The AI Reality for Most Specialty Carriers  

^{By Nick Mair}

Our recent attendance at the Insurtech America Symposium gave us an interesting look at the current, hard realities on the ground regarding AI for carriers as at Q1, 2026.   

Here’s my 2-minute take on the core themes emerging:  

1. Real AI usage is still limited.

Especially across specialty and E&S lines. There’s plenty of intent, but given we’re still so early in the cycle, there’s also too much “wait and see”.  Post-pandemic and post the great Insurtech wave of 2016-19, many carrier-side innovation teams were wound down. This means the burden of experimentation is now falling on the few carriers left who are still willing to take risk to experiment in order to prove what works for the rest. In a market where the average sales cycle of new technology can be 12-18 months for proven products, adoption will remain slow for a while yet until competitive advantage is established and tangible ROI is demonstrated.

2. Everyone has rebranded as an “AI company”.

Legacy vendors included. Which makes it harder, not easier, for prospective buyers to decide which software to back. Most native AI, when applied to specialty, is largely unproven, lacking access to the private, carrier side data it needs to learn and improve. In parallel, many legacy vendors are doing “add on” AI badly.  The smartest conversations I heard cut through the noise to ask the practical, back to basics questions i.e.:

• What are the longstanding, foundational problems we’re still trying to solve? 
• Which vendors are already making good inroads in solving them?  
• Which technologies already have proven adoption and multiple positive references? 

3. Trust in AI output remains a problem

Particularly in pricing. AI-led pricing has been around for a while now, acting as a useful check and balance to manual pricing, but ask any CUO or CFO and clear, provable P&L impact is still hard to nail down. Trust in AI involves multiple layers of transparency, but it starts by having daily confidence in the baseline data used by models. Which leads me to my final point…

4. It really does come back to the quality of your core, operational data! 

You won’t be surprised to hear me say this (again) but amongst it all, it was great to hear this point raised multiple times from the stage.

Kimberley Harris Ferrante of Gartner finished her excellent panel (Underwriting Technology at a Crossroads – Decision support or Digital Noise?) by cautioning  “Having clean, trustable data is still a problem our industry hasn’t solved….”. 

Meanwhile, Primerica’s Misty Sutton was more direct. “If your data isn’t cleaned up… AI is useless.”  

Wherever we are in the current AI cycle, this fundamental truth won’t change.  

In fact, when the current AI wave finally settles down to deliver measurable impact in BAU, consistent, clean and trustworthy data becomes more important than ever.  

To help you with your first step in ensuring clean data flows, whether it’s for efficient Underwriting Ops or to power your latest AI experiment, we are here to help!