The first question to ask when building technology is: what problem are we solving? The problem cannot be “we need to incorporate AI.” Better problem statements can be framed as “how might we?” How might we speed up cycle times? How might we reduce fraud? From there, the problem can be broken down into further parts. Once the problem has been scoped, it’s time to move to solutioning. And, while the solution is being built, AI may be one tool that can be applied to solve the particular problem.

AI transformation may be a mandate, but when it comes to implementation and leveraging it to its full potential, AI must be thought of as a tool that integrates into existing processes and systems to optimize them, rather than an end in and of itself. At Strala, we think about this from the bottom up.

Layer one: Data architecture

At its core, AI is a technology that makes predictive choices based on data. For the technology to be able to use data, that data must be available, clean, and consistently labeled. In insurance, that is often not the case. A single claim’s information may be spread across a policy system, a claims system, email threads, PDFs, photos, and adjuster notes, and each one may be formatted or labeled differently depending on the region, the line of business, or the person entering it. If an AI model is working from incomplete or inconsistent data, its outputs will be unreliable at best. In a regulated industry, it also matters that every decision can be traced back to the data behind it. Getting the data foundation right is not necessarily the obvious or exciting part of AI, but everything else depends on it.

Layer two: Systems

Most legacy companies have had to bolt several systems together as new technology and processes have become available. A common challenge with having all of these separate systems is that they can become siloed from each other. Some adjusters have shared that they have multiple logins to different systems they must use to get one claim done. Emails may be handled outside the system, which requires adjusters to download files and move them between systems. Fraud may need to be investigated on a separate platform. Estimates, too, may need to be made in a separate place. You can imagine that this contributes to slowdowns and errors in addition to being merely inconvenient.

Layer three: People

Technology is only as good as how much it is actually used. This means that people must adopt it. When people do not adopt technology, it is not typically because they don’t want to, but rather because they have not been set up to use it successfully. The technology must deliver on its promise, too, and actually solve problems that people face rather than creating new ones. It requires organizational systems that support adoption, in the sense that it must be integrated into the right places, and people must be trained.

A lot of insurance organizations today are focused on the people problem: is my team using AI? How can I make sure they use it? Where many fall short is on the supporting systems. Without systemic support that is built all the way down to the data architecture level, AI adoption is likely to fail (and we see this happening with many organizations that we talk to).