The CEO of SAP caused a stir in late January when he said, "The end of the keyboard is near.... “The future will be, for sure, that you are not typing any data information into an SAP system." He added that people will use their voices, not their fingers, to ask analytical questions of SAP systems, to trigger tasks, to make pipeline entries and more. 

A recent Wall Street Journal column carried the headline, "Our Gadgets Finally Speak Human, and Tech Will Never Be the Same." The columnist, Christopher Mims, wrote: "This is shaping up to be the year that AI makes talking as powerful as tapping and swiping. The shift could be as transformative for the tech industry as the introduction of the Mac, Windows or the iPhone."

While I'm not convinced the change will be quite as total or as fast as the SAP CEO says it will be, he and Mims are describing an important new wave of convenience and productivity that AI will provide.

Let's have a look at what will — and maybe won’t — happen.

Ever since I learned to type in ninth grade, I haven't been fully convinced I could have a serious thought without my fingers on a typewriter or a keyboard. I'm sentimental enough about my father's manual typewriter that I have one displayed on my bookshelves. But, if you step back and think about it for a moment, keyboards are extremely inefficient, especially as a way to control a computer. 

The core problem is that few people can type as fast as they think. But beyond that, we're always moving our cursors around to click on one thing or another. A touch screen can remove some of the inefficiency, but even then you're moving on and off the keyboard frequently. 

Voice is a much faster and more natural way to do a lot of the things we now do with our fingers — and the voice capabilities of AI have improved by leaps and bounds. As a result, the capabilities are starting to be built into software that we use for all sorts of tasks. 

I now dictate my text messages — saving me ever so much time by ending the trouble my fat fingers used to cause me. Voice is also showing up in word processing and email systems. Voice will allow me to tell my email that I want to reply to someone, then to dictate a quick response and be done with it. As the CEO of SAP says, data entry is also a natural for voice. No more peering over at forms, while trying to read the numbers on a sheet of paper and flipping your eyes back and forth between the paper and the screen to make sure you've typed the information correctly. You just hold the paper in front of the screen and read the data, which you can easily check with a glance. 

Combining the voice capabilities with generative AI, we'll be able to just ask for data, for access to corporate systems or for any number of other things that would have taken us much longer in the past. No single action will save a ton of time, but the efficiencies will add up, and considerable drudgery will disappear. The computer mouse was an exceptionally important invention, but voice is faster and more natural. 

The voice capabilities will soon be everywhere, because the arms race for AI dominance is still going full speed. In recent earnings announcements, big AI players said they were going to spend $650 billion on AI infrastructure in 2026. That's roughly the GDP of Belgium or Nigeria  — and those announcements were just from four companies. Eleven Labs, a startup using AI to translate speech to text and vice versa, just raised $500 million at an $11 billion valuation, more than three times the valuation when it raised capital a year ago. Mims, the WSJ columnist, says the AI voice capabilities will increasingly find their way into convenient hardware, including Meta's glasses. 

I still don't think keyboards will go away, at least not soon. I still can't really think without my fingers on the keys for something as long as what I'm writing here, and, in general, I think keyboards will be important for editing. Talking to my computer to dash off a quick, formulaic email is one thing. But editing is serious business, and it's easier just to use a mouse or a finger to get to the offending spot in the text than it would be to tell the computer to go to the second sentence of the third paragraph, to a specific word in the sentence that starts with XYZ. Even with the explosion of voice capabilities, it will take time for software developers to come up with the right mix of voice and touch commands. 

I also don't think, as I've written previously, that voice will be a great way to buy things, certainly not things as important and complex as insurance. The lure of voice for purchasing was behind a lot of Amazon's early efforts with its Echo devices, but voice-based buying hasn't really happened even for paper towels and dog food. For anything more complicated, a buyer wants to see all the details so they can weigh the various variables against each other. You need a screen for that, and likely a keyboard.

The transition will be a journey and will take the form of a voice/keyboard hybrid for the foreseeable future. But I'll certainly be happy to increasingly use my voice to control my devices, saving bits of time and of frustration along the way. 

Cheers,

Paul

Insurers are navigating fast-moving markets, shifting performance, and growing data volumes. Machine learning and AI can turn complexity into clarity by spotting what's changing, explaining why, and guiding action with confidence.

Used appropriately, machine learning speeds analysis, strengthens decisions, and reduces manual effort without compromising governance. I will outline five practical ways machine learning is advancing portfolio management today and how modern tooling helps teams manage the complexity that comes with larger model estates and richer data.

1. Identify emerging trends faster

When performance diverges from plan, the difference between a timely intervention and a late response often comes down to signal detection.

Machine learning helps surface early signs of change: whether it's a spike in repair cost inflation, a drop in claims frequency in a specific region, or a retention shift that alters portfolio mix.

By revealing meaningful patterns sooner, teams can protect margin, target profitable growth, and rebalance exposure before small deviations become big problems.

2. Strengthen risk assessment and segmentation

Machine learning brings sharper granularity to segmentation and risk assessment, enabling insurers to recalibrate assumptions as conditions evolve. The most effective approaches are built for insurance: explainable to business stakeholders, defensible to regulators, and practical for day-to-day use.

The payoff is clearer pricing and underwriting decisions across geographies, demographics, and product features—turning complex data into decisions that withstand scrutiny.

3. Accelerate decision-making with automation

Automation turns model monitoring from a periodic task into a continuous source of insight. Near-real-time reporting highlights what changed and where to act, while governed thresholds can trigger refits or reruns to keep performance on track.

This entire process can be automated. Automated model monitoring allows businesses reliable insights on business performance; portfolio and segment trends; and changing risk exposures, all while models can also be automatically rerun in the background if their performance degrades below a certain point. Analytics teams can get back to identifying new opportunities, rather than manually reviewing current model performance.

It's important to note monitoring is a key part of portfolio management in model-dense environments, but it's not the whole story. It complements broader actions such as dynamic pricing, geographic rebalancing, channel strategy and product design—helping teams move from observation to execution faster.

4. Manage complexity without the drag

As insurers expand their model estates and integrate richer data sources, the challenge shifts from building models to running them efficiently at scale. Many insurers now operate estates numbering in the hundreds, a testament to how quickly the industry has embraced machine learning. With scale comes complexity. More models mean more oversight, especially as newer model types tend to degrade faster. If models underperform, so will the business.

Modern platforms help insurers stay ahead by combining automation, versioning and governance, allowing teams to maintain transparency and control without slowing down. A governed environment for deploying AI and machine learning models, including Python, reduces IT bottlenecks while preserving auditability. Radar’s Python deployment component enables insurers to benefit from the flexibility and innovation of open source but in a controlled, robust manner that supports business critical decisions and generates real value. This combination of flexibility with control turns operational complexity into a strategic advantage.

5. Go beyond monitoring to steer the portfolio

Machine learning's role does not end with detecting issues. It also helps answer "what if?", from the impact of price changes and rate adoption to exposure limits and product mix.

By pairing trend detection with scenario testing, insurers can quantify trade-offs before rollout and build a continuous loop from insight to action to measurable outcomes. The result is a more responsive, disciplined approach to portfolio management that aligns daily decisions with strategic goals.

KEY TAKEAWAYS

• The membership economy is reshaping insurance by replacing transactional, annual renewals with continuing, affinity-driven relationships rooted in trust and shared identity.
• Subscription-based services offer insurers a sustainable growth path by embedding protection as a seamless benefit of membership rather than a standalone product shopped for based on price.
• The future of the insurance industry depends on shifting from policy-centric models to member-centric experiences that deliver continuous, visible value beyond claims.
• Affinity organizations hold a competitive advantage because their existing trust and engagement dramatically reduce churn and customer acquisition costs.
• Scalable insurance technology solutions are essential to enable frictionless self-service, personalization, and real-time engagement that modern members now expect.

For decades, insurance has been a transactional, "set it and forget it" chore tied to an annual renewal. But this episodic approach is increasingly out of step with how people actually live.

Today, the most successful protection models are built on affinity. Consumers aren't looking for another disconnected vendor; they want to leverage the memberships and associations they already trust. They make purchases within these trusted circles because of the value and identity those groups provide.

For the insurance sector, this means shifting from a passive contract to a seamless, value-added benefit of belonging, turning insurance from a standalone bill into a core advantage of the affinity relationship.

The Death of the Transactional Mindset

The rise of the subscription economy has fundamentally reshaped consumer behavior, moving the needle from ownership to access and from transactions to relationships. In most retail and service sectors, the "membership mindset" has eliminated the friction of the re-purchase decision. When a consumer subscribes, they are choosing to bypass the exhaustion of constant comparison. However, this loyalty is not granted for free; it is traded for continuing value, transparency, and ease of use.

Consider the Amazon Prime model: Members don't spend hours cross-referencing prices across a dozen different websites for every household item. Instead, they head straight to Amazon, use "Buy Now" for a frictionless checkout, and move on with their day. That loyalty is rooted in the belief that, as a member, they are already getting a competitive price and a level of convenience that far outweighs the potential of saving a few cents elsewhere.

To understand the future of the insurance industry, we must look at how subscription models bridge the loyalty gap. In the traditional model, the annual renewal cycle creates a natural friction point. It's at this moment where the lack of a deeper relationship encourages the customer to comparison shop.

Shifting to an affinity-based membership model changes this dynamic entirely. Instead of the renewal notice acting as a yearly "call to action" to find a cheaper alternative, the insurance remains anchored to the value of the organization itself. When protection is part of a trusted membership, the "choice" to stay is already made; the consumer remains because the insurance is a seamless part of a community they already value.

Trust as an Operating System in Affinity

While legacy insurers struggle to pivot away from actuarial-centric models, affinity groups possess a natural advantage. These organizations already have the most expensive ingredient in the subscription recipe: trust.

Traditional carriers spend billions on customer acquisition, often fighting for "switching" customers who are motivated solely by price. Affinity insurers, conversely, operate within a framework of pre-existing loyalty. By adopting a subscription-based model, these groups can embed insurance within a larger bundle of perks, such as exclusive content, community access, or professional tools.

This bundling changes the psychology of the consumer. If a policyholder feels the value of their membership every week through a discount portal or a professional resource, the underlying insurance product becomes "sticky" by association. The subscription model allows affinity groups to move away from being a mere distribution channel and toward becoming a holistic service provider.

The Role of Frictionless Technology

The transition from a policy-centric model to a member-centric one is impossible without a robust digital foundation. Insurance technology solutions are the backbone of this evolution. In a subscription framework, the "user experience" is the product. If a member can upgrade their Netflix plan in two clicks but must call a broker and wait 48 hours to adjust an insurance limit, the relationship is doomed to fail.

Modern insurance technology solutions enable the self-service capabilities that subscribers now demand. This includes flexible payment structures, transparent "tiers" of coverage, and the ability to pause or pivot protection as life circumstances change. Beyond the interface, technology allows for the data-driven personalization that defines the membership economy.

By leveraging member data and technology, insurers can move from being reactive, such as merely responding to a claim, to being proactive by offering a specific coverage adjustment based on a member's life stage or behavior.

Overcoming the "Invisibility" of Insurance

The greatest challenge in applying the subscription model to insurance is the lack of a tangible "delivery." Unlike a streaming service where you see the content daily, or a meal kit that arrives at your door, insurance is often invisible until something goes wrong. To ensure long-term success, insurers must find ways to communicate value between claims.

This requires a total rethinking of engagement. A successful insurance subscription should provide "living benefits." This might include risk mitigation alerts, wellness rewards, or integration with IoT devices that provide the member with a sense of security and utility on a regular Tuesday afternoon, not just when a basement floods.

Furthermore, the industry must grapple with legacy infrastructure. Most core systems were built for the rigid architecture of annual cycles. Moving to a subscription model requires an investment in agile billing systems and CRM platforms that can handle the high-frequency interactions of a membership relationship. It also requires a sophisticated approach to regulation, ensuring that monthly billing and "cancellation at will" policies comply with state-level consumer protection laws.

The Future of the Insurance Industry

The shift toward the membership economy requires the insurance industry to stop asking, "How do we sell a policy?" and start asking, "How do we become an indispensable part of this person's life?"

For affinity groups and forward-thinking carriers, the subscription model offers a path toward increased lifetime value and decreased churn. By mirroring the convenience of subscription-based services and leveraging insurance technology solutions, the industry can finally bridge the gap between what it sells and how people actually want to buy.

The future of the insurance industry is one where protection is seamless, the benefits are constant, and the relationship is renewed not by a contract, but by the continuing delivery of value. The insurers that thrive will be those that stop treating people like risks to be managed and start treating them like members to be served.

The middle market remains one of the most persistent protection gaps in the insurance industry. Middle-income households face meaningful financial exposure yet remain underinsured across life and annuity (L&A) insurance. Traditional insurance distribution models were designed for high-touch sales, longer underwriting cycles, and relatively high premiums. These models struggle to operate efficiently for lower-margin products that require speed, simplicity, and flexibility. As a result, many middle-market customers remain unserved, not because insurance is irrelevant but because existing distribution models cannot reach them at scale.

Traditional Distribution Model Wasn't Built for the Middle Market

Insurance has long been described as a product that is sold rather than bought. This dynamic becomes especially pronounced in the middle market, where customers often lack the urgency or familiarity that would prompt purchasing through conventional channels. At the same time, the economics of agent-led distribution are poorly suited to products with lower premiums and shorter lifecycles.

Middle-market customers typically require coverage that can be purchased quickly, adjusted over time, and delivered within digital experiences they already use. Traditional models, built around lengthy sales processes and manual servicing, are misaligned with these expectations.

How AI Makes Embedded Distribution Scalable

Embedded insurance is no longer a niche distribution experiment. Recent analysis estimates global embedded insurance sales of $87.4 billion, projected to grow at a 20% CAGR from 2023 to 2032. The channel is maturing quickly, but its strategic significance is not simply reach. It is that embedded models shift how protection is priced, sold, and supported when coverage is delivered inside third-party ecosystems rather than insurer-owned journeys.

As outlined by RGA, embedded programs generally take three forms. Soft-embedded (opt-in) insurance is presented contextually at the point of a primary purchase, such as travel insurance offered during flight booking. Hard-embedded (opt-out) insurance is included by default within a broader transaction, requiring the customer to actively decline coverage. Invisible insurance is embedded so deeply within the primary service that coverage is automatically activated based on participation in the service.

While embedded distribution expands access, it also introduces new operational pressures. Embedded products are high-volume, lower-premium, and event-driven, which compresses decision timelines and pushes servicing costs closer to economic limits. Servicing expectations remain high, even as tolerance for manual intervention declines.

This is where AI plays a critical role. Embedded insurance changes not only where insurance is offered, but how protection is evaluated, quoted, bound, and serviced. For insurers and MGAs, AI is applied across distributed journeys to coordinate decisioning, data access, and workflow execution across underwriting, policy issuance, servicing, and claims.

Applied at the workflow level, AI aligns automated decisions, business rules, and human oversight within a single operating flow. This supports consistency, traceability, and governance as products scale across partners and channels.

Embedded insurance only succeeds when the experience is seamless. For bite-sized products, this translates into near-100% straight-through processing at purchase and a largely touchless claims and servicing experience thereafter. AI enables this by coordinating real-time data access, automated decisioning, and workflow execution across the lifecycle, allowing insurers to scale volume without increasing operational friction.

Embedded Insurance in Practice

These models place different demands on insurers across product design, distribution, servicing, and compliance. They also reset expectations around speed, simplicity, and relevance, particularly in middle market segments where traditional distribution often struggles to operate economically. Boston Consulting Group has cited Prudential Financial’s Simplified Solutions initiative with Neutrinos as an example of how AI-enabled, embedded distribution models can expand access to life insurance in North America.

Market sizing reinforces the upside: Forrester estimates that building new solutions for just 1% of the roughly 4 billion underserved people globally could translate into approximately 40 million new customers, if insurers can deliver simpler products through scalable journeys.

Closing the Gap at Scale

Embedded insurance is increasingly proving itself as a scalable distribution model rather than a niche channel. However, distribution alone is insufficient. Insurers that invest in the operational foundations required to support high-volume embedded products, including AI-enabled workflow orchestration, explainability, and integration with existing systems, are better positioned to improve access and close long-standing protection gaps in the middle market.

70% of vehicle collisions are caused by inattention, including distraction, cell phone usage, and fatigue. Most of this risk develops silently, unseen by fleet managers and insurers alike, only potentially becoming visible once it has resulted in a claim.

That reality exposes a growing structural weakness in commercial motor insurance. If the majority of collision risk forms upstream of loss, underwriting frameworks built primarily on historical claims data are, by definition, incomplete. In a market grappling with rising severity, social inflation, and earnings volatility, this gap is no longer theoretical. It is material.

The industry's next competitive advantage will come from seeing risk earlier and acting on it before loss occurs.

Risk begins long before first notice of loss

For decades, underwriting has relied on lagging indicators such as loss runs, experience modifiers, and aggregated exposure metrics. These tools remain necessary, but they describe outcomes rather than causes. In commercial auto, collisions are rarely random events. They are typically preceded by identifiable behavioral patterns interacting with vehicle dynamics and environmental conditions.

Until recently, those precursors were largely inaccessible to insurers. Risk could be priced after the fact, but not meaningfully influenced in advance.

From descriptive telematics to predictive intelligence

Early telematics solutions represented an important step forward, providing visibility into speed, harsh braking, acceleration, and location. These signals improved transparency and gave insurers better behavioral proxies, but they remained descriptive. They explained what had already happened rather than what was about to happen.

Predictive artificial intelligence fundamentally changes that relationship with time.

By analyzing multiple contextual signals simultaneously, including driver attentiveness, following distance, vehicle movement, and road conditions, advanced AI systems can identify elevated collision risk as it forms. Crucially, this intelligence can be acted upon in real time, alerting drivers in the critical seconds before a potential impact. While that window is narrow, it is often enough to change the outcome entirely.

At Nauto, for example, AI models trained on more than 6 billion miles of global driving data have demonstrated the ability to detect imminent collision risk with over 99% accuracy, validated through independent research. In practice, interventions typically occur two to four seconds before the triggering event. Those few seconds frequently determine whether an incident becomes a near miss or a loss event.

This represents a shift from measuring perceived risk to actively influencing outcomes, and it has profound implications for underwriting.

What this changes for underwriting

Predictive behavioral intelligence introduces a new variable into the underwriting equation, real-time risk quality. Rather than inferring exposure solely from historical outcomes, underwriters can now access leading indicators of attentiveness, distraction, and behavioral discipline.

This capability is particularly valuable in portfolios with limited claims history, rapidly evolving operations, or exposure to emerging risk factors where backward-looking data provides limited guidance. Pricing becomes more responsive and more defensible. Fleets that demonstrate sustained behavioral improvement can be differentiated with greater confidence, while persistent risk signals can be addressed earlier through pricing, terms, or targeted intervention.

The result is a move away from portfolio-level averaging toward a more granular assessment of how risk is actually created.

For MGAs, predictive intelligence enables prevention to be embedded directly into product design, aligning delegated authority with real-world risk outcomes. For brokers, it strengthens the advisory role by grounding renewal conversations in objective, forward-looking evidence rather than retrospective explanation. Across the value chain, assumption is replaced with observation.

The economics of prevention

Prevention does not simply reduce claim counts. It changes claim outcomes.

Across fleets deploying Nauto's predictive AI, collision frequency reductions of between 40% and 60% are consistently observed. Importantly, the effect does not stop there. When collisions are avoided entirely, claims disappear. When incidents do occur, earlier intervention often reduces speed at impact, point of contact, and loss complexity, leading to materially lower claim severity.

This dual effect, fewer claims and less severe claims, alters loss cost trajectories in a way traditional risk controls rarely achieve. Secondary costs decline alongside primary losses. Vehicle downtime is reduced, supply-chain disruption is minimized, litigation exposure falls, and operational volatility softens. Claims that do occur are resolved faster and with greater confidence when supported by contextual video and AI-derived insight, reducing frictional cost and uncertainty.

At scale, these dynamics stabilize combined ratios and improve capital efficiency. In a market where margin expansion through pricing alone is increasingly constrained, prevention offers a durable alternative grounded in operational reality rather than actuarial optimism.

From recovery to partnership

This shift reflects a broader evolution in the role of insurance. Predictive AI does not replace underwriting judgment or risk management expertise. It enhances them. It allows insurers, brokers, MGAs, and fleet operators to share a clearer, real-time understanding of risk as it actually unfolds, rather than reconstructing it after the fact.

The insurers best positioned for the next decade will be those who underwrite behavior rather than history, reward prevention rather than recovery, and treat intelligence as something to be acted on, not archived. In an environment where risk is increasingly dynamic and unforgiving, resilience will belong to those who can see loss forming and intervene before it ever reaches the balance sheet.

Prevention is no longer an aspirational ideal. It is becoming a defining capability, and increasingly, a prerequisite for sustainable underwriting performance.

The hybrid fronting model is gaining traction in the re/insurance market, driven by a shift toward deeper risk alignment and the rapid expansion of MGAs.

As of 2026, approximately 25 major fronting carriers now operate in the U.S., with new ones launching frequently. The surge is tied to the expansion of MGAs, the tightening of reinsurance capacity, the growing need for efficient capital, and the availability of insurtech-enabled data modeling. Gallagher Re reports that hybrid fronting carriers generated nearly $28 billion in gross written premiums by the start of 2025, a significant portion of the total $100 billion-plus MGA market. A 2025 TMPAA survey reveals that 19% of program administrators now use hybrid fronting models for their operations.

What is a hybrid fronting carrier, and why is this model taking off now?

DEFINING A NEW MODEL

Hybrid fronting carriers are fully licensed and regulated insurance entities that provide rated paper to MGAs and program partners while retaining a share of the underwriting risk on their own balance sheets. Their retained portion of risk is typically 5% to 30%, and rated insurance paper is provided to MGAs/MGUs, captives, and programs. Hybrid fronting carriers use reinsurance or alternative capital to cover the remaining risk and maintain strong capital partner relationships with private equity firms or insurance-linked securities (ILS) to fund growth and operations.

Unlike traditional fronting carriers, which pass all risk along to reinsurers, hybrid fronting carriers retain a meaningful portion of exposure. In doing so, they share the risk and the incentives with other stakeholders across the insurance lifecycle. That retained risk drives tighter alignment with reinsurers and capital partners and positions hybrid fronts as not just enablers, but committed participants in the value chain. In other words, hybrid fronting carriers have more "skin in the game" than traditional fronting models. Ultimately, this sharing of risk results in more disciplined underwriting and oversight.

The hybrid fronting carrier model is becoming a more attractive option in the MGA and specialist program space, especially for launching new, niche, or complex insurance products quickly.

KEY DRIVERS OF THE HYBRID FRONTING CARRIER

There are multiple drivers behind the rise of the hybrid fronting carrier model. At the top of the list is capacity. Traditional reinsurers have been reducing capacity in challenging insurance markets, such as California, Texas, and Florida, where NatCat claims, including hurricanes, floods, and wildfires, have led them to reduce their exposure to tail risks. Others are pulling out of these markets entirely. As a result, demand for alternative capacity is rising as traditional insurers shift risk appetite or exit markets.

Hybrid fronting carriers are also filling an expanding gap in the excess & surplus (E&S) insurance market by serving as a bridge between specialized, high-risk, or niche business produced by Managing General Agents (MGAs) and the risk-bearing capital of reinsurers.

Another key driver is the acceleration of MGA growth, not just in the U.S., but globally. MGAs are growing rapidly and seeking flexible partners to develop new products and expand capacity. Hybrid fronting carriers supply MGAs with the rated paper and shoulder some of the risk, enabling them to launch niche or specialist programs, like cyber risks or complex casualty lines, at speeds much faster than traditional carriers with legacy systems. They are essential for navigating "volatile and emerging risks" where traditional insurers may have pulled back. This is crucial to MGAs' success, where the first to market with an innovative product wins the race.

Alternative capital fills these gaps for E&S insurance organizations and MGAs left by shrinking traditional supply. For capital providers, such as private equity and ILS investors, the hybrid model provides an efficient, scalable way to enter insurance markets without partnering with a traditional, full-scale carrier. Additionally, it streamlines cross-border expansion for MGAs by managing complex local licensing and regulatory compliance. Private equity and ILS investors are interested in hybrid fronts as capital-efficient, lower-risk insurance platforms.

BENEFITS FOR ALL PARTIES

Who wins with this model? Basically, all parties involved benefit. Hybrid fronting carriers present significant benefits to MGAs, program managers, and reinsurers. A shared alignment of interests ensures better underwriting and oversight, faster market access, and greater capital efficiency.

Other key benefits include:

• Multi-year capacity stability: While traditional reinsurers are tightening capacity, hybrid fronts can access a broader investor base. MGAs can reduce their reliance on a small group of traditional reinsurers.
• Faster product launches and distribution: Hybrid fronting carriers provide the necessary AM Best or S&P rating and state licensing required for MGAs to write business immediately. This bypasses the multi-year process an MGA would otherwise face to become a standalone licensed insurer.
• Access to reinsurance and capital markets: Unlike traditional insurers, which can face rigid internal governance, hybrid fronts act as conduits to global reinsurance markets. This provides MGAs with a broader pool of capital to back specialized or niche programs.
• Higher transparency and potential for profit-sharing between stakeholders: Unlike traditional models, where data is exchanged in fragmented silos using PDFs or monthly reports, modern hybrid carriers are built on technology and use unified underwriting command centers, making them extremely transparent. Because the hybrid front shares in any losses, it has a financial incentive to perform the same level of due diligence as a standard carrier, including thorough analytic and exposure reviews.
• Lower cost base compared to traditional carriers: The best hybrid fronts operate as lean, technology-driven entities that delegate expensive operational functions to specialized partners.

Fronting carriers play a crucial role in enabling MGA distribution by providing regulatory paper, compliance frameworks, reporting mechanisms, and risk-sharing structures necessary to launch new programs.

TECH-FIRST HYBRID FRONTING CARRIERS WILL WIN

However, with this model comes new complexity. Hybrid fronting carriers often manage numerous MGA relationships, demanding coordination of complex capacity flows, diverse systems, multi-territory compliance, and increasing calls for reporting clarity. These operational and regulatory demands frequently exceed the capabilities of older, legacy carrier infrastructure.

To thrive in this environment, hybrid fronting carriers require more than rated paper, risk, and capital. They'll need modern infrastructure that can support:

• Real-time digital dashboards provide immediate visibility into MGA, program, and portfolio performance, enabling faster, data-driven underwriting decisions and proactive intervention on underperforming segments. Because hybrid fronts have their own capital at stake, they need this transparency to effectively manage their own risk.
• Data transparency for reinsurers, regulators, and capital partners alike fosters trust, ensures compliance, and strengthens long-term capacity relationships.
• Automated bordereaux processing and streamlined delegated authority workflows reduce operational overhead, boost accuracy, and ensure audit readiness. Further, using a single, shared platform eliminates the need for manual bordereaux reporting, giving hybrid fronting carriers better data and a single source of truth for their reinsurance partners.
• Integration with reinsurers, TPAs, and third-party systems that streamline operations, accelerate speed to market, and enhance collaboration across the entire value chain.
• Scalable platforms that accommodate multi-entity, multi-jurisdiction operations — supporting rapid growth across regions, lines of business, and regulatory regimes without losing control.

With hybrid fronting carriers, technology isn't a support function—it's part of their very structure as an organization. It's an enabler of profitable growth, faster onboarding, and regulator-ready reporting. Carriers that invest early in building this connected foundation will be best positioned to scale. And those that can scale will be the most successful as they court private equity and ILS investors while bringing innovative MGAs and risk products to market.

Many insurers are in an AI pilot purgatory, where promising experiments rarely scale into everyday operations. The models perform adequately, and the business cases hold up. The primary barrier is data architecture. Systems built for reporting and analytics simply cannot support the demands of production AI.

Core insurance decisions depend on synthesizing information from multiple sources. Underwriters assess application data, loss histories, external data, and regulations to evaluate risk. Claims handlers review photos, repair estimates, medical notes, and witness statements to settle cases. Investigators pull together scattered, sometimes conflicting information to pursue recovery. Transforming this expertise into AI capability requires data architecture that supports learning, generation, and contextual requirements.

Why traditional architectures fall short

Organizations have long separated day-to-day transaction systems from analytical warehouses. This division supported dashboards and compliance reporting effectively. However, AI blurs these boundaries because it learns from historical patterns to make real-time operational decisions.

When AI evaluates a new claim, it needs current policy data, historical loss patterns, regulatory requirements, and market conditions simultaneously. It needs real-time transactional data integrated with comprehensive historical context.

Unstructured documents create an even larger hurdle. Applications, claims notes, legal filings, and reports hold the most valuable intel for decision making. Many architectures treat this as a mere storage challenge. AI needs to understand documents at a much deeper level - identifying key elements, mapping connections, and pulling meaning in real time alongside structured records.

This matters most for complex workflows. When AI processes legal documents, filings, or investigation files, it can handle work that once demanded years of specialist knowledge. Document intelligence must sit at the same level as core transactional and analytical data in the architecture to enable this.

What AI-ready data architecture requires

Getting data ready for AI means building four specific capabilities that work together. Each closes a critical gap between how traditional systems operate and what AI applications need to work in production.

1. Bridging architecture and data silos

AI applications need access to policy systems, claims platforms, finance, and external data without long delays. This means real-time operational data alongside historical context, structured tables alongside document content, internal data alongside external feeds. This doesn't mean consolidating everything into one repository or platform. Focus on connecting data where it lives, with clear tracking of its path and origins. This architecture enables AI to navigate existing systems securely with proper lineage and control.

2. Capturing and using expert knowledge

Every time an underwriter overrides an AI suggestion or a claims handler adjusts an estimate, that action contains valuable knowledge. The capability to capture, curate, and organize expert feedback into training datasets separates competitive AI from generic tools. Raw corrections alone aren't sufficient. The architecture must support structured approaches that validate expert feedback, enrich it with context and reasoning, and organize it into training datasets that prevent bias while maximizing learning signal.

3. Managing context data for AI

Experienced underwriters or claims adjusters don't evaluate evidence in isolation. They build a growing understanding as new information arrives, drawing inferences, applying rules, and tracking reasoning. AI needs the same ability: to maintain and evolve understanding throughout a process. Context is this accumulated understanding that an AI system builds, stores, and shares as it works through a process. AI requires context as a managed data type with its own lifecycle, access controls, and transformation rules.

4. Creating data environments for AI development and testing

Moving AI from pilots to production deployment requires infrastructure that can provision realistic environments on demand. The data architecture must support replicating production data with appropriate privacy controls and generating synthetic data for edge cases. As AI programs scale across use cases and product lines, the ability to spin up multiple isolated environments lets teams work in parallel without interference. Provisioning environments quickly with realistic data, then tearing them down when complete, becomes critical infrastructure for scaling AI operations.

Competitive reality

AI will transform insurance operations. Organizations that address these data foundations build compounding advantages: faster decisions, greater accuracy, reduced leakage, and teams freed for higher-impact work.

Start with use cases where the business case is clearest. Focus on the data and capabilities those require, and build incrementally on current investments.

Thirty percent of companies will eliminate remote work this year, and 83% of CEOs globally expect a return to full-time office work in 2027, according to two recent reports. Many insurers will be among those heading back to the status quo pre-COVID. 

But a lot of employees are pushing in the opposite direction. They not only want flexibility on where they work. They want flexibility on when they work. 

We hear all the time about the hundreds of thousands of insurance industry employees reaching retirement age and about all the difficulties in attracting the talent needed to replace them, so I suggest we don't dismiss the desire for time flexibility out of hand. Yes, it runs counter to the management reflex that wants to bring everyone back to the office so they can be seen and managed as a cohesive group. But insurance desperately needs an influx of talent, and, as the saying goes, you attract more bees with honey than vinegar — or, more bluntly, beggars can't be choosers.

Clearly, many parts of the insurance process can't happen whenever an employee chooses to work. Agents and brokers need to be available, for instance, whenever a client needs them. But many underwriters and claims representatives could do their work based on a caseload, rather than on office hours, especially now that generative AI can track down so much of the data for them. 

Whether to offer more flexibility, not less, is worth a thought.

My interest in the topic of flexibility was piqued by a smart column by Matthew Fray at Quartz (which supplied the statistics I quoted in the first sentence). It says:

"Work-life balance has overtaken salary and compensation as the leading priority cited by 65% of office workers globally, up from 59% four years ago, said Peter Miscovich, co-author of the book The Workplace You Need Now, and the executive managing director and global future of work leader at JLL, the commercial real estate giant.

"Employees increasingly value control over when they work such as start and stop times, protected focus blocks, and predictable personal-time boundaries, more than additional workplace location choice, Miscovich said."

I realize I have a bias about flexibility, having worked remotely and pretty much on my own schedule since I left the Wall Street Journal in the mid-'90s. The productivity of a writer is also awfully easy to track. You either produce, or you don't. Even at the WSJ, where I worked office hours, if I went a couple of weeks without a byline, I might get a call that began, "Pa-uu-ll, this is your faaaaather. I'm just calling to make sure my son is still employed." (Thanks so much, Dad.)

But I do think flexibility attracts and retains top talent and is possible in many parts of insurance processes. I'm thinking, in particular, of claims and underwriting. An experience manager knows what a claims rep or underwriter should be able to handle, not just based on the number of cases but on their complexity, so it should be possible to let them work largely on their own time in their own place. I'm sure other processes can allow for at least some additional measure of flexibility, too.

People should still come to the office for socialization purposes. Training of newbies probably needs to be largely done side by side. And anything that requires frequent interaction between employees obviously needs to be done in the same place at the same time — Zoom eliminates some of the need for being in the same place but by no means all. 

I realize that, in many types of jobs, there's a fear that employees will slack off if they're not under close supervision, and that surely happens. But we also see how compulsive people can be about keeping up with their email and other work even during off-hours, so I'd bet some people — especially the talented and ambitious — would work even harder if motivated by more flexibility. 

I harken back to an interview I did with Scott McNealy, at the time the CEO of Sun Microsystems, in 2001. In the days before everyone had a laptop they carted to and from work, McNealy had spent quite a bit of money buying home computers for his 40,000 employees. He caught some grief for the expense but seemed to me to have a pretty good justification.

"I do not want somebody at 10 o'clock at night who can't sleep, who wants to work because there's nothing good on TV, to not have the full capability to do everything he needs to do to get the job done," McNealy said. 

Worth a try?

Cheers,

Paul

In recent conversations, Brian Poppe at Mutual of Omaha highlighted AI adoption in four stages – a transition from AI being a fun tool that employees use to a final stage where AI-driven processes are integrated for seamless workflows. And certainly directionally, this appears to be correct – AI use cases have historically been focused on addressing specific inefficiencies and then scaling, which will eventually lead to AI driven processes.

While there is an acknowledgment that AI is a rocket that will move insurance in a truly transformative way, carriers are treating AI as if you are putting a new engine in an old car: it performs better, but it is still driven the same way.

AI-driven processes raise an entirely different question – do legacy operating models make sense for insurance carriers in the age of AI? And the answer is – no, they do not.

When we think of an operating model, definitionally we should think of the way in which people, processes, technology, and capabilities are arranged within an organization to deliver value to customers. The evolution of AI will be AI assisting with a task to developing a process that is driven by AI. But operating models are derived from the assumption that processes are human-driven.

Consider the underwriting process and how the evolution of technology has changed it. Historically, you needed many underwriters to carefully review applications, assess the risk, and then provide a quote on pricing against that risk. RPA reduced the manual effort and increased productivity, but the process remained the same. Then automation and enhanced underwriting (e.g., algorithmic, usage-based, simplified, etc.) were implemented to provide faster underwriting, but the organization did not necessarily change to reflect these changes. Instead, carriers have viewed this from the lens of capacity and workforce management.

But if you were building a new insurance carrier today, would you structure underwriting in the same way that it is today? Most likely, no. And as AI evolves over time, you would certainly design a different operating model.

In other words, processes designed around AI and technology would require a quite different organization than human-driven processes. The more carriers lean into AI-driven processes, the more the legacy operating model makes little sense.

The Legacy Trap: Why Current Models Are Not Changing

If we accept that AI-integrated processes are directionally where the insurance industry is headed, then the question is why haven't carriers designed new models? There are several reasons why organizations are not evolving:

1. Organizational Resistance: AI-driven processes come with an uncomfortable question – what is the role of a human in this new environment? Most assume that it means that AI is "coming for their role," and to some extent, they may be correct. But that assumption hinges on two beliefs – that all capabilities can be automated and that all automated capabilities no longer require people. Neither of these beliefs is true.

2. Lack of Success With AI: There is an often-cited statistic that 95% of all AI projects do not make it from pilot to tangible, measurable ROI. This suggests that although carriers are investing in AI and understand its capabilities, they are not finding success at scale, delaying transition to AI-driven processes and capabilities. While this suggests that the AI-driven process may not be as close as some believe, it would be incorrect to dismiss it as hype. Executives and insurance leaders only need to be directionally right, and innovation in the space should be balanced with an AI strategy on what to invest in and how to prioritize.

3. Unproven Models: Insurance carriers are conservative – an op model built on new technology is a significant risk and has not aligned with traditional automation strategies. Typically, a process is automated and then resources are reallocated or modified once the investment has generated ROI. But there is evidence of carriers operating in dual environments with new operating models, in what some have called a "two highway" approach – a legacy environment for in-force business coupled with a new environment for new products. A new target operating model does not need to be an enterprise effort initially – it may be useful to design a different model in a specific business unit to run in parallel to assess strengths and weaknesses before eventually scaling it.

Building AI-Native Op Models: A Practical Framework

If carriers accept that integrated AI processes creating new workflows is the future, then part of the planning effort must be an exercise developing a new target operating model. As carriers seek assistance with developing these models, there are five key principles that will lead to the greatest chance of success:

1. Realize Directionality Is More Important Than Timing: Carriers do not need to know exactly when a transformation will occur, they only need to think in terms of where AI is moving directionally. Consider various capabilities in insurance. Operational support of the insurance model is likely headed toward significant automation of processes, while sales and marketing are likely to remain less automated in the future. From an operating model perspective, that likely means that AI driven processes will push workflow in the back office (think of new business submission or policy administration), while in the sales capability, you are more likely making the agent/advisor/broker more efficient (e.g., next best actions, generating marketing material with existing pre-approved templates).

2. Ignore Biases and Existing Requirements: One of the most difficult aspects of designing a new operating model in general is getting stakeholders to leave "the way it has always been done" at the door. Remember that this is a white space exercise and should be framed as such. For example, policy servicing should initially be thought of in terms of desired customer/agent experience, not how that service is delivered. When framed appropriately, carriers can focus on what they want to achieve and then assess how they would achieve it.

3. Understand the Hard Lines: For some carriers, there are hard rules that they will not consider. For example, risk appetite in underwriting may make some AI-driven processes impossible, or there may be a decision to create a large case workflow that is human driven to provide white-glove treatment for a particular agent class. Understanding enterprise non-negotiables upfront eliminates downstream decision-making on the op model.

4. Embrace Uncertainty: Carriers must understand they are blazing a new path forward. There is no cookie-cutter approach to a new operating model. While there are proven approaches, the result is that you may no longer have a clear benchmark. AI is introducing uncertainty and the only thing that we know is that it will transform the way that insurance carriers operate. The introduction of AI-driven processes will inevitably create a feedback workflow connecting actuarial product design, underwriting, and claims to create real-time adjustments to initial assumptions. The long-term consequences are unknown, but carriers still have to develop these capabilities to compete in the market.

5. Iterate, Iterate, Iterate: While there is directional design, understand that operating models evolve as new data is presented. While there are assumptions that sales (particularly personal lines) will continue to be driven through agents and brokers, significant change in customer dynamics or technology could change these assumptions. Additionally, end-state operating models make assumptions on where technology will be, not where technology is today. That may mean an agile approach to op model development.

The process of developing these operating models will not be instant. But carriers must begin the process of reassessing how they are organized to meet client needs in the age of AI. Digitally native carriers like MGT Insurance (organization built around AI stack to support small businesses) and Ethos (organization built around underwriting that can be done in five minutes) are already further along in this journey than legacy insurers, and the consequences may mean bloated organizations, reduced profitability, and an inability to compete in the marketplace, particularly in price sensitive markets. Embracing AI while ignoring op model transformation is only delaying the inevitable. As AI evolves, what assumptions in your op model might need rethinking?