Is your AI underwriting tool making better decisions or just faster ones?

In this article:

• At the Insurtech America Symposium, four insurance executives had a candid conversation about a crucial gap in the industry: carriers are investing heavily in underwriting AI, but most still can't prove it's making them better at the actual job

• Most carriers are only running one AI strategy. The ones that pull ahead over the next five years will be running two simultaneously, and the difference between them is bigger than most leadership teams realize

• Underwriters are spending more than half their time on tasks that have nothing to do with underwriting. The panel gets specific about how to fix that and why getting it wrong means your best people never make it to the decisions that actually matter

• Bad data doesn't just slow down AI. It breaks it. The panel lays out a practical, no-jargon framework for knowing whether your data is actually ready to drive automated decisions before you find out the hard way that it isn't

The question sounds simple. The answer is anything but.

‍

At the Insurtech America Symposium, a room full of insurance executives sat down to wrestle with one of the industry's most pressing tensions: underwriting technology has never been more capable, yet many carriers still can't say with confidence that their AI tools are actually improving risk selection. They're moving faster. They're reducing admin time. But are they making smarter underwriting decisions?

‍

That was the central tension in "Underwriting Technology at a Crossroads: Decision Support or Digital Noise?" a session featuring Don Keleman, Managing Director US at Trendtracker, alongside Michelle Shaver, SVP of Commercial Lines Small Business at Encova, Sashi Aiyathurai, Executive Vice President and CUO at PMA Companies, and Jie Gao, Director of Advanced Sales Analytics at SunLife. The conversation was candid, specific, and loaded with practical insight for any carrier navigating this space right now.

‍

Here's what was worth writing down.

‍

1. Stop measuring AI by what it does, measure it by what it changes

Most AI underwriting tools on the market today are solving for speed: faster document retrieval, faster application ingestion, faster data consolidation. These are real gains. But speed is a byproduct of good AI, not the point of it.

‍

The panel pushed hard on this distinction. The question carriers should be asking isn't "how much time does this tool save per policy?" It's "is this changing the quality of our risk selection?". Those are very different evaluations, and most carriers are only running the first one. When evaluating any AI underwriting tool, the smarter move is to add an outcomes layer to your assessment criteria to define what better underwriting decisions actually look like for your book, whether that's loss ratio improvement, more accurate pricing on complex accounts, or appetite expansion into new risk categories, and hold vendors accountable to those metrics - not just efficiency benchmarks.

‍

2. The strategic opportunity most carriers are missing

Don Keleman, Managing Director US at Trendtracker, reframed the AI conversation in a way that cut through the operational noise dominating most industry discussions:

‍

"A lot of the focus these days for AI solutions is on the day-to-day underwriting side how do we optimize underwriting, how do we save 30 seconds within a policy. But from our perspective, it's also about utilizing AI to really understand the types of risks that we should be underwriting two to five years from now. Really using emerging signals, emerging threats and then trying to understand the type of impact that has in your specific industry, connected to your geography, connected to your specific use cases."

‍

This is where the real whitespace is. Most AI deployments are pointed at the transaction layer making existing processes faster. Very few are pointed at the strategic layer helping chief underwriting officers identify which risk categories to build appetite for before competitors spot the same opportunity, or which exposures to cut before losses start accumulating. The carriers closing that gap are running two parallel AI strategies: one optimizing day-to-day operations, a second feeding strategic intelligence to leadership so portfolio decisions are proactive, not reactive. If your current AI investment only covers the first track, that's the gap to close next.

‍

3. Fix the admin burden first then elevate human judgment

Michelle Shaver, SVP of Commercial Lines Small Business at Encova, put numbers to a problem the whole room recognized:

‍

"If you look at a small business underwriter, my estimate is 50 to 60% of their time is on administrative tasks that are not really making the great decisions that you want them to be focused on at the underwriting desk. So what we're doing and what the industry should do in small business is implementing AI tools to eliminate that administrative task so that your underwriters can focus on the decision-making task."

‍

One concrete example from the session: in small commercial, underwriters routinely spend 20 minutes before every referral decision manually pulling data from Google Maps, company websites, Verisk, and other sources then trying to synthesize it all before making a call. AI tools that consolidate this onto a single dashboard cut that preparation time dramatically, without removing the underwriter from the decision itself. The principle scales beyond small business: audit where your underwriters' time actually goes before you invest in AI. If the majority is administrative, that's where you'll see the fastest, most measurable ROI and the capacity you free up is what allows human judgment to land where it actually matters.

‍

4. Human-in-the-loop needs to move Up, not out

The panel pushed back on the binary framing of "human vs. automated" that dominates most AI discussions. The more useful question is where in the decision chain human judgment adds the most value and whether you're deploying it there.

‍

In small commercial, where average premiums sit around $4,000 and minimum-premium policies number in the thousands, putting a human in the loop on every referral isn't economically viable. Straight-through processing for a meaningful share of the book isn't optional it's a margin requirement. AI supplements the rules engines already doing that work, improving them by catching trends faster and validating data accuracy at the element level.

‍

In complex commercial and specialty lines, the calculus is different. Sashi Aiyathurai, EVP and CUO at PMA Companies, framed the risk plainly: in medical underwriting, there are documented cases where near-total adherence to model outputs without sufficient human review led to damaging consequences and litigation. AI in these environments works best as a decision-support tool, not a decision-replacement tool.

‍

Don added the layer that often gets skipped entirely:

"From our side, it's more like how do we make sure that the strategic decision-makers are actually really in the loop. If we look at the chief underwriting officer how do we make sure that they are aware of all these strategic opportunities that are emerging for the company, where they should pivot and where they should pursue? How do they know which type of risk to underwrite before your competitors are potentially seeing it?"

‍

The practical implication: map your human-in-the-loop design by account complexity and decision stakes. For low-complexity, high-volume business, keep human attention at the rules and governance level, not the transaction level. For complex accounts, protect the underwriter's role in final decisioning. And at the leadership level, create a structured process for AI-generated strategic signals to reach the CUO not just the desk.

‍

5. Don't add AI into a broken process

One of the panel's most consistent themes: the biggest implementation mistake carriers make isn't choosing the wrong tool. It's deploying the right tool into an unredesigned workflow.

Sashi Aiyathurai named the vendor-side version of this problem directly:

‍

"I often see solutions looking for a problem. I would love more and more to work together to really solve for the problems I have with solutions that can be customized to the needs that I have because a lot of times there's a challenge in trying to take this solution and plug it into the current things that are facing us."

‍

The pattern plays out on the carrier side too: an AI tool improves one step of the underwriting process by ten minutes, while the surrounding workflow remains fragmented and inefficient. Net result: marginal.

‍

Jie Gao, Director of Advanced Sales Analytics at SunLife, put her finger on why this keeps happening: legacy systems have forced carriers into modular, piecemeal technology deployments for years, creating fragmented data environments that now slow down AI adoption. Underwriters end up navigating multiple disconnected applications their legacy system, a new AI recommendation layer, several data sources with no coherent user experience tying it together. When that's the reality, adoption suffers and the investment stalls regardless of how good the underlying tool is.

‍

The fix isn't just better technology selection. It's workflow redesign before deployment. Map the full end-to-end process any AI tool touches. Design the user experience so recommendations surface where underwriters already work. And if the underlying process is broken, fix it before you automate it.

‍

6. Data quality is the foundation. Most carriers haven't built it yet.

The panel was direct: clean, trustworthy data is the precondition for AI that actually works and it remains unsolved for most of the industry.

‍

Michelle Shaver outlined a practical framework. Two metrics matter most: fill rate (how often a data element returns a result from your vendor) and accuracy rate (how confident you can be that the result is correct). High fill rates with low accuracy are dangerous you're making decisions on bad inputs. Her recommendation: purchase the same data element from multiple vendors, compare outputs, and use agreement between sources as a proxy for confidence. Where sources align, you have higher-confidence data you can act on with less human review. Where they diverge, that's where human judgment needs to stay in the loop. Conducting that kind of data readiness assessment at the element level, not just the vendor level is what separates carriers that can trust their AI outputs from those that can't.

‍

What the next five years look like and what it means for you now

The panel's closing predictions landed with some urgency. From Don’s perspective, the carriers that win five years from now are running two parallel AI strategies simultaneously one optimizing day-to-day underwriting operations, one feeding strategic intelligence to leadership. Carriers doing only one will be competitive. Carriers doing both will be hard to catch.

‍

Jie Gao flagged a shift worth watching closely, most AI underwriting tools today are built on general large language models. She expects the next five years to bring significant fine-tuning of those models on carriers' own internal data and sees this as a potential inflection point. The question she's watching: which line of business moves first, and which carriers are positioned to capture the advantage when it happens. For any carrier investing in data infrastructure now, that's the payoff on the horizon.

‍

Michelle Shaver closed with a warning that reframed the stakes entirely:

"I worry about AI and a subset of companies investing millions of dollars in AI, taking their expense ratio down drastically, and being able to price accounts and do it much more efficiently than the rest of the carriers in the marketplace. Jump into AI and use tools that will make you more efficient so that five, ten years from now, we still have a wonderful competitive marketplace in the United States versus a few monopolies running the insurance industry."

‍

The market is not going to wait for the middle of the pack to catch up. The carriers moving fastest on AI with disciplined data foundations, redesigned workflows, and strategic intelligence feeding leadership decisions are building advantages that compound.

‍

Practical takeaways: What this means for underwriters

A few questions and actions worth taking back to your organization.

‍

On evaluating AI tools

• Stop measuring AI by time saved per policy. Track impact on underwriting outcomes: loss ratios, pricing accuracy, appetite quality. If your vendors cannot articulate success in those terms, push harder.

‍

On human-in-the-loop design

• Map where human judgment sits today, then ask if that is where it adds the most value. For high-volume, low-complexity business, move human attention to the governance level, not the transaction level. For complex and specialty lines, protect underwriter involvement in final decisioning.
• Build a mechanism for AI-generated strategic signals to reach the CUO regularly. Most organizations do not have one yet.

‍

On data readiness

• Assess data quality at the element level, not just the vendor level. Fill rate and accuracy rate are the two metrics that matter. Cross-validate high-stakes inputs across multiple vendors before trusting them to drive automated decisions.

‍

On implementation and strategy

• Map the full workflow before deploying any new tool. AI dropped into a broken process will underperform regardless of quality.
• If your AI strategy is entirely operational, you have a gap. The carriers that win will be running two tracks simultaneously: day-to-day underwriting optimization and strategic risk intelligence feeding leadership decisions.

‍

Conclusion

The most important question to bring back from this session isn't "should we invest in AI?" That debate is over. The question is whether your AI investment is pointed at the right problems and whether it's reaching deep enough to actually change underwriting outcomes, not just underwriting speed.

Efficiency is the floor. Decision quality is the ceiling. The carriers that understand the difference and build toward both are the ones that will still be relevant when the next panel convenes five years from now.

‍

About Trendtracker

Trendtracker is an always-on AI strategy analyst platform purpose-built for innovation, risk, and strategy teams at enterprise insurers. By continuously scanning emerging signals across industries, geographies, and risk categories, Trendtracker helps carriers identify underwriting opportunities and threats before they become visible in loss data enabling smarter appetite decisions and first-mover positioning.

‍