BCG's March 2026 report, "The AI-First Property and Casualty Insurer," sequences P&C adoption into three levels and attaches $35 to $60 billion of potential US operating cost reduction to them. McKinsey's April 2026 "Can Agentic AI (Finally) Modernize Core Technologies in Insurance?" reaches the same carrier audience with a different architecture. The interesting number in BCG's report is not the savings estimate. It is the share of the problem BCG says is not technology at all.

Key Takeaways

  • $35 to $60 billion in potential US operating cost reduction is BCG's headline, spread across a value chain where the largest single claim is up to 70% real-time resolution of simple claims.
  • 10% algorithms, 20% technology and data, 70% people and process is BCG's resource split, and it states that human and organizational factors account for 70% of scaling challenges among insurers.
  • Up to 3 percentage points of loss ratio improvement is credited to better use of unstructured submission data, the one projection on the list that is a pricing outcome rather than an expense outcome.
  • 38% of P&C insurers are realizing AI value at scale on BCG's count, against a Sedgwick-derived 7% at full-scale deployment. The definitions of scale differ; the gap does not close.
  • A write-off exceeding $500 million after eight years is BCG's own cited case for a Central European core system program, which is the risk the productivity ranges sit on top of.

The Blueprint and the Numbers Attached to It

BCG sorts use cases into three levels meant to be pursued in sequence, while allowing that carriers with mature data infrastructure can run several at once across different functions.

Deploy embeds generative AI in everyday tasks: underwriting, claims processing, customer support. These are the cases a carrier can implement on existing infrastructure with measurable return inside months. Reshape redefines the processes themselves, with a human-in-the-loop methodology, and changes job descriptions rather than task lists. Invent creates new products and business models, including migration off legacy systems, and BCG recommends three-month pilots before commitment. Few carriers have reached it at scale, which BCG says plainly.

The value chain projections underneath the framework are more granular than consulting estimates usually get:

Value Chain Function Key Metric BCG Projection
Underwriting efficiency Complex-line productivity improvement Up to 36%
Underwriting quality Loss ratio improvement from unstructured data utilization Up to 3 percentage points
Quote turnaround Time reduction Up to 60%
Underwriter capacity freed Time redirected from data gathering to judgment ~20%
Customer service productivity Overall agent productivity gains Exceeding 30%
Claims cost reduction Operational cost cuts 30% to 50%
Claims processing speed End-to-end cycle acceleration Up to 50%
Simple claims resolution Real-time resolution capability Up to 70%
IT migration time Legacy-to-modern transition duration 50% reduction
IT migration cost Program cost reduction 30% reduction

The 3-percentage-point loss ratio line is the one a pricing actuary should read twice, because it is the only entry that is not an expense claim. The mechanism is information: if underwriters can process submission data that previously went unread, such as statement-of-values attachments, loss run narratives and inspection photos, the asymmetry between carrier and risk narrows and individual risk pricing should improve relative to manual class rate application.

Where the 10-20-70 Split Lands on a Rate Filing

BCG's resource formula is the part of the report that constrains everything else in it: 10% algorithms, 20% technology and data, 70% people and process.

That split reprices the projections. The technology spend that draws board attention is a fifth of the work; the rest is change management, role redefinition, training and governance. BCG staffs it with three named oversight roles, review-and-approve, exception handling and quality calibration, each of which is a headcount line rather than a software line.

The consequence for rate filings is a timing problem, not a technology one. Morgan Stanley projects $9.3 billion in AI-generated operating income for P&C insurers by 2030, driven mostly by expense ratio reductions of roughly 200 basis points. A pricing actuary building prospective expense loads has to decide when those basis points become recognizable. Book them early and the rate is inadequate if the gains lag. Book them late and a faster carrier files under you.

The 10-20-70 split argues for booking late, and the external evidence agrees. Capgemini's May 2026 P&C research found 72% of carrier AI investment going to technology and infrastructure against 28% to change management, and 47% of employees given AI tools reporting unchanged workdays after 18 months of access. The carriers Capgemini calls intelligence trailblazers, its top 10% by maturity, are four times more likely to fund change management beyond basic training, and show 21% higher revenue growth and roughly 51% greater share price appreciation over three years.

BCG's own scale count is the same signal from the other direction: 38% of P&C insurers realizing AI value at scale, against a Sedgwick-derived finding that only 7% have reached full-scale deployment. The definitions of scale differ across those surveys. The distance between them is where an expense assumption goes wrong.

The Failures BCG Cites Are the Constraint

BCG supplies its own counterweight, and it is more useful than the projections.

A Central European insurer's core system program ran eight years before producing a write-off exceeding $500 million. A Southern European insurer's claims platform completed at 500% over budget. BCG traces both to the same cause: the programs started without sufficient understanding of what the legacy system actually contained. McKinsey names the same bottleneck, that in policy administration migrations the loops of discovery, mapping, testing, reconciliation and cutover consume the time rather than the coding, and puts testing and reconciliation at a 15% to 90% productivity range, the widest on its list.

The architectural fork matters here. BCG's zero-based design rebuilds around the outcome rather than replicating legacy configuration, which is the version that gives actuaries an opening to modernize rating algorithms and territory definitions instead of recreating rate tables. McKinsey's modular agent library preserves existing logic at high fidelity and accelerates the mechanical work around it. The first carries more execution risk against exactly the failure mode BCG documents.

For reserving, the cutover creates a problem the frameworks do not address. Claims cost reductions of 30% to 50% and processing acceleration of up to 50% do not change ultimate loss, but they change when it is recognized, and during a phased migration part of the book is running the legacy workflow while part runs the new one. That is a mixed-population development triangle, and the loss development factors fitted to the old one describe neither.

Further Reading on actuary.info