Where insurers are actually cashing in on AI

Liberty Mutual credits AI as a key factor behind its 43% profit surge. Manulife expects $717 million in value from AI by 2027 through its work with Microsoft. But not every AI story is a success; what are these companies doing right that others get wrong?

At Liberty Mutual, the answer is treating AI as a deep integration rather than a series of standalone initiatives. Manulife is also going big, and has eliminated more than 400,000 manual intake and indexing tasks in its U.S. claims operations. But these may be outliers; 77% of financial services leaders say their AI investments don't have a measurable ROI, according to research from PwC.

Read more from the past week's coverage:

Liberty Mutual's embedded AI strategy drives 43% profit jump

Liberty Mutual reported $2.6 billion in net income for Q2 2026, up 42.8% year over year, and $13.3 billion in revenue, up 6% — with CEO Timothy Sweeney crediting embedded AI as a key operational driver. Rather than treating AI as a standalone initiative, the carrier has integrated it directly into underwriting, claims and service workflows, backed by model validation, governance frameworks and human accountability structures. The approach offers a template for peers: AI scaled without enterprise-wide integration and governance infrastructure risks underdelivering. Liberty Mutual's tools include an internal generative AI assistant and a consumer-facing conversational quoting app launched in May 2026.
Read more: Liberty Mutual provided gen AI to all employees and profit surged

Manulife targets $717M in AI value by 2027 with Microsoft deal

Manulife's five-year expansion of its Microsoft partnership offers a data-backed blueprint for large-scale AI deployment in financial services. The insurer is rolling out Microsoft 365 Copilot to more than 30,000 employees and deploying Microsoft Agent 365 for enterprise AI governance. Results already reported: developers using GitHub Copilot increased productivity approximately 30%, and AI eliminated more than 400,000 manual intake and indexing tasks in U.S. claims operations. Manulife reported approximately $213 million in enterprise AI value by the end of 2025 and projects roughly $717 million by 2027. The rollout pairs adoption programs with published Responsible AI Principles across four governance pillars.
Read more: How Manulife will expand use of AI with Microsoft partnership

77% of AI investments lack measurable ROI, PwC finds

Despite widespread AI adoption, 77% of financial services leaders report their AI investments aren't delivering measurable ROI, according to a PwC survey of more than 1,000 executives. The disconnect stems from tracking the wrong metrics. Rather than measuring task automation alone, insurers gain clearer returns by monitoring premiums written per underwriter, claims handled per adjuster, quote turnaround times and revenue per employee. PwC's insurance advisory practice also cautions that automating repetitive tasks creates value only when employees are redirected toward higher-value work. Firms that establish performance baselines and set defined financial targets before deployment will outperform those adopting AI purely to match competitors.
Read more: Why insurers struggle to measure AI's ROI: PwC

AI-driven MMM cuts wasted ad spend, lifts premium 30%

Insurance carriers relying on last-click or rules-based attribution are systematically misfiring on budgets that can run into the hundreds of millions annually — with some attribution models overstating channel contribution by more than 40%. Modern marketing mix modeling (MMM), powered by Bayesian regression, automated data pipelines and agentic AI, is correcting that. Early carrier implementations report 10% to 30% gains in incremental premium from the same budgets, reallocated correctly. Effective deployment requires a unified spend-to-policy data foundation, triangulated measurement across MMM, incrementality testing and attribution, and model governance built to withstand regulatory scrutiny from day one.
Read more: How insurers use AI with MMM to fix wasted marketing spend

Auditability is key as AI tackles property data errors

Insurers are deploying computer vision, synthetic data generation and rule-based AI to reconcile conflicting property data — mismatched coordinates, inconsistent terminology and derived spatial values that can distort risk assessments. Experts recommend codifying data-sourcing rules, such as "source of truth" hierarchies and recency rules, to ensure decisions remain consistent and transparent. Synthetic data can fill gaps created by emerging risks and secondary perils tied to climate events. Critically, the processes used to produce and transform property data must be fully documented: Without an auditability trail, downstream users — human or AI — cannot determine whether a given data point is fit for purpose.
Read more: Synthetic data from AI can fix costly errors in real data

Metadata forensics can help close AI fraud's detection gap

With 99% of insurers reporting encounters with manipulated or AI-altered claim documentation and only 32% confident in their ability to detect deepfake images, according to Verisk, the pressure to modernize fraud detection is acute. Metadata forensics — examining image creation dates, location data and file-modification timestamps — offers a scalable first line of defense. Carriers are building repeatable scoring systems that automatically flag anomalies and trigger manual or on-site review. Proprietary apps that capture damage photos in real time embed verifiable metadata at the source. A balanced approach — pairing technology with agent training and policyholder communication — avoids creating friction that slows legitimate claims.
Read more: Is it evidence or AI? Why insurers need metadata forensics

94% of small businesses trust AI insurance advice, survey finds

Small-business owners are arriving at agent conversations better informed than ever — 36% have used AI tools to research business insurance, versus 31% who consult an agent first, according to an ERGO NEXT survey of 501 U.S. small-business owners. Notably, 94% trust the AI-generated information they receive. For agents and carriers, the implication is a shift in engagement strategy: lead with complex coverage analysis and relationship-building rather than foundational education. The data also signals a persistent protection gap — 25% of small businesses carry no insurance and 64% of those with coverage are underinsured, with cost and confusion cited as primary barriers.
Read more: Why small businesses trust AI advice but still want agents

Organize data now to gain AI edge, insurtech CEO says

Go Abacus CEO David Moscatelli, whose Chicago-based company raised a $5 million Series A in November 2025, argues that insurers' most pressing priority is data readiness — not AI adoption itself. Executives who centralize and organize institutional data first will be positioned to capitalize on AI's precision in risk pricing and actuarial forecasting. Moscatelli warns that excessive caution around technology adoption is itself a material risk. Near-term opportunities include AI-assisted cross-selling — auto policyholders who also own homes, boats or RVs — and improving coverage explainability to reduce customer service friction. Go Abacus deploys on-premise AI infrastructure to avoid sending sensitive data to third-party cloud platforms.
Read more: Don't let risk aversion stall innovation: Go Abacus's David Moscatelli

AI drift threatens claims, underwriting as revenue hits $1T

With unmonitored AI revenue projected to surge from $150 billion to $1 trillion by 2028, model drift poses a direct threat to claims accuracy and underwriting integrity. Drift occurs when identical queries yield inconsistent outputs — a 2% swing in claims rejection rates can signal a problem requiring immediate review. To counter it, carriers should conduct A/B testing to verify AI agents are pulling from consistent data sources, build "golden sets" of pre-reviewed underwriting cases covering edge cases, varied risk types and demographics, and track output versions to identify input changes. Human oversight of AI workflows remains essential throughout.
Read more: How AI model drift will thrive as AI revenues reach $1 trillion

Keep humans accountable as AI accelerates insurance workflows

Rapid AI adoption risks displacing the professional accountability insurers owe customers, agents, regulators and shareholders. The more productive approach: map friction points first — tasks that are repetitive, slow or judgment-light — then deploy AI there while keeping final decisions with credentialed professionals. At Managing Success Insurance (MSI), AI cut real estate portfolio quote turnaround from four to five days to under one day, and reduced high-net-worth home risk file review from 20–30 minutes to a few minutes — without removing underwriter sign-off. Insurers that create internal AI champion groups, drawn from frontline staff rather than executive committees, convert general enthusiasm into testable operational improvements faster.
Read more: Why losing accountability is the real AI risk

Chemical incidents expose gaps in pollution liability coverage

Recent chemical plant emergencies in Washington and California — including a tank failure that killed 11 people and an overheating incident that prompted mass evacuations — are prompting renewed scrutiny of pollution liability coverage. Most general liability policies contain full pollution exclusions, yet many businesses with chemical exposures carry no standalone pollution coverage. Blended GL/pollution policies offer partial protection but share aggregate limits, leaving insureds vulnerable when large claims arise. A standalone pollution policy provides dedicated limits and the broadest protection. Brokers have an opening now to audit clients' existing coverage, identify gaps and initiate risk tolerance conversations before the next incident occurs.
Read more: Why brokers and insureds need a new pollution liability strategy

This roundup was created with AI assistance. A Digital Insurance editor reviewed each item before publication.