Insurers have invested heavily in Gen/Agentic AI on the promise of transformation. Innovation labs were established. Pilots have been encouraging. Vendor demos generate excitement. Yet for most carriers, brokers, and MGAs, the returns are difficult to identify. Boards, investors, and clients are now asking the obvious question: where is the payoff?

The commitments insurers are making to the market suggests they believe the answer is within reach. Chubb, for instance, has publicly outlined plans for up to 20% headcount reduction over the next three to four years, targeting automation of 85% of its major underwriting and claims processes and projecting run-rate expense savings of approximately 1.5 combined ratio points.¹ These are not aspirational slide decks. They are investor-grade commitments.

What is striking is that the barriers to scaling AI are not primarily technological. The models are improving rapidly. Pockets of genuine innovation exist across the industry. While Gen/Agentic AI’s growth is unprecedented, the obstacles are all too familiar; as research consistently affirms, the challenges are organizational and capability related. People and data are what stall progress, not the tools themselves. The cost to build is lower now than it has ever been, raising expectations for accelerated value creation and sharpening the scrutiny on organizations that have yet to move beyond experimentation. Failing to invest in capabilities and operating model reinvention will ensure these experiments remain exactly that—expensive experiments.

Insurers that get the organizational dimensions right will pull ahead. Those stuck in pilot purgatory risk ceding ground to more agile competitors and Insurtech entrants that do not carry the same legacy baggage. 

Insurers are spending on Gen/Agentic AI. Period. But ROI is elusive. The gap between ambition and return is, above all, an operating model problem. Three imperatives consistently separate the leaders from the laggards, and all three are organized around that diagnosis: creating quick wins to establish credibility, establishing the data foundation to scale momentum, and building the operating model to sustain it.

AI Imperative 1 – Quick Wins: Prioritize and Deploy with Strategic Discipline

The first operating model discipline is to know where to focus. There is no shortage of AI use cases for insurers to pursue – what many lack is a disciplined framework for deciding which ones deserve real investment. The numbers are stark: about 30% of AI initiatives in insurance progress beyond the pilot stage. Organizations that do manage to scale outperform their peers by 3 to 5x on productivity and efficiency metrics.²

A large part of the problem is proliferation without prioritization. When dozens of disconnected pilots are running across different business units, each with their own budget and sponsor, the result is fragmented effort, diluted resources, and disappointing ROI. Effective leaders take a different approach; they prioritize ruthlessly, evaluating use cases against economic impact, product and line-of-business fit, geographic and regulatory context, and scale of deployment.

The broader market is already voting with its capital. In Q3 2025, roughly 75% of Insurtech funding went to AI-centered companies.³ The question for established insurers is no longer whether to invest in AI, but how quickly they can scale before competitive dynamics cement the early movers’ advantage.

One practical lesson we see repeatedly: it is better to aim small and miss small. Start with focused pilots, use those successes to build momentum and organizational trust, then scale. Tackling the largest problems first with costly, enterprise-wide rollouts tends to produce the opposite of what was intended.

AI Imperative 2 – Scale Momentum: Establish the Data Foundation and Governance

The second operating model discipline is a maniacal focus on data governance. If there is one thing that derails AI initiatives before they reach production, it is data – not the algorithms, not the computer, not the vendor. This binding constraint is not a revelatory observation, but it remains a stubbornly underappreciated one.

Legacy administration systems, siloed platforms, and vast stores of unstructured data complicate how information is captured, stored, and accessed. This is a primary reason pilots fail to graduate. Leaders are addressing it: investing in modern, modular data architecture, moving away from monolithic data warehouses, building rigorous governance frameworks, and creating reusable “data products” organized around specific use cases rather than generic enterprise schemas.

The regulatory environment is adding urgency. In the US, 23 states have adopted the NAIC’s AI Model Bulletin, and a NAIC “AI Systems Evaluation Tool” pilot is expected to launch in 2026, marking a shift from principles-based guidance to active supervisory scrutiny.^* In Europe, the EU AI Act classifies AI used for insurance risk assessment and pricing as “high-risk,” with mandatory compliance obligations taking effect as early as August 2026.^** Carriers that have not invested in data lineage, model documentation, and audit trails will find compliance painful and expensive.

Data and AI governance can no longer be treated as a purely compliance exercise; they must evolve in lockstep with the technology as insurers progress from traditional predictive models to generative and agentic AI architectures.

AI Imperative 3 – Continuous Improvement: Build a Product Operating Model Designed for AI-Driven Transformation

Many of the insurance operating models and value chains were designed for a different era, and business is trying to catch up. Adopting a product operating model — built for speed, cross-functional collaboration, and iterative delivery — is not one option among many. It is the precondition to deliver lasting value.

This last imperative is where most companies stumble. The technology works well enough. The data can be fixed. However, the operating model – the way decisions get made, teams are structured, and work gets done– is frequently misaligned with what AI requires. Unlike a technology gap, an organizational gap cannot be closed with a vendor contract. What is needed is adoption of a product operating model, built for speed, cross-functional collaboration, and iterative delivery.

The traditional operating model is hierarchical, siloed by line of business, and built on long product cycles. That structure is poorly suited to the speed and iteration that AI demands. While insurers continue to spend on traditional AI, current and future investments are tilting heavily toward generative and agentic AI. Despite this, more than 40% of insurers reported inadequate internal skills and expertise to implement these technologies effectively,^*** adding significant risk to the success of these investments.

The talent challenge cuts both ways. The industry must develop “AI translators” in the team who understand both insurance operations and AI capabilities. Critical thinking and a bias toward collaboration are essential skills for the “human-in-the-loop”. At the same time, a wave of retirements is creating real urgency to capture and transfer institutional knowledge to junior staff before it walks out the door. As we argued in “The Underwriter Redefined,” recasting roles as tech-enabled, advisory, and client-facing reshapes the value proposition for the next generation of talent.

An effective product operating model should be characterized by five things: proximity to the customer, sales-led front-line empowerment where AI augments underwriters and adjusters rather than replacing them, product owner-driven agile delivery with cross-functional squads replacing siloed handoffs, AI Centers of Excellence for enterprise governance, and an organizational culture robust enough to continuously embrace change.

One final point that gets overlooked: change management costs should be budgeted as a meaningful proportion of total AI investment and management attention, not treated as an afterthought. Organizations that underinvest in adoption consistently fail to realize returns, regardless of how beneficial technology promises to be. Business must embed a commitment to changing behaviors and ways of working. Creating a culture that embraces experimentation and failure, one that learns and iterates, is hard. Product Operating Model principles create the transparency and user focus that drives cross-functional buy-in. Adoption is what drives benefits, and the carriers that ultimately win with AI will be the ones that pair sound technology with the organizational agility and discipline to put it to work.

SOURCES

¹ “Chubb to cut up to 20% of workforce in ‘radical’ AI drive,” Insurance Business, 2025.

² “Only 30% of insurer AI projects make it past pilot stage,” Insurance Business, 2025.

³ “Top 10 Insurance Industry Trends Shaping Underwriting in 2026,” Send Technology, 2025.

^* “Tracking the Evolution of AI Insurance Regulation,” Fenwick, 2025.

^** “The EU AI Act and Insurance,” Harvard Data Science Review, 2025.

^*** “Insurance in the AI Era,” IBM Institute for Business Value, 2025.