August 1, 2015
How Google Thinks About Insurance
by Paul Carroll
For those of us wondering what Google plans to do in insurance — isn’t that all of us? — it’s worth looking at the company’s Project Sunroof. The project uses exceptionally sophisticated mapping data to determine which homeowners would most benefit from solar panels and, in the process, may provide some insight into how Google is approaching insurance.
To me, there are two key aspects of Project Sunroof. The first is that Google is taking a bottom-up approach that could inform a lot of decisions about insurance (while insurers traditionally go top-down). The second is that Google is being unusually smart about combining layers of information — some proprietary, some in the public domain; some new, some long-available — to produce what my frequent co-author Chunka Mui has, with a little help from me, labeled “emergent knowledge.” (“Big data” is the term commonly used, but data isn’t very interesting, while knowledge is. And the size of the database doesn’t matter. What matters is using developments in technology to look in the right places to find the right data to answer the right questions so that revelations emerge.)
Top-down vs. bottom-up
Insurers typically start with pools of risk. They’re getting much more sophisticated about subdividing those pools into ever smaller groups, but the thinking is still along the lines of “drivers without moving violations who travel 11,000 to 12,000 miles a year in generally suburban conditions.” Insurers will keep getting more and more specific and produce more and smaller pools but are still going from the top down.
Now look at Project Sunroof. Google is modeling the world in three dimensions and using that model to generate information house by house based on totally personalized criteria: on the square footage on the roof that would be available for solar panels, on the amount of cloud cover that is expected to obscure the sun above that house, on the effectiveness of the sunlight that will hit the roof (incorporating calculations based on temperature and on how the angle of the sun changes each day) and on any shade that would be cast on those panels from other structures. Although the article doesn’t say so, I assume Google calculates potential savings on solar based on the rates of each local utility. In any case, there are no pools in sight for Google — unless you want it to tell you about those in the backyards.
That same model of the world could be the basis for a house-by-house, car-by-car, person-by-person approach to insurance for Google. And, if this approach works, Google will gain the sort of information advantage that has proved to be almost impossible to overcome. Even the largest insurers would have a hard time spending the money that Google has to map the U.S. by having cars drive every single street to take pictures and collect data, by making a series of acquisitions of data providers and by employing a small army of people to manually fix errors and update maps — and Google would still have a years-long head start in developing its model of the world. Microsoft has thrown billions of dollars at search engines, but even Microsoft couldn’t overcome the fact that Google’s dominant share meant it was always learning and improving faster than Microsoft’s Bing. Apple’s map services were ridiculed, by comparison with Google’s, when Apple launched them in September 2012. Apple is now at least in Google’s ballpark on mapping, but no insurer can come close to Apple’s resources — a $646 billion market valuation and $202.8 billion of cash in the bank. That’s “billion,” with a “b.”
Google obviously begins with a huge asset because of its prescient decision years ago to map the entire U.S. and because of the recent work that has made that map 3D. But Google is also taking data wherever it can get it.
I know from some work I did at the Department of Energy in 2010 that national maps of sunlight have been available for years, and they have surely become far more detailed as the interest in solar power has spread, so I assume Google didn’t have to generate those maps on its own. Temperature maps are also in the public domain. (Especially high or low temperatures degrade the performance of solar panels.) Those maps will become increasingly granular as they incorporate data from smartphones and other widely used devices that can act as sensors — temperature will no longer be what the weather station reports from the Detroit airport; temperature will be known house by house. Overhead photos from satellites and, in some cases, drones are widely available, so Google can use those to check square footage of roofs, to see which direction the solar panels would point and so on. Google can collect information on rates from state utility commissions, where utilities have to make regular filings.
It’s easy to imagine Google layering similar types of information onto its map of the world for insurance purposes. In response to the federal Data.gov initiative, governments at all levels are making more information available digitally, so Google could incorporate lots of data about where and when accidents occur, where break-ins happen, where and when muggings occur and so on.
Google could incorporate private work that is taking a 3D approach to flood risk (whether your house is three feet higher or lower than the average in a neighborhood can make all the difference) and is being much more discriminating about earthquake risk. Google could add information, from public or private sources, on the age of homes, type of pipes used, appliances, etc. to flesh out its understanding of the risks in homes.
And, of course, Google will have lots of very precise information of its own to add to its model of the world, based on, for instance, what it knows about where your smartphone is and can infer about where you park your car, where and when you drive, etc.
Once you take all this information and map it to such a precise model, there will surely be some non-obvious and highly valuable insights.
WWGD: What Will Google Do?
Looking at Project Sunroof still doesn’t say a lot about how Google will attack insurance. Will it just sell increasingly targeted and valuable ads? Will it sell leads? Will it become a broker? Will it do more?
But I think it’s safe to say that, whatever Google does, its starting point will the most sophisticated model of the world — and that model will always be improving.