(Image credit: IIR/Gemini.)
As I prepare for InsureTech Connect (ITC) 2026, an InsurTech vendor recently reached out to ask whether I would be publishing another Insurance AI Sauce Report—the post-event breakdown I typically write after attending ITC NYC in June.
My answer surprised him: I’ve decided it’s time to move on to a new topic.
It isn’t because I’ve lost interest in AI—far from it. I remain deeply fascinated by where the technology is heading, particularly emerging paradigms such as neuro-symbolic AI. The reason for retiring the report is simpler and more fundamental:
AI is no longer a differentiator in InsurTech discussions. Everyone has AI—even solo practitioners like me. The core question is no longer whether you have AI, but what real problem you are solving with it.
The Sea of Sameness on the Exhibition Floor
My decision crystallized after speaking with more than a dozen AI-focused vendors on the trade-show floor at ITC NYC. Vendor after vendor delivered nearly identical pitches.
One conversation in particular stuck with me. A very earnest product marketer walked me through his company’s platform:
“We provide a workflow solution that uses generative AI to perform tasks.”
“We use agentic connections to streamline activities.”
“We make things faster, cheaper and better for carriers.”
It sounded great on paper. But when I asked him directly, “What is your core differentiator?” he paused, looked at me and candidly admitted: “I have no idea.”
Of the more than 12 vendors I interviewed, almost all told the same generic story.
Except one.
The Power of the Sharp, Well-Defined Problem
The single vendor that stood out wasn’t interesting because of its AI stack—it was using the same underlying models as everyone else. What made it different was its focus on a specific business domain: a distinct set of complex legal problems within insurance.
Rather than having a marketing team look at macro industry trends and produce generic slogans about faster time to market or operational efficiency, this team sat down and asked a better question: Where is a specific, high-friction pain point that really needs solving?
It chose a problem that was perhaps less broadly applicable across every carrier but offered substantially greater value to the customers who needed it.
That mindset represents the dividing line between InsurTech companies that will thrive and those that will fade. Unless the goal is simply to be a generic utility provider, companies must apply AI to a sharp, well-defined problem set.
The Evolution of AI in Insurance: Rules vs. Reason
This shift in focus brings us to a major technical challenge facing the insurance industry: transparency and certainty. While the inductive logic of GenAI offers significant capabilities, it is not inherently transparent or certain.
To understand why this presents a hurdle, it helps to compare two broad approaches to artificial intelligence: symbolic AI and generative AI.
Symbolic AI, or rules engines, relies on deductive reasoning. Developers build a framework based on expected behaviors and known parameters, enabling the system to apply explicit rules consistently within those boundaries.
The advantage is transparency and clearly defined logic. The drawback is that considerable upfront engineering is required to build and maintain the rule sets. If the rules encounter unexpected data outside their framework, the workflow may fail or require human intervention. The system can be reliable within its defined boundaries but fragile at scale.
Generative AI, or pattern-matching systems, relies largely on inductive reasoning. Instead of following only human-coded rules, the model observes large datasets, learns patterns and builds an internal framework for handling novel and complex scenarios.
The advantage is flexibility and adaptability across messy, unstructured data. The drawback is that inductive reasoning can produce confidently incorrect results. Even when an answer is correct, a GenAI model may not provide a sufficiently clear explanation of how it reached a particular conclusion.
The “Show Your Work” Dilemma
This lack of transparency is already causing friction in regulated sectors. Legal challenges involving automated healthcare decisions have increased scrutiny of how insurers document and explain model-assisted determinations across large numbers of files.
When pressed, a pure GenAI system may point to statistical relationships among relevant factors. But regulated insurance decisions often require more: a traceable rationale, consistent application of rules and an audit trail that can be examined by regulators, courts and internal reviewers.
Combining Pattern Recognition with Explicit Logic
Neuro-symbolic AI isn’t a brand-new technology. It is a hybrid approach intended to bridge two established paradigms of artificial intelligence:
Neural networks, including GenAI: Effective at parsing unstructured data, spotting nuanced patterns and making sense of messy real-world inputs.
Symbolic AI, including rules engines: Effective at enforcing explicit logic, mathematical constraints and regulatory or business rules.
By combining these two paradigms, insurers may not have to choose between rigid rule engines and opaque black boxes. A neural engine can handle the work of reading complex policies, medical records or claims notes, while a symbolic layer applies business logic and defined guardrails. Properly implemented, this approach can improve traceability and produce a clearer audit trail for system outputs.
Neuro-symbolic systems can be organized in several ways, including parallel, pipeline and interactive architectures:
Parallel: The neural and symbolic components run simultaneously, providing complementary insights that are later synthesized. This can work in some scenarios and is often an initial step, but because the systems operate independently, integration and synthesis can become cumbersome.
Pipeline: The neural component processes and structures the data, which is then passed to the symbolic component for rules-based validation. This creates a defined connection and dependency between the two components and can make effective use of their respective strengths.
Interactive: The neural and symbolic components communicate iteratively, with the symbolic system constraining and guiding the neural model’s reasoning. This approach may support greater automation, self-correction and resilience in straight-through processes as the two models apply inductive and deductive reasoning to address exceptions.
Looking Ahead to ITC 2026
As we gear up for ITC 2026, my challenge to vendors and carriers alike is to shift the conversation away from surface-level AI buzzwords.
To vendors: Stop telling the market that you “use GenAI and agentic workflows to make things faster, cheaper and better.” Everyone is doing that. Instead, show us the specific, high-friction problem you solve better than anyone else—and demonstrate how your architecture supports transparency and defensibility when regulators come knocking.
To carriers: Stop evaluating solutions primarily according to the sophistication of their underlying large language models. Start looking for platforms that combine flexibility with explainable, well-controlled outputs suited to the business problem at hand.
AI is no longer the story—it’s just the engine. The real distinction will belong to those who know exactly where to apply it.