I spend a lot of my day in conversations with carriers, MGAs, and program administrators about the application of Artificial Intelligence (AI) in insurance operations. Whether that is how to underwrite a risk better, improving the business portfolio,utilizing LLMs for data analysis, applying agentic AI to improve operational efficiency, increasing employee capacity or accuracy, or automating complex administrative tasks that overburden daily workflows. Most of those conversations are still investigating what is possible, what is on someone’s roadmap, what a pilot might show six months from now. Useful, but often theoretical. I also hear a lot of conflicting information coming from the market on how to best utilize AI or the benefits that an insurer will actually achieve. That’s why the conversation Meredith Barnes-Cook hosted for Datos Insights on April 2nd was refreshing.
Peeyush Rai and I had the opportunity to join Sean Rand, CTO of Trium Property, to discuss real-world implementation of AI into an insurer’s underwriting operations and the impact they were seeing. Sean talked about what they are doing in production today and how it was affecting their employees. Not a concept, not a roadmap — a live agentic workflow touching one hundred percent of submissions and underwriter analysis, service requests, and claims details before a human even reads an email.
5% of carriers have achieved widespread AI deployment.
30-40% in-appetite submissions never get quoted.
— Meredith Barnes-Cook, Senior Principal, Datos Insights
Meredith set the stage with two statistics that echoed what I am hearing on a daily basis. Only about five percent of carriers have achieved anything close to widespread AI deployment, despite tremendous investment and desire. And 30-40 percent of in-appetite submissions never get quoted — not because the underwriter didn’t want the business, but because the deal was already done with someone else by the time the submission worked its way through the process.
The conversation with Meredith, Sean, and Peeyush clarified a few key learnings I have discovered, and I want to take a moment to reflect on our conversation and share a few critical insights from the webinar.
Sean’s framing on security regarding the application of AI was the most direct I have heard from the carrier side in a while:
"We could go out and pick up a SaaS product and stick it over the top of all of our data, all of our infrastructure if we wish to, and that would be a huge security concern to us.”
— Sean Rand, CTO, Trium Property
A lot of conversations around AI focus on getting employees to leverage LLMs to simplify their workloads, or buying new/updated solutions that have bolted on copilots. But what does that mean for your data? Do you know where it is going, or who has access to it? Trium Property built their architecture so they own the infrastructure, they manage the data. They invoke capabilities, like Weav.ai, through APIs and rigorously control data access. The AI comes to the data, not the other way around. It is a harder road than bolting on a SaaS product, but it is the road that holds up to the security standards our industry requires. Even when customers use our UI, we focus heavily on security and where the data is flowing. It’s why we maintain SOC II Type II for the safety of our customers. Security must come first.
Good data security demands good data architecture. Trium started with what Sean called a greenfield mindset — no legacy platform constraints, and a clear view of what they wanted to bring to market. They started with the problem they wanted to solve and what the solution looked like. Then figured out how to connect in the services and intelligence they needed to create success. Working with solutions who provide both API-first connectivity, integrations into core systems, easy to implement workbenches as needed, and deep strategic partnerships with implementation providers ensures the flexibility to get new technology into place quickly and efficiently. The velocity that unlocked for Sean was striking:
"We’ve done more probably in three months than I maybe have managed to do in three years prior."
— Sean Rand, CTO, Trium Property
His advice to any carrier CTO or chief underwriting officer starting this evaluation was direct and to the point:
"Think of the portfolio. Don’t think of the technology."
— Sean Rand, CTO, Trium Property
The health and growth of the portfolio should be the primary focus. Technology should enhance that focus — never distract from it and never become a vanity project or a checkbox in strategic priorities. That last thing insurers need is another stand-alone solution that doesn’t quite work and is never fully adopted.
More than eighty percent of my conversations with carriers involve existing workbenches, core systems, and investments the carrier wants to keep. That is simply the reality of our industry, and it should be embraced, not apologized for. Although Trium may have started as a rare greenfield case, Sean was clear on how he looks at technology adoption:
"We haven’t removed our technology footprint or reduced it… we wanna enhance it. We’re not gonna remove those partners."
— Sean Rand, CTO, Trium Property
The right architecture orchestrates across what is already working — cat models, rating engines, data providers, core policy systems — and adds the agentic layer underneath, systemically. Find the biggest bottleneck first (in almost every carrier, that is submission ingestion). Prove value against real data. Move to the next bottleneck. Keep going. On the ingestion side alone, we are seeing eight to ten times the productivity of the traditional workflow.
Real value is key. Meredith was direct when discussing early engagements with new solution providers:
"Start out with a proof of value, not a proof of concept. Use real submissions. Use your guidelines. And see if the output reflects how your best underwriters think."
— Meredith Barnes-Cook, Senior Principal, Datos Insights
Generic benchmarks and vendor promises do not tell you anything. What matters is whether the output maps to how your best people make decisions leading to greater value for the business — and that is the real power of an agentic system working behind the scenes.
Meredith’s advice on navigating the vendor landscape when it comes to AI implementations drives this home:
"Look past what the solution providers are calling things and get into asking about the problems they solve."
— Meredith Barnes-Cook, Senior Principal, Datos Insights
Walk any conference floor right now and everything is labeled with AI. That noise makes it genuinely hard to figure out what any of it does. When I sit down with an insurer, one of the most useful things I do may seem counterintuitive: I set the word “AI” aside for a minute. The question I like to start with is simpler: what business problem are you trying to solve, and what are the key use cases that are causing you pain and sleepless nights?
That framing tends to simplify the conversation, and it brings a level of pragmatism that ends up being the major theme in everything we do with our customers and partners. From there, the technical conversations become much easier.
Rethinking the workflow and utilizing Agentic-AI to initiate, analyze, recommend, score, and automate as much as possible frees underwriters, auditors, and adjusters to focus on what needs their expertise, act faster, and be more informed. AI-in-the-Loop designed to empower people is unlocking the massive productivity gains we are seeing.
In the end, it’s less about “Human-in-the-Loop” and more about “AI-in-the-Human-Loop”. It becomes critical to focus on how new technology can empower the talent you have. Meredith also discussed how the talent dimension is even more urgent than most AI conversations acknowledge:
"Experienced underwriters are becoming retirement eligible in significant numbers… we’re redefining what entry level looks like."
— Meredith Barnes-Cook, Senior Principal, Datos Insights
The underwriters who carry decades of institutional knowledge are heading toward retirement. New hires are being asked to perform tasks that previously required intermediate experience, but without the same hands-on foundation to rely on. Across the conversations I have had over the last 20 years, talent maintenance used to be a secondary topic — today it is very much a primary concern. Underwriting virtual assistants taking on a tranche of the renewal book, codifying the guidance that would otherwise live in five different senior underwriters’ heads: this is becoming practical work today, not a future-state slide.
And the reason it mattered was not purely technical. It comes down to one business question, the one Trium Property started their transformation with:
"How can we grow 5x, 10x, 20x with the same group of people?"
— Sean Rand, CTO, Trium Property
Neither Peeyush nor I expected the webinar to lean as hard into the human side of the AI debate as it did. But Sean and Meredith kept returning to it, and I think that is telling.
Sean was unusually direct about how dangerous it is to let job-replacement framing take hold inside a leadership team:
"If you take a junior underwriter and you think we could potentially do ninety percent of that junior underwriter’s job with AI and just have that simplistic approach — that’s usually scary, and that’s actually dangerous as a leadership group."
— Sean Rand, CTO, Trium Property
The framing that works at Trium is different, and it matches what I see with every insurer that is getting real traction today. AI supercharges people, it does not replace them. In specialty markets — hurricane, wind, flood, the kinds of exposures Trium writes — you depend on the underwriter’s judgment and experience. AI simply allows getting to that judgment faster, more consistent, transparent, and easier to repeat and audit.
Sean also made a point about product ownership that I want to underline. At Trium, the scorecards, knowledge graphs, rules, and versioning are owned by the underwriting and operations teams — not IT:
"The technology team will not be building all of this out when it comes to knowledge graphs, scorecards, rules, versioning… and therefore the underwriting team and ops team are leading us."
— Sean Rand, CTO, Trium Property
When underwriters own the decisioning logic, they are not waiting on IT to update a rule. That shift in ownership changes how fast things move — and, honestly, it changes how engaged the underwriters are, because they are shaping the system they use every day. When AI can take the mundane administrative work off their plate; the true talent stays firmly in the seat where it belongs.
A few things the webinar reinforced for me:
What I am thinking about differently after this conversation: how much of the real work is change management. The technology itself can work. The harder question is whether the organization is ready to use it. Meredith put it as simply as anyone could:
"Just like insurance is all about people, technology is all about people."
— Meredith Barnes-Cook, Senior Principal, Datos Insights
It’s ironic that Artificial Intelligence may actually help us become more human-focused in the end. I for one, am glad to be helping carriers, MGAs, MGUs, and Program Administrators explore the boundaries of how AI can make insurance better.
Peeyush and I are grateful to Meredith and the Datos Insights team for a conversation that stayed grounded. And to Sean for being willing to speak candidly about what is actually working in production — that kind of honesty is what moves our industry forward.
Watch the full recording of the Datos Insights webinar, “Rethinking Underwriting Architecture: When Workbenches Become Intelligence Platforms,” — worth the hour for anyone in underwriting technology or operations leadership.
------------------
Jeff Heine is a Founding Member and the Chief Revenue Officer of Weav.ai. This post reflects on the Datos Insights webinar “Rethinking Underwriting Architecture: When Workbenches Become Intelligence Platforms,” recorded April 2, 2026, featuring Sean Rand (CTO, Trium Property), Peeyush Rai (CEO & Founder, Weav.ai), and Meredith Barnes-Cook (Senior Principal, Datos Insights).