Introduction: The Promise and the Gap

Suppose you’re a C-suite executive at a carrier. In that case, you’ve seen the pitch: AI in insurance (from advanced machine learning to generative AI to fully autonomous “Insurance Agentic AI”) can reshape underwriting accuracy, accelerate claims, personalize distribution and fundamentally shift the cost base.

In fact, according to McKinsey & Company, the industry is “at an inflection point” where AI can no longer be treated as a novelty. And yet, the vast majority of insurers never move beyond the pilot stage. They dabble with models, proofs of concept, chatbots, clever dashboards—and then stall. The risk? AI becomes a check-the-box project rather than a source of competitive differentiation.

That gap—between promise and scaled reality—is the subject of this piece. And we’ll break it down not with fluff, but with fresh data, tangible patterns and actionable suggestions for how an insurer can move from “one use-case wonder” to “enterprise AI machine.”

Reality Check: Pilots Blossoming, Enterprise-scale Scarce

Across the industry, the story is sobering:

• BCG reports that while ~85% of insurers have initiated AI projects, only about 10% have fully scaled AI across operations—and they estimate the global value for underwriting alone could reach US$160 billion by 2030. (Source)
• In a 2024 survey of 200 U.S. insurers by Deloitte Touche Tohmatsu Limited, 76% said they’d implemented some form of generative AI in business functions—but the majority remain in scoping or pilot mode, not enterprise deployment. (Source)
• According to BCG again, the industry has the most robust early adoption of AI across sectors—but many of those efforts stall because of organisational or individual resistance. (Source)

So the question is not “Does AI matter for insurance?” It clearly does. The question is “Why do so many efforts die before scale—and how do you avoid that fate?”

Four Root Causes of the Scaling Failure in Insurance AI

Data, Legacy Systems and Workflow Fragility

One consistent theme: insurers often discover that “data readiness” is far more than ingesting a few datasets. It’s workflow readiness, core-system readiness, and access control readiness.

As one practitioner’s note puts it:

“We imagine data sitting in databases. For AI-ready data, we must think about data in motion — the processes around training, versioning, accessing AI models, transaction history.” (Source).

Many carriers still operate on decade-old core systems, siloed underwriting, claims, and distribution, and heavily manual processes that prevent real-time machine learning. Weak authentication and access controls for AI agents further compound these issues—slowing innovation, accuracy, and responsiveness across the enterprise.

If an insurer cannot pull clean, timely, high-trust data into an AI workflow, then even the best model won’t scale.

Use-case Fragmentation vs Domain Transformation

It’s tempting to launch dozens of pilots—chatbots, fraud detection, claim triage—but what matters is moving from “pilot” to “domain-scale” (claims, underwriting, and distribution). McKinsey authors say insurers are stuck in “pilot purgatory” because they prioritize technology over business value. BCG’s report echoes this: insurers that invest heavily in AI and scale it tend to focus on core domains, not isolated widgets. In short, you don’t scale by doing lots of disconnected experiments—you scale by rewiring a business domain with AI-enabled workflows.

Operating Model, Culture and Talent Gaps

Scaling AI demands more than data scientists. It demands a mindset shift, new roles (e.g., AI product owners and model operators), agile squads and business-IT collaboration. While many insurers are training in-house talent, the bigger issue is aligning business, operations and IT around a shared vision. If your operating model remains “business defines use case → IT builds model → pause”, you will never embed AI into the fabric of how things get done.

Governance, Trust, Risk and Regulatory Drag

Insurance is a risk-centric business. So when AI enters the picture, questions arise: model risk, fairness, privacy, interpretability and regulatory compliance. If those questions aren’t addressed early, AI stalls.

And the emerging risk landscape is real—such as AI tool errors, bias, exclusion of customers and “uninsurable” populations. In other words: if you ignore the trust/risk/governance piece, pilots may work—but scale will not.

The Fix: A CXO-Ready Framework for Scaling AI in Insurance

Here’s how you move from stalled pilots to enterprise-scale impact. These aren’t platitudes—they reflect what the data and consulting firms recommend.

Choose Strategic Domains & Link to Business Metrics

Rather than spreading across 20 modest pilots, pick 1-3 domains (e.g., claims automation, underwriting decisioning, distribution channel augmentation).

• Define upfront: What business metric will improve? (e.g., expense ratio, new-business conversion, retention uplift).
• Link that metric to volume, time-to-value and reuse potential. McKinsey notes that carriers who treat AI strategically—rather than tactically—get “outsized value.”
• One of our clients set a target: “reduce first-notice-of-loss (FNOL) to paid claim cycle by 25 % within 12 months via an agentic-AI triage engine.” That clarity forces prioritisation.

Build a Modular “AI-ready” Architecture

Scaling AI isn’t about one monolithic model; it’s about reusable components: data pipelines, microservices, workflow engines and monitoring dashboards. Architect with reuse in mind: set up a foundation that supports agentic AI workflows across domains.

Think: plug-in modules, versioning, experiment sandboxes, governance hooks, and production pipelines.

Rework Operating Model, Embed Change and Product Mindset

Scaling happens via teams that treat AI like a product—not a one-off project. That means cross-functional squads (business + data + IT + compliance), agile rhythms (sprints, MVPs, continuous improvement), and empowerment.

Change management is essential: for example, when an AI-enabled underwriting assistant goes live, underwriters must trust it, understand it, and adopt it. Without that, the model sits unused.

Also, define roles such as AI product owner, model owner and business champion. Create incentives for adoption and usage.

BCG report suggests that carriers scaling AI invest ~30 % more in AI capabilities than their peers (Source). That means budgeting for talent, change and measurement—not just technology.

Design Adoption, Trust and Risk-controls From Day One

Don’t bolt on governance and compliance later. Build it in from the start. That means:

• Data governance (lineage, quality, permissions)
• Model governance (versioning, audit trail, interpretability)
• Ethical guardrails (bias detection, fairness metrics)
• Operational monitoring (performance drift, cost/benefit, adherence)
• Regulatory readiness (state/federal, global)

This is non-negotiable in insurance. If you launch an agentic AI “underwriting bot” and the board asks how you monitor its decisions, you need an answer on day one — or risk scaling being blocked.

Focus on Value Metrics, Learn Fast, Scale Smart

Measurement matters. Too often, insurers focus on “number of models built” or “hours saved” instead of business outcome (e.g., claims severity down, conversion up and retention improved). Deloitte’s “Resources, Responsibility, Returns” model emphasizes that returns are the core readiness metric.  Tap pilots to learn (what worked, what didn’t), develop a playbook, and have criteria for “scale escalation.” Don’t attempt full-scale roll-out without clear pilot success and reuse potential.

Conclusion: The Tipping Point is Now

We’re in a moment where AI in insurance is no longer optional—whether it’s underwriting precision, claims automation, customer-centric distribution, or even new business models built around “Insurance Agentic AI.”

The challenge is not “Can we use AI?” but “Can we scale AI so that it meaningfully moves the needle for business?”
In short, you’re ahead of the crowd if you’re a CXO asking:

• Which domain will we focus on first for scale, not experimentation?
• Do we have a data, architecture and workflow foundation that supports that domain?
• Do we have an operating model and culture built for rapid iteration and adoption?
• Have we embedded trust, governance and risk controls from the start?
• Are our KPIs tied to business outcomes and do we have a roadmap to scale beyond one use-case?

Explore the Sutherland Insurance AI Hub

If you’re ready to move from proof-of-concept to performance, explore the Sutherland Insurance AI Hub — an enterprise-grade ecosystem built for insurers who want to scale responsible, explainable, and agentic AI across claims, underwriting, and customer experience.

Designed with compliance, interoperability, and human-in-the-loop governance at its core, the Hub helps carriers go beyond pilots — transforming AI from an experiment into an operational capability that delivers measurable value.

Visit Sutherland’s Insurance AI Hub to see how leading insurers are scaling trust, speed, and intelligence — all at once.