October 31, 2019 8 min read

Geospatial AI for Underwriting and Rating: Build or Buy?

Written By CAPE Analytics

By Bobby Davis, Senior Product Manager 

As AI technologies continue to mature and proliferate, the insurance industry is starting to latch on to their business benefits. Forward-leaning carriers have taken steps to test and operationalize AI across nearly every underwriting workflow. As this shift occurs, insurers are left with a pivotal choice: should they buy an AI solution or build their own?

A “build” approach promises tantalizing benefits. Carriers who choose this path would experience the freedom of a blank canvas wherein they can design the exact system for their needs, with fewer external constraints imposed by solution providers. In other words, this is the innovation team’s dream scenario. In the best case, an in-house capability becomes a competitive edge that enables the carrier to consistently underwrite policies more effectively and profitably than others.

Despite these seeming advantages, building an in-house solution presents many disadvantages and comes with steep opportunity costs. 

First, an in-house AI capability requires a significant and sustained R&D commitment. Although tools such as TensorFlow serve as a useful springboard for AI product development, industry-specific requirements often make it necessary to build an extensive degree of customization on top of the out-of-the-box functionality. For example, the state of Florida mandates a homeowner policy credit for wind mitigation if 90% or more of the roof system perimeter consists of a hip shape. A generic roof detection AI model based on aerial imagery is unlikely to classify roof shape, let alone classify hip roofs in accordance with the required definition for the Florida market. Building a highly accurate, tailored roof shape model entails a large, multi-quarter undertaking involving model architecture design, label taxonomy definition, ground truth data collection, model training, and data pipeline integration. This timeline assumes the availability of experienced technical staff who have deployed and scaled AI products in the past. Similar levels of effort should be anticipated for other AI product development initiatives.

Second, and perhaps more importantly, investing in an in-house AI capability doesn’t stop with the R&D phase; it necessitates constant caring and feeding throughout the entire product lifecycle to support ongoing operations and maintenance. IT teams that claim victory after deploying an AI solution will instead find that the job has just begun – among other maintenance responsibilities, they will be on the hook to periodically test and validate AI model performance, properly version control their AI models, increase system throughput, and meet stringent data pipeline uptime requirements, all while continuing to build out additional useful features. In the end, it will turn into a multi-year effort that must go beyond innovation and IT teams in order to be successful; if the initiative is not core to a carrier’s DNA, it will most likely fail. 

Given the tough trade-offs outlined above, it is prudent for carriers to explore external solution providers as a viable alternative to building an in-house AI product development capability.

At Cape, we invest (and re-invest) considerable R&D towards developing purpose-built data products for the insurance industry. Since each of our products serve many insurance customers, we are effectively amortizing our product development investments — projects with a small payoff for any single carrier become no-brainers if dozens of customers stand to benefit. With every new customer, Cape’s economies of scale increase, resulting in even more products delivered faster. Meanwhile, we always maintain our same laser focus of tailoring our AI products for insurance workflows, thus maximizing our customers’ return on their investment in Cape. 

Similarly, Cape’s continued existence as a company is predicated on our ability to keep customers happy throughout day-to-day operations. Our engineering staff includes a sizable contingent who live and breathe our data platform, continuously making our AI-derived data pipeline more reliable and accessible. The result of this dedication is an enterprise-grade AI product that customers can operationalize on day one with minimal integration. There is no need to wait quarters  (if not years) for an internal AI solution to be designed, developed, and deployed.

For carriers, a misguided “build” versus “buy” decision can have a deleterious business impact. A carrier choosing to develop its own bespoke policy administration system ten years ago, instead of relying on up-and-coming third-party platforms such as Guidewire or Duck Creek, would sink enormous resources into building and maintaining their internal system. Now, the buy option clearly outweighs the build, though it may not have been as obvious a decade ago. As migrations from internal system to 3rd party provider takes place, it will consume additional time, resources, lost productivity, and organizational attention.

Similar “build” versus “buy” shifts regularly appear within the enterprise software industry. Auth0 is one example in the unsexy user identity and access management space; where developing and maintaining an in-house authentication system is viewed as a chore by internal IT teams, Auth0 presents an opportunity to eliminate this distraction with a far superior product. SendGrid has done the same thing for system-generated emails; why even consider building your own email infrastructure if a better off-the-shelf solution already exists? Auth0 and SendGrid are very different companies but share an important trait in that their products are built around a specific set of enterprise capabilities that, while essential to company operations, are not core to their customers’ competencies. 

Ultimately, AI build versus buy decisions for insurance carriers should be based on top-level company strategy and the core competencies needed to support this direction. A carrier that wishes to augment their existing underwriting core competency with more accurate, AI-derived data may very well land on a different end of the build vs buy spectrum than, say, a hypothetical carrier that makes the drastic decision to immediately automate all of their underwriting and rating functions through an AI platform. A carrier must ask themselves whether their AI strategy affords a sufficiently compelling, unique, and sustainable competitive advantage to justify the opportunity costs that come with building a custom solution instead of purchasing a commercial product. Especially one offered by a solution provider that specializes in solving that same problem, solves it exceedingly well, and continues to rapidly add new capabilities based on the collective experiences, needs, and learnings of all of its carrier customers.