The Conundrum - Reality of Insurance & AI
In case you missed it, the NAIC recently announced a data call for homeowner insurance data. There is no doubt many interesting and somewhat actionable insights will be refined from this data.
But here is the unfortunate reality: patterns and trends at the industry level can be radically different than the situation your own company faces.
Around 20 years ago, I first witnessed the power of using technology to see across an entire population of claim notes for a single company. Seriously, I was blown away by the ability to see so much information so clearly.
Just like looking at any insights derived at a high level, the euphoria quickly dissipates when you double click a few things and start reviewing the details and the circumstances.
The specific facts and circumstances of individual claims or accounts are seemingly immeasurable which directly erode the confidence of the insights displaying at a high level.
The conundrum of always "need more data" versus "more data equals more nuance" has always existed in insurance. It always will.
And, here comes "AI".
For this post, let's focus on the Generative AI wave sweeping World.
The low cost of compute and storage have amplified the inherent reality that if a gen AI model is given enough data, it can figure out most anything. Especially when most people no longer worry about the degree of accuracy.
Thus, it's all about the data.
On that note, every insurance executive has the same problem. The head of AI is correct in declaring that AI can solve most every problem - assuming it has enough data.
At a high level, a lot of data is available. At a company level, the data is much more nuanced.
Yes, I fully understand that AI can and will be trained. The key questions that need to be asked: how long will it take? how much money will it cost? how will errors be identified and corrected? how much will the errors cost?
And many more questions could be asked.
Just like the NAIC is going to serve up some nuggets from the HO data pull, so will most any model served up by tech companies and consultants.
However, the execs handling claims, underwriting, & pricing won't survive long managing decisions from the perspective of one-time, high level insights.
On the other hand, having a consistent, reliable-yet-dynamic stream of measurements around key business activities will produce robust results that outperform the competition.
Don't be fooled by the conundrum: Focus. Measure. Adjust. Repeat.