Most insurers have deployed AI in some form and most are seeing measurable results from it. The harder problem - making it work at scale across an entire organisation - is where the majority are stalling.
Accenture surveyed 263 senior insurance executives with direct accountability for AI, data, technology and business transformation across the Americas, Europe and Asia-Pacific between October and November 2025, supplemented by in-depth interviews with 15 executives from leading global carriers. The resulting report, How Insurers Drive Revenue by Deploying AI with Intent, finds 81% of respondents have achieved at least a 5% improvement in gross written premiums from AI and data initiatives. Yet only 23% have scaled AI enterprise-wide, with capability still fragmented across individual teams rather than embedded across underwriting, claims, distribution and operations as a whole.
Accenture has a commercial interest in the narrative that insurers need more AI transformation support - that is the work it sells. The underlying data is worth taking seriously regardless, because the gap it documents is real and has been independently corroborated: Datos Insights found earlier this year that while nearly 80% of large insurers have rolled out AI-assisted processing workflows, fewer than half have extended those programmes across more than one business function.
Half of Accenture's respondents cited legacy system integration as their top barrier to scaling AI, with data quality and accessibility the second most common obstacle at 45%. Separately, 70% of insurers run targeted AI skills initiatives, but only 14% have scaled these enterprise-wide - meaning AI expertise remains concentrated in specialist groups rather than distributed across the people who actually make underwriting decisions, settle claims, or manage distributor relationships day to day.
The scale of that skills gap is worth separating from the technology question, because they have different solutions. Legacy system problems are ultimately capital allocation problems - expensive, slow to fix, but understood. The skills gap is a workforce and culture problem that tends to outlast the system replacement programmes meant to solve it.
Accenture's parallel Pulse of Change survey found 86% of insurance employees said AI tools had increased their productivity, and 83% of insurers had reduced underwriting turnaround times by at least 5%. That points to a market where individual use cases are clearly working, and where the constraint is not whether AI can deliver value but whether insurers have the organisational infrastructure to make that value consistent and measurable across functions.
More than half of respondents, 56%, have implemented formal AI governance frameworks. Accenture frames this as a competitive advantage. It is more accurately described as a baseline that regulators and counterparties are beginning to expect - the FCA has been deepening its AI engagement, the SMCR already requires accountability for material decisions regardless of whether AI assists them, and the Treasury Select Committee's January 2026 report noted that over three-quarters of UK financial services firms already use AI for core functions.
What the 44% without a formal governance framework have not yet built is an audit trail for decisions that will increasingly be scrutinised after the fact. That is less a commercial differentiator and more an exposure that has not yet been tested in a significant claims or conduct dispute.
The 81% seeing revenue gains versus 23% achieving enterprise-wide integration is not a contradiction. Most of the revenue gains are coming from specific, bounded applications - pricing models, cross-sell propensity scoring, fraud detection at the claims intake stage. Those work at the use-case level and their results are measurable. Enterprise-wide integration requires something different: consistent data architecture, model governance that travels across business units, and enough AI fluency in the broader workforce that underwriters, claims handlers and actuaries can interrogate outputs rather than simply accept them.
That is a slower and harder problem than deploying another use case. The 23% figure is not an indictment of the market's AI ambition. It is an accurate reflection of how much organisational work sits between a successful pilot and a capability that runs reliably at scale.
For brokers placing business with major carriers, the practical read from this research is that AI-assisted pricing and risk assessment is now operating in most large markets - but that the consistency and governance of those systems varies considerably from firm to firm, and asking how a carrier's AI capability is governed is an increasingly relevant question at renewal and at tender.
