain area?
This gets even more powerful when combined with insurer’s proprietary data and other sensors such as IoT and GPS locations from smartphones. More insights lead to better predictions and better customer experiences and new products. For example, parametric insurance depends on having an accurate picture of what happened in a given place at a given point in time. Some customers can’t afford to wait for an adjuster to start funding a response to, say, a catastrophic windstorm or flood.
It’s also exciting to think about mitigating risk, such as identifying trees and brush near power-lines which can be cleared before they create a catastrophic wildfire. (I live in California where this is a big issue.) Here’s another example of risk mitigation: our portfolio company Fantasmo creates inexpensive, very high resolution 3D city maps and a technology to geo-locate, also using vision, down to 10 centimeter resolution. They’ve been working with micro-mobility companies and municipalities to keep scooters off of sidewalks and other places where they pose a hazard to pedestrians.
Computer vision continues to get faster and cheaper, allowing real-time site monitoring and deeper analysis. The pace of development in computer vision creates massive opportunities and significant challenges. For many industries such as Insurance, harnessing these technologies isn’t just a nice to have but a requirement for remaining competitive. Fortunately, startups are seeing this too and bringing solutions that would be hard and maybe unnecessary to build on your own.
