Generic AI is reaching its limit in insurance. It can summarize a submission, draft a customer response or extract information from a claims file. What it cannot do independently is make the high-stakes decisions that determine insurance performance.
Those decisions require more than a plausible answer. They depend on an insurer's data, pricing and underwriting rules, risk appetite, regulatory obligations, portfolio position and operational context. Without that grounding, AI may produce an answer that sounds credible but cannot be trusted, explained or acted upon.
According to Grant Thornton's 2026 AI Impact Survey, 44% of insurance executives said governance or compliance challenges had caused AI projects to fail or underperform, while 56% identified regulatory uncertainty as a leading barrier to scaling AI.
This is driving the industry toward vertical AI built specifically for insurance. Unlike general-purpose tools, vertical AI can apply intelligence within the context, controls and accountability that real insurance decisions demand.
The Danger of Faster Silos
Most insurers already have vast amounts of data, but rarely a complete view at the moment a decision must be made. Policy data may sit in one system, claims history in another and customer interactions somewhere else. Portfolio information and external risk signals may also arrive on different cycles.
Each source tells part of the story, but none necessarily reflects the full business context.
When insurers place AI on top of that partial view, they may produce an answer more quickly without producing a better decision. In high-stakes decision making, speed cannot compensate for missing context.
Insurance has traditionally been organized around separate disciplines, including pricing, underwriting, claims, customer engagement and portfolio management, even though the outcomes of those functions are closely connected. A pricing decision affects conversion and retention. An underwriting decision changes portfolio quality. A claims experience can influence renewal, customer value and future risk.
The business understands these relationships, but its systems do not always reflect them.
The risk is that insurers layer AI onto the same fragmented operating model. Pricing, claims, underwriting and service may each gain tools, while the organization still lacks a shared understanding of the decision being made. Individual tasks become faster, but the underlying disconnect remains.
That is the limit of horizontal AI. It can improve an activity without improving the enterprise decision surrounding it.
Insurance AI Should Follow the Decision
Policy administration, billing, claims and CRM platforms remain essential to insurance operations. They preserve records, process transactions and provide the stability insurers rely on. AI should strengthen the work performed within and across those systems, rather than attempt to replace them.
These platforms are highly effective at recording what has happened. The greater opportunity lies in helping insurers determine what should happen next.
A renewal decision, for example, may require policy history, recent claims, customer value, portfolio concentration, emerging risk signals, a pricing model, underwriting appetite and current regulatory requirements. No single platform owns all of that context.
The architecture must therefore be decision-centered rather than core-centered. Insurers need to identify the intelligence, rules, workflows and human expertise that must come together around each decision, then orchestrate those elements across the systems they already use.
Vertical AI extends far beyond adding insurance terminology to a general-purpose model. It is designed around how insurance work is performed, how decisions are governed and how their consequences are managed.
A rate change, for example, affects more than the price presented to a customer. It can influence portfolio composition, retention, profitability, fairness and regulatory filings. The AI supporting that decision must understand those connections.
Bringing the necessary context into the workflow may require predictive AI for risk and pricing, generative AI for explanation and guidance, and agentic AI for bounded, multistep execution. The type of model matters less than the insurer's ability to orchestrate the right intelligence around the decision.
The decision must come first.
Governance Must Travel With The Action
Insurance has always delegated authority within defined boundaries. Controls throughout the process allow insurers to distribute decision-making without losing accountability.
Agentic AI changes the risk profile because it can take action rather than simply provide advice. When an AI system can initiate or execute steps in real time, reviewing its activity after the fact may be too late.
Governance must therefore be embedded within the decision itself. It should define which data can be used, which models and rules may be applied, when human review is required and how the complete decision path can be reconstructed.
Not every AI component will be deterministic, but the route from input to action must be controlled wherever the decision demands it. Probabilistic intelligence requires deterministic guardrails, including defined permissions, monitoring, escalation paths, and the ability for a person to intervene.
The regulatory direction is already clear. The NAIC's model bulletin calls on insurers to establish governance, documentation, testing, validation and third-party oversight for AI systems supporting decisions that affect consumers. It also reinforces that insurers remain accountable for those decisions, regardless of the technology used to support them.
Human oversight must be designed into the operating model. This does not mean asking someone to rubber-stamp every AI-generated output. It means defining what can be automated, what must be reviewed, when escalation is required, and who ultimately carries responsibility.
Done well, this approach also gives experienced underwriters, actuaries and claims professionals greater leverage. Their hard-won expertise can become reusable decision logic, business rules and escalation paths, rather than remaining trapped in documents, manual processes or the knowledge of a small number of individuals.
Measure Decisions, Not Deployments
The insurance industry has invested heavily in data, analytics, predictive models, cloud platforms and core modernization. The remaining gap lies between those assets and the decisions that run the business.
A vertical approach allows insurers to advance AI without waiting for a multiyear core replacement. It can draw on the systems, data, and models insurers already trust, then orchestrate them around the decisions that need to improve today.
This makes existing investments more valuable while giving the business greater flexibility to adapt, without destabilizing the systems that keep operations running.
Success should therefore be measured by the quality and business impact of the decisions being improved, not by the number of AI tools deployed. That may include stronger pricing performance, more consistent underwriting, faster product changes, improved customer outcomes, or better portfolio management.
Insurance is an industry of connected consequences. AI built for isolated tasks will miss those connections, regardless of how capable the underlying model may be.
Insurance-native, vertical AI begins with the decision. It brings together domain context, predictive and generative intelligence, business rules, workflows, governance, and human judgment at the point where action occurs.
That is how insurers can move beyond isolated AI use cases and turn AI into a trusted, scalable capability for the decisions that matter most.
