A recent Celent report on technology trends in Property & Casualty insurance describes a strategic transition that is already underway: insurers are moving away from operating as simple claims payers to become architects of resilience—meaning companies that help their clients prevent risk before it materializes, rather than just compensating them after the fact. In parallel, other industry analyses in Latin America converge on a similar point: artificial intelligence is ceasing to be an isolated proof of concept and is embedding itself at the core of the business, present in underwriting, claims management, customer service, and back-office operations.

For the life and health sectors, this transition has a more concrete name: the evolution toward continuous underwriting. Instead of assessing risk once a year using a static snapshot of the customer, it begins to be evaluated dynamically, using real-time data from wearables, electronic health records, and wellness apps.

All of this sounds, with good reason, like a real upgrade for both the customer and the business. However, the same Celent report that describes this transformation warns, quite bluntly, about its accompanying risks: a lack of model transparency, unverifiable biases or errors, over-reliance on tools, and the erosion of the human skills that previously anchored those very same decisions.

This is not a generic "use AI with caution" warning. It is a precise description of what happens when an insurer automates underwriting or claims management faster than it strengthens its capacity to audit those decisions. In a regulated sector—where every underwriting choice or claims payout can end up reviewed by a regulator or disputed by a client—this gap between speed and traceability is not a minor technical detail: it is the core risk of the entire transformation.

A business rule in insurance—defining which policy gets approved, under what limit, and with which exceptions—is simultaneously a compliance rule. When that rule is coded without treating both elements as a single, unified responsibility, the result isn’t just a faster decision engine: it’s a faster decision engine that is significantly harder to audit when a decision is questioned.

In the projects where we have worked with configurable rule engines for policy approvals—fully integrated with legacy systems and featuring auditable traceability by design, rather than tacked on later—the lesson is always the same: traceability cannot be added to an already-built decision system as if it were a separate module. It is either designed alongside the business rule from day one, or you end up paying the steep price of rebuilding it under regulatory pressure.

We continue to deepen our expertise in this sector with the same discipline and honesty we already apply to banking: we prefer to build that depth project by project, backed by real evidence, rather than over-promising on an industry track record that we are still actively consolidating.

• Can your underwriting or claims model explain, line by line, why it made a specific decision? Not "in general," but in the exact case a regulator or customer might dispute.

• What happens if your real-time data source fails or arrives incomplete? A continuous underwriting model is only as reliable as the least reliable data point it receives.

• Who audits the model—and how frequently—once it is already in production? Governance is not a document you sign before launch; it is an ongoing discipline that continues long after.

Becoming an architect of resilience is absolutely the right ambition for the sector. However, business resilience depends entirely on the engineering behind each AI model being just as auditable as it is fast. An underwriting or claims model that decides correctly most of the time—but cannot explain why—is not solving the industry's problem; it is merely postponing it until the first serious audit.