When executives ask my team where they should deploy AI first, they usually point at the decision itself. Underwriting AI. Claims decision making AI. Lending decision making AI. The decision is the visible part of the workflow, and the cost of the decision is what shows up in the operating budget.

In my experience working with insurers and adjacent industries over 10 years, this is not where the highest-value early AI deployment lives. The decision is rarely the most time-consuming part of the workflow. The time consumption sits in the work that has to happen before the decision can be made at all.

I call this the data-collection tax. It is the time professionals spend gathering, reconciling, and verifying information before they can apply their judgment to it.

How the tax shows up

In claims, an adjuster spends most of the working day collecting facts, photographs, repair estimates, and policy details. The decision takes minutes. In underwriting, an underwriter spends the day collecting financial statements, loss histories, and exposure data. The decision takes minutes.

This pattern is not unique to insurance. It shows up wherever knowledge workers apply professional judgment to a defined question. Litigators collect facts and apply legal reasoning. Doctors collect symptoms, test results, and history and apply clinical reasoning. Risk analysts collect operational data and apply probabilistic reasoning. In each case, the collection time is the larger share of the workflow. The reasoning time is the smaller share, but it is where the value lives.

When my team published research from 20 US insurance leaders this June, we asked where the biggest gap was between how claims decisions get made today and how the leaders would want them made. The most common answer, named by close to two-thirds of respondents, was adjuster time spent gathering data instead of evaluating the claim.

That answer landed clearly ahead of every other operational complaint, including cycle time, decision consistency, and regulatory documentation effort. The bottleneck in their operations is not the decision. It is the work that has to happen before the decision.

Why AI fits here

The reasoning is operational and regulatory, and it generalizes across industries.

Operationally, AI handles structured but tedious data-collection work better than it handles judgment-intensive decision making work. Parsing a repair estimate in a PDF, extracting fields from a medical chart, reconciling a customer record across three systems of record, summarizing a deposition transcript, flagging a transaction against a watchlist. These are tasks where the input is unstructured, the output is structured, and the success criterion is verifiable. They are also tasks where the cost of an AI error is bounded. A wrong field extraction can be corrected by the human who reviews it. A wrong document summary can be checked against the source.

Compare this with AI deployed at the decision itself. The input is structured but contextual. The output is a judgment. The success criterion is contested. The cost of an error is high and often regulated. Most enterprise AI deployments that stall in pilot stall here.

Regulatorily, AI deployed on data-collection work attracts much less scrutiny than AI deployed at the decision. The NAIC Model Bulletin on AI in insurance, the EU AI Act, the NIST AI Risk Management Framework, and various federal and state-level expectations all focus their highest scrutiny on AI systems that make or substantially influence decisions affecting consumers. AI that pre-processes documents, intakes information, or routes work to the right human is generally outside the highest scrutiny tier. This means the deployment can ship faster, with less governance overhead, while still delivering measurable value.

Four categories of opportunity

Four categories of data-collection work cover most of the opportunity in most enterprises.

The first is document interpretation. Any workflow that requires a professional to read and extract information from unstructured documents has a data-collection tax. Policy documents in insurance. Loan applications in banking. AI can extract structured information in seconds, leaving the professional to verify it and apply judgment.

The second is intake and triage. Any workflow that begins with a customer providing information through multiple channels, in inconsistent formats, has a data-collection tax. First notice of loss in insurance. Application intake in lending. Patient onboarding in healthcare. AI can normalize the input, route it to the right specialist, and pre-populate the work queue.

The third is system reconciliation. Any workflow that requires manual reconciliation across multiple systems of record has a data-collection tax. Customer records across policy administration, billing, and claims in insurance. Patient records across electronic health records, billing, and pharmacy in healthcare. AI can perform this reconciliation continuously in the background.

The fourth is audit trail generation. Any workflow that requires documentation of how a decision was reached has a data-collection tax. Regulatory documentation in insurance and financial services. Clinical documentation in healthcare. Compliance documentation in any regulated industry. AI can generate the documentation as a byproduct of the decision rather than as a follow-on task.

Each of these has the same operational profile. The data foundation work is bounded. The output is verifiable. The regulatory exposure is low. The measurement is clear. The professional whose time is freed can spend more of the day on the higher-value work the organization is actually paying for.

A framework shift

The framework shift this enables is small in description and large in consequence. Most enterprises think about AI as automation of decisions. The more productive frame is AI as liberation of decision-makers. The decision itself is rarely the bottleneck. The data-collection tax is.

If you are sitting on an AI proposal that targets the decision itself, take a hard look at what happens before the decision. The faster, cheaper, less risky deployment is probably already in front of you.

Insurance industry leaders have barely achieved a tentative comfort level with artificial intelligence (AI) that talks (conversational AI), before being urged into AI that acts (GenAI and agentic AI). Tool use changes the operational profile immediately. Interfaces, permissions, logs, and change control start carrying real risk, and teams feel that risk fast.

Adoption pressure keeps climbing. According to McKinsey's State of AI in 2025 report, 62% of organizations are now experimenting with or piloting AI agents, and 23% report scaling agentic systems in at least one business function. The execution gap underneath those numbers is wide. Recent research from Adaline Labs notes that on τ-bench, a standard AI agent evaluation benchmark, well-trained language models succeed on roughly 25% of tasks, and the majority of those failures trace back to tool selection errors rather than execution errors.

Insurance, however, likes complexity, and therefore, adds a second constraint: Successful scaling of AI remains rare. Only 7% of insurers have brought AI systems to scale, and about two-thirds remain in piloting, according to Boston Consulting Group's Sept. 2025 report.

Part of this problem is due to unexpected results when AI is deployed "in the wild," so to speak. If used correctly, Model Context Protocol (MCP) can reduce the opportunity for AI to get out of its intended "box," and dull the sharp edge of errors or breaks produced when unanticipated switches get flipped. Five of the most common breaks include using ambiguous parameters that trigger unpredictable behavior, not implementing an MCP server to control tool calls, failure to log requests and responses blocking incident response, relying on unversioned interfaces, and missing timeouts and fallbacks that turn tool use into workflow friction.

Break 1: Ambiguous Parameters Trigger Unpredictable Behavior

Loose tool contracts invite ambiguity. Engineers define "address" in one system and "streetAddress" in another. Teams accept free-form strings where a workflow expects structured fields. Tool calls pass basic tests, then drift when real-world inputs hit the boundary. The Adaline Labs research found that parameter-level selection errors cascade to a wrong final answer roughly 62% of the time, citing the AgentProp-Bench benchmark for tool-using agents, and the wrong tool call is rarely the end of the failure.

Teams can stop this drift by making sure the fields in the response have clear, human-understandable names so the large language model (LLM) can parse them as needed. This is also true of the output values. Make sure the meaning of each value is clear, and don't rely on documentation or other external information to interpret match codes or other field values.

Action: Clearly name or label fields to ensure the LLM interprets match codes or other field values correctly.

Break 2: Implement an MCP Server to Oversee Tool Use

Tool-using AI generates requests at scale, but bad inputs rarely arrive one at a time. They arrive in bursts. A tool that accepts invalid payloads creates downstream damage in the form of partial records, misrouted work, duplicate entries, and manual cleanup.

Teams prevent this by "supervising" the LLM through the creation of an MCP server. Think of the LLM as an eager intern. Interns typically want to work hard, but don't have the experience to validate information independently. In the ideal scenario, your MCP server assumes the role of a gatekeeper, validating every request before passing it to downstream resources. Regulated workflows benefit from explicit controls, such as allowlists or authorizations for actions, strict parameter constraints, and business-rule validation where the tool touches risk decisions.

Action: Implement an MCP server to act as gatekeeper over tool calls while also validating requests and responses.

Break 3: Weak Logging Blocks Incident Response and Auditability

Trust collapses when teams can't reconstruct what happened. If you treat the LLM as you would any other employee, it needs to answer basic questions on demand, including who called what, with which parameters, what the tool returned, and what action executed next.

Employees need clear authorities and logs that support replay, audit, and performance analysis. Correlation IDs should tie tool calls to workflow events and downstream outcomes.

Action: Log every request and response with stable identifiers, timestamps, version tags, and access context. Store logs in a system that supports retention and retrieval requirements.

Break 4: Unversioned Interfaces Create Regressions that Stall Releases

Tool contracts evolve. Parameters change. Downstream systems break. Engineers lose weeks chasing regressions that started as "small" changes to an interface.

Versioning eliminates most of this pain. Teams should treat tool contracts like public APIs. Version tags, change notes, and deprecation windows keep consumers stable. Contract tests should verify behavior against expected inputs and outputs.

Action: Publish versioned tool definitions, enforce backward compatibility policies, and run automated contract tests in CI.

Break 5: Missing Timeouts and Fallbacks Turn Tool Use into Workflow Friction

Every tool call inevitably carries latency and failure risk. Networks fail, dependencies degrade, and a workflow that waits indefinitely builds backlogs and frustrates users.

Proactive teams can keep workflows moving by defining timeouts and explicit fallbacks. A fallback can route work to review, request missing fields, or return a structured "cannot complete" response that preserves workflow continuity.

Action: Set latency budgets for each tool, enforce timeouts, implement retries with backoff where appropriate, and define a deterministic fallback for every failure case.

What Teams Should Build Next

Tool-using AI succeeds when teams build execution infrastructure that behaves like software, supports governance, and scales across workflows.

1. Inventory Tools and Classify Risk. Separate read-only tools from side-effecting tools. Apply stricter constraints to side-effecting actions.
2. Write Strict Schemas and Enforce Validation. Treat schema design as a first-class engineering artifact.
3. Implement Traceable Logging and Replay. Store enough data to reproduce an incident and prove what happened.
4. Version Contracts and Automate Regression Testing. Protect downstream workflows from breakage.
5. Define Timeouts, Fallbacks, and Failure Semantics. Keep workflows moving during degradation.

MCP is intended to standardize how AI applications discover tools and exchange structured inputs and outputs. Teams, however, still need engineering discipline to keep tool use safe, repeatable, and resilient. Fortunately, MCP can also provide the interface layer that standardizes tool discovery and invocation.

Using MCP, teams create safety and scalability through contract discipline and operational safeguards. Tool use will move into production quickly, and teams that ship reliably will treat tool contracts, validation, audit trails, and change control as core infrastructure.

As more P&C insurers enter the rapidly growing pet insurance market, many assume they can extend existing platforms, processes, and operational models to support pet insurance as just another specialty line of business.

In practice, this assumption breaks down quickly. Pet insurance is not simply another P&C product. It represents a fundamentally different operating model, one that more accurately reflects health insurance rather than traditional property and casualty.

That distinction matters. While insurers may be underwriting pet risk, they are also being asked to deliver a continuing care experience. Traditional P&C products are built around episodic events. Auto claims happen occasionally. Home claims are infrequent. Even travel insurance is event-driven. Pet insurance is different.

It is defined by continuing, high-frequency interactions between pet owners, vets, and insurers. Chronic conditions, repeat prescriptions, regular check-ups, specialist referrals, and continuing treatment plans transform insurance from a reactive product into an active service. This is not a product extension problem. It is an operating model mismatch.

Traditional P&C platforms were designed for annual renewals, infrequent claims, standardized documents, batch processing, and slower workflows. Pet insurance requires continuous engagement, real-time decision-making, dynamic interactions across multiple parties, and high-volume, non-standardized inputs. Trying to run one on the other creates structural inefficiency.

Why the Traditional Model Breaks at Scale

Historically, reimbursement has been the dominant operating model for much of the pet insurance market. The customer pays the vet bill upfront. They submit paperwork to the insurer. The insurer reviews the invoice, checks coverage, validates the claim, and reimburses the customer days or weeks later. That model is becoming increasingly unsustainable.

Pet parents now expect the same level of convenience and immediacy they experience in other areas of their lives — from real-time payments and instant approvals to transparent decisions. Waiting weeks for reimbursement is no longer just inconvenient; it is misaligned with modern expectations. More importantly, it creates operational strain.

Vet invoices rarely follow standard templates. Different clinics use different terminology, coding structures, and levels of detail. Many insurers still rely on manual processes to interpret invoices, validate treatments, check policy terms, and identify potential fraud or leakage. That creates a long chain of administrative effort.

As volumes grow, this creates a structural problem: operational costs scale with volume. Manual effort increases. Delays become inevitable. Fraud and leakage become harder to detect, and insurers are forced into a false trade-off. Move faster and lose control. Add controls and degrade customer experience. In many cases, they end up losing both.

Pet Insurance Is Increasingly Operating Like Retail Health

Some insurers have already demonstrated what success looks like. The challenge is that much of the market is still trying to scale pet insurance on infrastructure built for fundamentally different products. That matters because the next competitive frontier in pet insurance increasingly resembles a health insurance model rather than traditional P&C.

In modern healthcare, eligibility is checked in real time. Treatment decisions are made at the point of care. Payments are integrated into the care journey. Data is structured, codified, and continuously analyzed. Pet insurance is moving in the same direction. Instead of reimbursement-heavy workflows, insurers are shifting toward real-time coverage validation, point-of-care payment models, integrated vet-insurer-customer interactions, automated adjudication, and structured data capture.

Delivering this is not about digitizing existing workflows. It requires a fundamentally different technology foundation. When invoices are digitized and normalized, it becomes easier to identify anomalies, understand treatment trends, manage fraud risk, and control claims costs. Decisions can be made faster and with greater confidence.

This is where the economics of pet insurance begin to change. Decisions move from reactive to proactive. Control improves without slowing down the customer experience. This is where the real competitive advantage emerges, not just in pricing, but in operational execution.

Why This Matters Now

Pet insurance is growing rapidly, but growth alone does not guarantee profitability.

As more P&C insurers seek to capitalize on the pet opportunity, many risk underestimating how fundamentally different the category is from their existing lines of business. Insurers that continue to treat pet as a standard P&C extension risk carrying forward too much friction, too much manual work, and too much hidden cost.

The firms that win will be the ones that recognize pet insurance for what it increasingly is: not a niche version of home or auto insurance, but a customer-centric health insurance product built around continuing engagement, continuous service, and point-of-care decision-making.

That requires a different mindset, a different operating model, and a new health-oriented, service-driven framework. The next generation of pet insurance will not be defined by who can process the most claims, but by insurers that can most effectively remove friction from the care journey, connect customers, vets, and insurers in real time, enable decision-making at the point of care, and operate efficiently at scale.

This requires platforms built for real-time interaction, continuous data exchange, and event-driven execution — capabilities that traditional P&C systems were never designed to support.

Insurers that continue to treat pet insurance as a simple extension of P&C will struggle to scale profitably. Those that embrace its evolution toward a health-driven, continuous care model will define the future of the market.

America has entered its Peak65 era. With a record 4.2 million baby boomers celebrating their 65th birthday in 2025, the number of people entering retirement age in the coming years is expected to continue rising.

Regardless of whether someone is retiring immediately or continuing to work, the moment demands a new approach to coming lifestyle changes. It's also a time when people reassess their risk tolerance to ensure their savings strategy can support an uncertain retirement horizon.

Annuities have long been a versatile financial tool for this demographic, providing predictable income and financial security at a time when heightened concern for stability is warranted. And demand in the market has followed suit, with last year's U.S. annuity sales surpassing $460 billion for the fourth straight year.

Yet despite the demand, consumers lack confidence in understanding their annuity plans.

For insurers, this demographic shift presents a unique opportunity to engage and educate clients, becoming a trusted source for their future financial security. To capitalize on the momentum, advisors must address the knowledge gap with personalization, transparency, and trust.

The costly impact of miscommunication

Turning 65 is a gateway moment when new financial responsibilities and freedoms emerge – and when high-stakes decisions can have long-term consequences.

But from initial education to continuing policy communications, many consumers are left with a limited understanding of how their annuity works – as well as how it fits into their broader financial plan.

Recent data underscores the scale of the issue. According to Smart Communications' 2026 Customer Experience Benchmark Report, only 50% of consumers rate the communications they receive from insurers as very good or excellent, a 10-point decline from the previous year. This drop becomes significant given that 86% of customers say communications are important to their overall experience with a company. With interactions between insurers and consumers often taking place during emotionally charged, stressful or life-changing moments, communications are not a secondary concern, but central to maintaining an open and trusting relationship.

When communication fails, it creates inefficiency, with 46% of consumers contacting customer support teams when communications are difficult to understand. Confusion can also lead to frustration or disengagement, with many abandoning processes altogether. In fact, 62% of consumers say they would switch providers if communications do not meet expectations.

Shifting from first impression to long-term engagement

As with many insurance products, misconceptions about annuities are not the exception for many clients. As a trusted advisor, it's important to understand where these gaps show up and how to address them.

For instance, it's a common assumption that annuities guarantee fixed lifetime payments, when in reality payouts can vary depending on contract structures or underlying investment performance. Misunderstandings like this highlight a broader issue: too often, information is explained in ways that are technically accurate but not easy to understand. It's here that a personalized approach, which lays out a client's plan and outcome, would drive far better sentiment and results.

Breakdowns in understanding can occur at any stage of the customer's journey, and when they aren't clearly addressed, confidence and trust begin to diminish:

• In the pre-purchase stage, communication often falls short in helping consumers understand what an annuity is and how it fits into their broader retirement strategy.
• At the point of sale, complexity peaks. Contract terms, fees, and payout structures are frequently presented all at once, creating an overwhelming amount of possibly new information at the exact moment a decision is required.
• In the post-purchase stage, communication often becomes more reactive than proactive. Statements, updates, and policy documents may lack clarity or context, making it difficult for customers to interpret performance, understand their income options, or feel confident in their long-term plan.

Rebuilding client confidence through better communication

To sustain growth and retention in the Peak65 era, insurers must fundamentally rethink how they communicate with customers. Communication is no longer a supporting function but a strategic differentiator, where clear, timely, and relevant interactions not only improve customer understanding but also directly reduce operational strain, increase engagement, and build the trust required for long-term customer loyalty.

Technology has been a key enabler of this shift, but only when applied with intention. Modern CRM systems, AI-driven tools, and digital onboarding platforms allow insurers and advisors to deliver more personalized, responsive, and consistent experiences. These capabilities enable meeting customers where they are and tailoring communications to their needs.

At its core, closing the gap requires simplifying complexity without losing meaning. Plain-language explanations and outcome-based messaging make products easier to understand. Consistency across channels ensures clients don't encounter gaps or repetition.

It also means simplifying where and how clients can access information about their plans. Reducing friction in every interaction, such as onboarding or accessing documents, helps eliminate gaps before they arise.

As millions more Americans reach retirement age, insurers should look to deliver clarity at every stage of the customer journey. In a market defined by uncertainty, the ability to turn complexity into confidence will be the defining advantage.

Much of the AI initiatives in insurance today appropriately focus on use cases such as underwriting, claims, and policy servicing. Those critical areas deserve attention and action. Yet there are equally important considerations in how carriers deliver implementations, launch products, and release changes into production within accelerated timeframes.

For many carriers, the hardest part is not identifying what needs to change. It is getting that change through the enterprise delivery cycle. New products, state rollouts, regulatory updates, and competitive responses all depend on complex and involved work that begins upstream, long before a system goes live.

This is where timelines compound, dependencies stack up, and delivery risk escalates with every handoff.

The pattern is familiar. Discovery runs eight to 12 weeks, and a typical core platform implementation can run 15 to 25 months before a carrier sees production value.

Each phase waits for the previous one to mature or complete. By the time delivery reaches testing, the original business intent has often been interpreted, translated, and reworked several times.

Agentic AI creates an opening to rethink that model, and agentic AI with your data remaining within your infrastructure is a significant advantage.

With AI agents, systems do not just assist with tasks; they participate across connected workflows in the software development life cycle (SDLC).

Instead of treating delivery as a sequence of separate phases, carriers can begin to move more work in parallel. Analysis, design, build, and testing no longer have to advance only in a straight line. As outputs become more connected, teams can start earlier, validate sooner, and reduce the lag between one phase and the next. The shift is not only about speed. It is about creating a delivery model that is more continuous, more responsive, and less dependent on serial handoffs.

This is how agentic AI helps de-risk the SDLC. When work products stay connected from the start, fewer issues get rediscovered late in the cycle. Changes can be carried through more systematically, and teams enter user acceptance testing (UAT) with stronger alignment across requirements, build, and testing. In practice, that means less rework, fewer downstream surprises, and more confidence as delivery moves toward production.

Of course, none of this works without industry context and enterprise readiness. Any agentic platform used in insurance must reflect how products are actually built, configured, and tested, with awareness of policy lifecycle behavior, underwriting rules, rating structures, regulatory requirements, and platform-specific models. Purpose-built platforms can bring the enterprise guardrails carriers need to use these capabilities at scale: security-first architecture, governance, observability, auditability, controlled access, and deployment models that keep data within the carrier's environment.

The early results are already showing what this can look like in practice. One tier-one carrier saw more than 30 detailed requirement documents produced in less than two weeks, work that would have taken 20 to 30 business analysts two to three months. The value is not only faster documentation, but earlier alignment across the work that follows.

Product launches move faster. State rollouts are accelerated. Platform implementations that once took years can reach production in a fraction of that time. More importantly, delivery becomes less fragile and better able to absorb change dynamically without resetting the entire timeline.

Carriers that rethink delivery first will move faster, reduce deployment risk, and keep pace with regulatory, product, and seemingly constant market changes.

Climate risk has become one of the most frequently discussed topics in insurance, but from where I sit as a climate scientist who has spent two decades studying the performance and use of climate models, including their ability to represent weather extremes, climate risk is still widely misunderstood. 

In the recent Send Infuse webinar, I joined the conversation to help bridge the gap between climate science and underwriting practice. What became clear is that climate risk is no longer a distant scenario or a theoretical modeling exercise. It is already shaping losses, challenging long‑held assumptions, and exposing structural weaknesses in how insurers make decisions.

The industry, from a scientific perspective, is being asked to do something incredibly difficult, which is to try and understand climate change impacts in financial material terms. This translation of climate to financial risk involves the use of tools designed to represent past relationships between hazard and loss. These tools are now applied to explore a future that may look very different from today. Can these tools be informative about future losses? This tension sits at the heart of today's climate-risk challenge.

Climate Losses Are Surpassing Expectations, and It's Not Just Exposure Growth

Recent events illustrate the scale of the shift. Italy's 2023 floods, France's 2022 hailstorms, and the Los Angeles wildfires of early 2025 all produced losses far beyond what models anticipated. These aren't isolated anomalies, and they reflect a pattern we see across multiple regions and perils.

The science tells us hazard behavior is changing. In many places, we are observing nonlinear shifts — including threshold effects, intensification patterns, and clustering of extremes — that traditional catastrophe models were never designed to capture. In a transient climate, it is reasonable to regularly reassess models to assure users that they are still fit for purpose.

This is why I often emphasize that climate risk cannot be reduced to a new data layer or a simple model adjustment. It requires a deeper understanding of how physical systems behave under stress, and how those behaviors diverge from historical norms.

Why Traditional Models Struggle with Climate Scenarios

One of the most common misconceptions I encounter is the belief that we can simply "climate-adjust" existing catastrophe models and use them to explore future scenarios. But catastrophe models were built for a specific purpose: to price financial loss. They are essentially statistical models calibrated to a current reality — in terms of hazard but also socio-economic conditions that shape building standards, defense infrastructure, disaster response readiness, and ability to afford insurance, etc.

Climate models, or ensembles of climate models, by contrast, are designed to explore a plausible uncertainty space (socio-economically, and physical model response). Importantly, the socio-economic scenarios are not assigned probabilities or likelihoods. When regulators ask insurers to assess long-term climate scenarios, there is a clash of scientific perspectives. The mindset of insurance is focused on estimating a likely loss using predictive tools. From a climate science perspective, approaches and tools are designed for understanding system response. By exploring the plausible, firms can assess their ability to meet different types of challenges should they eventuate. If deriving a single estimate of future loss, how meaningful is this value for risk management when in fact many others are plausible?

This is why I have said that "the tools are not developed to be applied in a climate scenario approach where we don't understand how losses might occur." It is reasonable to assume that near-term projections are serviceable, but the further we look into the future, the more cautious we must be about the assumptions implicitly embedded in our tools.

The insurance sector is now at the stage where many other industries have found themselves. They are using familiar tools to answer unfamiliar questions. Eventually, organizations realize that the questions themselves need to change to seek information that is, if not predictable, at least actionable. This implies a shift from a predictive mode to an anticipatory mode of management for the far-forward time horizons.

The Real Challenge Is Governance, Not Just Analytics

While the scientific limitations are real, the operational challenges inside insurance organizations are equally significant. Climate-relevant insights often fail to travel through the organization in a meaningful way. Underwriters may be doing sophisticated work such as running local flood maps, tracking wildfire accumulation, and monitoring aggregates, but that insight rarely reaches reserving, capital, or the board. This is not a failure of effort. It is a failure of structure.

Insurance organizations are highly optimized for efficiency. Information flows sequentially, not collaboratively. Each function performs its role well, but climate risk requires a different kind of conversation — one that cuts across silos and acknowledges uncertainty rather than seeking to reduce it. We need to connect the chain links together to talk about how we do things in a more meaningful way. If we don't, climate risk becomes fragmented, and the organization loses the ability to see the full picture.

Climate Risk Is About Resilience, Not Just Pricing

One of the most important shifts insurers must make is recognizing that climate risk is not solely a pricing problem. It is a resilience problem. And resilience operates across multiple time horizons.

Near-term risks may be manageable with existing tools, but long-term risks require different approaches to monitoring climate drivers, understanding adaptation pathways, and then recognizing that exposure, hazard, and vulnerability will all evolve. It is unrealistic to assume that building standards, land-use planning, or national adaptation strategies will remain static. Yet many scenario exercises implicitly make that assumption.

This is why I encourage firms to think beyond model outputs and ask: What aspects of risk are we capturing? What aspects are we missing? And what decisions are we trying to inform?

A Path Forward: More Cross-Fertilization, More Curiosity

The insurance sector is early in its climate-risk journey, but it is not alone. Sectors such as water management, infrastructure planning, and emergency response have been grappling with deep uncertainty for years. There is much to learn from their approaches, particularly around decision-making under uncertainty.

We need more cross-fertilization between climate adaptation communities and the insurance sector. We need more conversations that bring together scientists, underwriters, actuaries, and risk managers. And we need leadership that recognizes climate risk as a strategic issue, not a compliance burden.

Climate risk is more than a model output. It is a shift in how we understand the world, how we interpret uncertainty, and how we make decisions in the face of change.

The Bottom Line

Climate risk now lives inside the underwriting workflow because the physical world is changing faster than our tools and structures were designed to handle. The challenge is not simply to refine models, but to rethink how information flows, how decisions are made, and how uncertainty is embraced rather than ignored.

As I said during the Infuse webinar, “climate risk is more than data layers and output from catastrophe models. It’s about understanding what risks might look like at different time horizons.” That understanding requires curiosity, collaboration, and a willingness to evolve.

The insurance sector has the expertise and the ingenuity to meet this challenge. It now requires new conversations and new ways of listening.

For an industry built on the precise calculation of risk, the persistence of claims leakage presents a striking paradox.

Defined as the difference between what a carrier actually pays and what it should have paid under the terms of a policy, leakage is not a marginal concern. Across the industry, it typically accounts for between 7% and 14% of total claims payouts, against insured losses that regularly exceed $100 billion annually.

But these potential financial exposures only tell part of the story. Claims leakage also corrodes the policyholder relationship at its most critical moment. There's an old expression in insurance: "pricing brings customers in the door; the claims experience keeps them there."

In a personal lines market where consumers have genuine options, and where bundled coverages make a dissatisfied policyholder an attractive acquisition target for competitors, the cost of a poor claims experience extends well beyond the individual transaction. Overpayment is one problem. Losing a customer who felt underserved by a confusing or inconsistent process is quite another.

Addressing leakage begins by distinguishing between its two forms.

• Hard leakage involves payouts where coverage did not exist: misinterpreted exclusions, undetected fraud, or losses simply not covered under the policy.
• Soft leakage is more diffuse: unvetted vendor invoices, missed deductibles, duplicate payments, and claims settled without adequate documentation. Both are damaging, both are preventable, and both have the same underlying causes.

The Claims Leakage Troika

Leakage is rarely the product of deliberate action. More often, it results from operational inefficiencies that compound across three connected domains.

1. Process Gaps

Inefficient claims processes create the conditions for financial leakage before an adjuster ever reviews a file. Missing or incomplete First Notice of Loss (FNOL) documentation sets a claim on the wrong trajectory from the outset; incomplete data cascades through the life of the claim, making accurate reserving and coverage determination harder at every subsequent stage.

Duplicate payments, processing the same service, invoice, or repair twice, remain a persistent problem in organizations relying on disjointed or manual payment systems. They are often discovered only in retrospect, if at all. Equally costly is unpursued subrogation: when third-party liability goes unidentified, the carrier absorbs costs it was entitled to recover. The failure is not always one of knowledge; more often it reflects workload constraints that push subrogation opportunities down the priority stack until the recovery window closes.

Prolonged claims management cycles compound the problem in another direction. Delays stack caseloads, generate non-productive administrative costs, and create pressure that pushes adjusters toward expediency, rather than accuracy. Process inefficiency, in this sense, is not just a cost driver; it is a leakage accelerant.

Vendor and supplier management introduces a parallel set of process vulnerabilities. Without pre-negotiated contracts or a carefully cultivated repair network, cost variances can go unchecked and overpayment often becomes the path of least resistance. Even where vendor agreements exist, administrative staff (and in some cases, claims adjusters) are often left manually processing invoices and attaching copies to individual claim files, adding another layer of inefficiency to an already strained operation.

2. The Human Factor

Claims adjusting is, by regulatory mandate in the majority of U.S. states, a human decision-making function. This regulation is concrete recognition that coverage determinations carry weight for policyholders. Human decision-making under sustained pressure increases the likelihood of claim overpayments.

This form of workload stacking, driven by staffing constraints, high caseload volumes, and persistent recruitment challenges across the industry, forces adjusters into assuming a reactive posture, rather than an analytical one. Among the most recognizable symptoms of this is stair-step reserving: the gradual, incremental upward adjustment of claim reserves without the structured analysis that should accompany each change. Reserves set too low, then corrected too late, distort loss ratios and undermine financial planning across the portfolio.

High talent turnover rates exacerbate this problem. Less experienced adjusters, saddled with intensifying claim file management responsibilities, are more likely to falter when executing error-prone tasks, expanding the risk of misapplied coverage terms, overlooked exclusions, or claims settlements executed without comprehensive documentation review.

Turnover also forces adjusters into unfamiliar lines of coverage: for example, a casualty-focused adjuster suddenly handling a surge of property claims, thereby adding to the ever-present risk of claim mishandling and overpayment.

Beyond these operational pressures, inconsistent decision-making injects a subtler form of friction. As adjusters apply subjective interpretations to materially similar claims, they introduce what behavioral economists call "noise" into the system, a form of variance unrelated to the merits of individual claims but instead tied to human inconsistency under pressure. Unlike stair-step reserving or reserve drift, this form of leakage does not leave an obvious trail, making it among the most difficult cost inefficiencies to detect and correct – especially at scale.

None of these outcomes reflect individual failure. They instead expose structural weaknesses: inadequate staffing, insufficient training investment, and the compounding effects of a talent pipeline that has historically not kept pace with demand. These are fundamental management and resourcing challenges, not adjuster accountability problems.

3. Technology Misalignment

Technology can be a potential remedy. But when poorly deployed, it can exacerbate leakage. Legacy claims systems that don't integrate seamlessly with modern data environments leave adjusters manually processing PDFs, photographs, and demand letters without the benefit of system-wide cross-checking. When claims, policy, and payment systems operate in "silos," enforcing compliance in real time becomes nearly impossible, and these gaps between systems open the possibility for serious financial consequences.

Effective fraud detection is another casualty of technological misalignment. Identifying sophisticated fraud, including inflated claims, duplicate billing across carriers, and staged losses, requires pattern recognition across large datasets. Without adequate analytics capabilities and well-developed fraud risk assessment processes, these claims pass through undetected and are absorbed as legitimate losses.

Carriers have increasingly turned to modern, automated workflows to address these vulnerabilities. But the fundamental limitation of technology in this context is well understood by experienced claims leaders: systems are only as effective as the quality of data that powers them and the judgment of the people interpreting their outputs. Automation accelerates decisions, but it does not inherently improve them.

No Shortcuts

AI does not "solve" claims leakage. Experienced people, working from complete and accurate data, using well-designed processes, do.

The insurance industry is in the early stages of significant technological change. AI and machine learning are reshaping claims processing, delivering faster triage, better anomaly detection, and more consistent documentation review. These are real capabilities. The risk behind their implementation lies in treating technology adoption as a discrete goal rather than a component in a broader operational strategy.

The downstream effects of misaligned claims handling are significant. Inconsistent outcomes invite regulatory scrutiny for inequitable treatment of policyholders, creating fertile ground for litigation. Indemnity and defense costs in contested claims continue to climb, and bad faith exposure, where a carrier's handling process is found to have produced an unfair outcome, carries penalties that can dwarf the original claim value.

Critically, regulators and courts do not distinguish between decisions made by a human adjuster and those influenced or generated by an automated system. For this reason, a chain of accountability is first, last, and always human-led.

Standards published by the NAIC, the Insurance Crime Bureau, and other regulatory and advisory bodies provide a framework for consistent, compliant claims handling. But frameworks only deliver results when they are deliberately embedded into daily practice. Standardizing decision-making criteria, calibrating adjuster judgment through structured peer review, and maintaining traceable audit trails are operational disciplines that leverage compliance guidelines to reduce revenue leakage and improve the customer claims experience overall.

Technology extends the capacity of skilled adjusters. It does not replace them.

The carriers best positioned to reduce leakage are those that approach the challenge as a compound problem demanding multifaceted solutions. Adjuster training that builds genuine coverage expertise, not just process familiarity, is irreplaceable. Technology investments that connect systems, surface anomalies, and reduce manual burden extend the capacity of skilled adjusters rather than attempt to replace them. And the organizational discipline to balance judgment with pace, to resist the pressure to close files quickly at the expense of accuracy, requires leadership commitment that no software platform alone can provide.

Claims leakage is not a new problem. What has changed is that the industry has developed a deeper understanding of its key drivers, true costs, and the tools and techniques that make reducing leakage achievable. These results come when people, processes, and technology are aligned toward the same goal: paying precisely what is owed to preserve the financial capacity to serve more policyholders at their moment of greatest need.

Many insurers are rushing to deploy AI before they have taken the time to ensure they can trust what it will say.

The industry is sitting on decades of expertise, but it is scattered. It lives in underwriting guidelines that have been updated over time, claims notes written by different adjusters, engineering reports, inspection writeups, loss history, and a lot of individual judgment that never made it into a system at all. Then, we layer AI on top of it and expect clean, confident answers.

Because of this, users are hitting massive roadblocks when the underlying knowledge is inconsistent, disconnected, or hard to interpret. What many still don't realize is that AI is simply not designed to fix human errors stemming from baseline data inputs. In fact, it only makes it easier to surface those incomplete or misleading answers faster. And in insurance, those answers can show up in major underwriting decisions, claims handling, client advice, and risk recommendations, which can have a negative domino effect on critical decision making in a highly regulated industry.

That said, the issue is not whether AI is "ready." The question is whether insurers have done the groundwork to make their knowledge usable and trustworthy in the first place.

A recent initiative at FM is a solid example of this foundational work in practice. The company enabled more than 1,500 engineers to access decades of engineering knowledge through an AI-assisted platform. The real impact was consistency and confidence, with speed as an added benefit. Engineers could get to the right answer faster, but more importantly, they could trust what they were seeing.

This is a persistent challenge that exists across insurance. Whether it is underwriting appetite, claims precedent, loss control recommendations, or how you respond to a client in a complex scenario, the success of AI depends entirely on what lies beneath the surface.

Here are three things insurers should consider before these tools start influencing real work:

1. Stop organizing knowledge for storage. Start organizing it for judgment.

Most knowledge environments are built like filing cabinets, which works until someone actually needs to make a decision.

An underwriter is typically not thinking, "let me go find one document." They are weighing signals. Prior experience, similar risks, exceptions, engineering insight, loss trends. The same goes for claims, risk engineering, etc. The work is contextual versus linear.

If your knowledge is stored in disconnected places, AI will pull from it in disconnected ways. The answer might look polished, but it will not reflect how real decisions get made.

What worked in the FM deployment is that they did not just make content searchable. They connected it in a way that matched how engineers think about risk. That required a different level of effort.

Insurers need to take the same approach. If you want AI to support underwriting or claims, then your knowledge has to be structured around real decision paths that move beyond document repositories.

2. Assume your answers will be wrong and build for that reality.

This is where a lot of organizations get too comfortable too quickly. There is an assumption that if you are pulling from approved content, the answers will align. That is not a safe assumption once you are generating responses dynamically.

In insurance, being "mostly right" is not good enough. If an AI tool points an underwriter in the wrong direction, or gives a claims professional an incomplete answer, the downstream effect is real. It affects decisions, consistency, and in some cases client outcomes.

The teams that are getting this right treat accuracy as something that has to be earned every day. They test outputs, create feedback loops, and involve subject matter experts in refining results. They also assume gaps will exist and design for continuous improvement.

At FM, trust came from validating how their system performed in real-world scenarios and improving it over time.

That is the mindset insurers need. Not "is this ready to launch," but "how will we know when this is wrong and how fast can we correct it."

3. Decide up front where AI stops and people take over.

A lot of governance conversations sound good on paper but fall apart in practice because they're too vague. In insurance, you need clear, concrete lines to protect the integrity of the work. Where is AI allowed to assist? Where is human judgment required? What needs to be reviewed before it is used in a decision? What level of confidence is acceptable?

If an AI tool is helping draft an initial view of a risk, that is one thing. If it starts affecting pricing, coverage interpretation, or claims outcomes without clear oversight, that is something else entirely.

The strongest implementations treat AI as an extension of the professional. That means building in transparency, showing where information came from, and making it easy for people to challenge or verify what they are seeing.

The closer these tools get to core insurance functions, the tighter that control needs to be.

In summary, there is a lot of energy around AI right now, and for good reason, with tremendous opportunity on the horizon. In parallel, industry rigor should never be sacrificed for speed, and demands an established, accurate data foundation prior to deploying AI tools.

The firms that will get this right are not the ones rolling out the most tools the fastest. They are the ones doing the less visible work, taking the time to structure knowledge properly, putting real discipline around accuracy, and being explicit about where AI fits into decisions and where it does not.

Done right, AI becomes a force multiplier for your best people, makes expertise more accessible, and makes decisions more consistent.

If you skip it, you are just introducing a new layer of risk into an industry that already understands how costly that can be. That is the tradeoff. And it is one insurers need to think through carefully before they deploy at scale.

AI is reshaping how insurance organizations operate, and the impact on the workforce is already visible. Submission intake, loss run processing, claims setup, and certificate issuance are all work that once required significant manual effort and is increasingly handled by AI.

For insurance leaders, that shift raises a question that doesn't get enough attention: what does it mean for the people doing that work today, and what does the team look like on the other side of this transition?

How AI Is Redesigning Insurance Teams

Carriers implementing AI are reporting real operational gains. But the more consequential change isn't in throughput numbers. It's in what AI is asking employees to do differently. As AI takes over repetitive, preparation-heavy work, the nature of insurance jobs is changing. Roles are consolidating in some areas and expanding in others. New responsibilities are emerging that didn't exist five years ago.

Insurance is not a sector where automation simply eliminates positions. The industry faces a significant and continuing talent shortage, and the work that remains after AI handles the routine tasks requires exactly the skills that are hardest to replace, such as judgment, relationships, and contextual expertise. An employee who previously scanned and indexed incoming mail can be retrained to manage agent relationships or work toward becoming an underwriter. The role doesn't disappear. It evolves into something with greater value and long-term potential.

What that evolution requires is careful planning. Leaders who treat workforce strategy as a byproduct of their AI rollout will find themselves reacting to disruption rather than managing it. The staffing model has to be built alongside the technology strategy, not after it.

Planning for the AI-Driven Workforce Shift in Insurance

AI is fundamentally changing how insurance companies operate. That's why workforce planning can't be an afterthought to AI strategy. It has to run alongside it. Leaders need to think through what new or redefined roles are required, how many resources each will take, who needs training or support, and what triggers staffing changes as automation matures.

Work through the steps below in sequence. Early steps focus on assessment and transition management. Later steps shift toward building the talent infrastructure your organization will need as AI becomes embedded in daily operations.

Step 1: Conduct a role impact assessment

Start by mapping which tasks AI will take over, which it will assist with, and which it won't touch. Categorize roles accordingly as fully automated, partially assisted, or unaffected and be specific about what that means for the people in them.

This isn't about deciding who stays and who goes. It's about understanding what the insurance workforce looks like after AI is in place, so you can make informed decisions about redeployment, retraining, and hiring.

Step 2: Define your transition timeline and pause non-essential hiring

Set clear rollout milestones, identify which manual processes will be phased out and when, and determine at what point staffing changes are likely to occur. Freeze hiring for roles likely to shift before a replacement is made. Use that window to re-scope open positions and determine whether temporary staffing can bridge the gap while AI implementation matures.

Ambiguity about timing creates more anxiety than the changes themselves. A clear timeline gives employees and managers something concrete to plan around.

Step 3: Communicate early, transparently, and with resources to back it up

Tell your teams what's changing, what isn't, and what the plan looks like before the rumors fill in the gaps. Employees don't resist AI because they're opposed to technology. They resist because they're uncertain about what it means for them.

Share timelines, explain how roles will evolve, and provide resources that help employees address their concerns directly, including access to career counseling, answers to common questions, and a clear point of contact for follow-up. Transparency without support is just announcements, and both are required.

Step 4: Crosstrain and redeploy ahead of the curve

Train staff for adjacent roles and build AI-specific skills including data validation, output review, and exception handling before those skills become urgent. Upskilling existing employees is almost always faster and less disruptive than hiring externally for roles that require both insurance knowledge and AI fluency.

Starting that transition early gives employees time to grow into new responsibilities rather than being pushed into them under pressure. The organizations that handle this well aren't waiting for automation to force the conversation. They're building the capability now, while there's still room to do it thoughtfully.

Step 5: Establish human-in-the-loop protocols

Assign staff to monitor AI outputs, manage escalations, and maintain compliance during early AI adoption. Human-in-the-loop oversight isn't a workaround for immature technology. It's a deliberate design choice that keeps your experts in control of decisions that matter while AI agents handle the volume work.

As model accuracy improves and teams build confidence in AI performance, the volume of exceptions requiring human review typically decreases. But getting there requires intentional oversight in the early stages, and the people doing that work need clear protocols and defined decision authority.

Step 6: Review licensing and regulatory requirements before roles shift

Redeploying staff isn't always as straightforward as rewriting a job description. In insurance, certain functions including claims adjudication in particular carry state licensing and certification requirements that follow the work, not just the title.

Identify which positions are governed by these requirements early, and factor compliance into any transition timeline. Incorporate continuing review of licensing, certification, and regulatory obligations into workforce planning so that all AI-enabled functions continue to meet jurisdictional standards as responsibilities evolve.

Step 7: Redesign job roles and career paths

As automation takes over document-heavy and rules-based work, job descriptions need to reflect what employees are actually responsible for now. Revise them to emphasize strategic tasks, AI collaboration, and the new responsibilities emerging from automation.

Define clear career paths so employees understand where their roles are heading and what development is available to them. This step matters both for retention and for accountability. If the job description still describes work the AI is now doing, you've created confusion about expectations, performance standards, and growth.

Step 8: Align staffing strategy with your long-term AI roadmap

Make sure staffing plans support the company's evolving AI strategy, automation goals, and customer experience priorities together, not just the technology rollout in isolation. Use your AI adoption roadmap and industry trends to anticipate the skills and capabilities your organization will need 12 to 24 months out across departments.

Coordinate staffing planning directly with IT and transformation teams so that AI tools and human skills develop in parallel, not in sequence. Roles focused on AI oversight, responsible use, and cross-functional coordination are already emerging at leading carriers. Building a talent pipeline for those positions now is far less costly than filling them reactively later.

Step 9: Build continuing reskilling into the operating model

Workforce transformation isn't a one-time training initiative. Implement structured, continuing reskilling opportunities focused on digital fluency, AI literacy, and adaptability that keep pace with how AI capabilities and workflows continue to change.

Invest in internships and rotational programs across functions to create a continuous pipeline of talent prepared for emerging roles. Quarterly skill refreshes, real-world scenario exercises, and internal knowledge-sharing between teams reinforce a culture where continuous learning is part of the job, not a response to disruption.

AI adoption also offers a concrete opportunity to strengthen diversity, inclusion, and succession planning. By intentionally retraining individuals from underrepresented groups for emerging roles, organizations can build more inclusive leadership pipelines while developing the talent the business genuinely needs. These goals aren't in tension. They reinforce each other.

Step 10: Develop AI oversight and governance roles

As automation scales, organizations need people whose job is to make sure it's working correctly, fairly, and in compliance with regulatory expectations. These aren't IT roles, they're operational ones.

AI workflow owners monitor agent performance and manage exceptions. Responsible AI officers oversee fairness, explainability, and audit readiness. Change and workforce transformation leads maintain employee engagement through continuing transitions. Creating these roles proactively rather than reactively is what separates organizations that scale AI confidently from those that stall.

Step 11: Monitor weekly, document continuously, and adjust based on reality

Track AI versus human workload on a weekly cadence and refine the plan based on what's actually happening, not the original projections. Record role changes, skill gaps, and planning insights as they emerge.

That documentation serves two purposes. Since early AI deployments rarely perform exactly as planned, it guides immediate staffing decisions and becomes the institutional knowledge that makes the next AI deployment faster and more predictable than the last.

The good news is that most insurance organizations still have time to get this right. According to Bevaya's State of AI Adoption in Insurance 2026 Report, 93% of insurers are now actively engaged with AI, and more than half are either testing or running AI in production.

The industry is moving, and that momentum creates a real opportunity for leaders who build their staffing model now, while AI deployment is still maturing, to get ahead of the workforce shifts rather than react to them. Companies that plan early will move through each phase with greater confidence, stronger retention, and less disruption than those who treat workforce planning as something to sort out later.

Balancing Today's Needs with Tomorrow's AI-Enabled Insurance Workforce

AI adoption in insurance is not a future event. It's happening now, at scale, across every major line of business. The carriers and TPAs that come out ahead won't just be the ones that deployed AI first. They'll be the ones that brought their workforce with them, retraining people before the pressure hit, redesigning roles before confusion set in, and building the governance and oversight structures that make automation sustainable.

Insurance has always been a people business. That doesn't change because AI can now process thousands of documents in the time it once took to process 10. What changes is what people are asked to do with their time, their expertise, and their judgment. Leaders who plan for that shift now, with the same rigor they bring to their technology investments, will build organizations that are faster, more resilient, and better positioned to deliver on the promises the industry is built on.

Transformation projects, especially when they're "side of desk," can seem like an unwelcome responsibility and a distraction from the core work of managing a function, yet for operational executives these are frequently an opportunity for senior leadership to evaluate suitability for more strategic roles. More often than not these projects involve a requirement to re-frame well-known challenges, to communicate a vision that sticks, and to demonstrate a capability to see across the business rather than purely within one's own vertical. 

Modeling as a means of facilitating shared understanding among different stakeholders is a key tool for insurance leaders to master.

When we think of models, we might bring to mind financial forecasts or visualizations, but a model can also take the form of an analogy, a prototype, a system map, or even a road map or a project plan. We define a model as a representation of the structure and underlying rules of a concept (or set of concepts). Modeling involves mapping an idea's structure onto a form that makes its defining conditions - the underlying rules that make it possible - easier to see, explore, or test. Moreover, insofar as a model is an abstraction, it can also help to bridge communicative divides between stakeholders with different contexts or different perspectives, thus making it easier to arrive at a shared understanding - which is essential for senior leader effectiveness. Models are tools that bridge mindsets.

Project management is people management

Over the past few years, I've worked on complex transformation programs in which securing alignment on objectives and purpose was as critical as agreement on how to execute. A recent example highlights what may be a familiar challenge for many executives. Our client, Alison, was caught in the middle of a project management nightmare. She had a reputation for successful IT delivery, but the majority of these projects had been focused within her function. Now she was involved in overseeing a multi-departmental program, coordinating the replacement of a legacy underwriting platform. On the face of it, this would appear to have been a simple re-platforming exercise, except for the fact that over time it had become a fundamental source of data for a range of different systems across the enterprise, from policy and CMS systems through to the master ledger. Replacing this system involved changes across the enterprise, and as such, required understanding and buy-in from stakeholders who had limited awareness or meaningful incentive to engage.

The program had executive-level support, which helped open doors and secure time with platform owners and the business, but as the initiative moved from concept to an execution plan, misalignment emerged between finance, operations, and IT leaders. Each stakeholder had different expectations:

• Finance pressed for outcomes that would require custom development beyond out-of-the-box capabilities of the recommended platform replacement;
• Forecast efficiencies were contested by operations leadership, affecting the program rationale and budget;
• IT's preferred vendor shortlist differed from operations' panel - and the latter had engaged their own leadership team to add pressure - raising concerns about impacts to scope, cost, and schedule.

We'll return to Alison's case, but when faced with this level of disconnect, our instinct can be to go on the offensive and tackle differences in perspective as skirmishes to negotiate. Another way of considering this situation, however, is to frame it as perceptual misalignment. In short, each party was interpreting the problem from their own point of view, informed by their own context, assumptions, and expectations. Bringing that to the surface - both for yourself and for your counterparty - can be helpful to isolate legitimate points of difference and exploring how to overcome these. Models are an effective tool to help us achieve this.

Why we see the same thing differently

Each of us perceives the same external environment, but we interpret it through our own lens - personal, for instance, but also institutional, such as different teams or functions that set our context - and these unique lenses lead to differences in understanding. Those differences are valuable because they help us see alternatives outside of our own way of interpreting the world (hence the value of diversity) - but what's more compelling than difference in and of itself is why we interpret our environments differently. Bringing those assumptions and framing conditions to the surface is where the value lies.

Recent research in cognitive psychology supports this focus. Pioneering work by scientists such as Karl Friston, Anil Seth, and Andy Clark asserts that our moment-to-moment experience is informed not only by the data that our physical senses collect but through a complex process of predictive modeling that shares an affinity with Bayesian statistical analysis. They argue that our experience of reality is an interpretation and as such is informed, to an extent, by underlying expectations and assumptions and beliefs. Moreover, while we may see the same world "out there" - outside of the skull-bound brain that manifests our experience but doesn't literally "see" it - we each have our own model of that reality. When we engage with each other we're often challenged by the problem of aligning these interpretive models to arrive at a shared understanding. Starting out with this recognition - that there is an inherent difference in the way each of us "reads" the world - helps us to understand the nature of the divide and leads us to explore why we're interpreting the same situation differently and how we can get onto the same page.

To see this in practice, consider an underwriter and a claims manager separately reviewing the wording of a revised commercial property policy. The underwriter sees a risk selection and pricing instrument, with a mix of conditions, sub-limits, and exclusions. He may consider how the language allows for exposure in line with the firm's risk appetite and how definitions protect against adverse selection. The claims manager reads the same document through a different lens. Drawing on years of experience at point of loss, she sees where clients are likely to contest coverage language, where exclusion wording is ambiguous enough to invite dispute, and where the gap between policyholder expectation and actual coverage may drive complaints. What looks like careful risk architecture to the underwriter is, to the claims manager, a map of potential friction. Our perception is shaped by our background, training, and the mental models we've built over time. Shared models can help bridge that perceptual divide. Analogy, deployed carefully, is one effective tool for achieving this.

Analogy can be a powerful communication and alignment tool when it's deployed with some constraints. We define an analogy as a relational structure between two models (which may be concepts, ideas, stories and so on). Analogies can be effective to help us isolate a concept from a familiar context to analyze it carefully (we can think of this like a discrete "lab environment"), or to explore abstract ideas in terms of more familiar constructs, or even to visualize problems that otherwise couldn't be observed. Think of Einstein's thought experiments when developing general relativity, for instance, imagining two lightning bolts striking the end of a train and thereby exploring the idea that simultaneity is relative and not absolute. Similarly, in one well-known example, mapping a radiology problem onto a military campaign helped to generate options for treating a cancerous tumor without killing the patient. This is the art of constructing an analogy as a model for exploration and communication - isolating the fundamental conditions and rules and consciously mapping them across to a different environment to retain focus and facilitate transfer of insights back to your original problem. When we consider the objective of achieving alignment between stakeholders, mapping a core topic into an unfamiliar setting can help to create a level perceptual playing field.

So far we've considered how modeling can help us to acknowledge and address differences in interpretation of a shared topic. Yet not all stakeholder alignment challenges are the same - there's no "one size fits all" approach. Paul Carlile's insight that misalignment may take different forms serves as a useful guide.

Addressing misalignment with models

If we aim to align stakeholders with distinctly different perspectives, and we intend to use models to do so, it helps to think of these differing viewpoints as being separated by boundaries. Carlile defines three levels of boundaries in this context (syntactic, semantic, and pragmatic). For accessibility, we may re-interpret these boundary categories as defined by:

• A common language
• An agreed meaning (but different languages exist)
• Contextual differences

The first level – "a common language" (Carlile's syntactic) – applies when all parties share a similar understanding, both language and familiarity with the concepts involved, such that communication may involve direct knowledge-sharing without significant modification. A straightforward project status update may be one example of this, shared with leads of different functions who are already familiar with a program.

The second level – "an agreed meaning (but different languages exist)" (Carlile's semantic) – applies to situations in which various parties to a communication may be more or less "on the same page" but arrive at this from different perspectives or different ways of framing it. In this instance, there is a need to translate concepts from one form to another that is more readily understood. Analogy can be an effective tool in this case.

The third level - "contextual differences" (Carlile's pragmatic) - describes circumstances in which some form of transformation is required to communicate and engage. This transformation may require a change or compromise from one or more parties and may involve some cost. Models may help to bring differences to the surface that can be addressed and required changes (such as modifications to the concept design or to the budget or timeframes) may be negotiated to achieve alignment.

Returning to Alison's case, the problem that she faced was that her stakeholders were located across the organization and held very different interpretations of the IT program's impact and outcome for their own functions. As an IT leader who worked regularly with the operations team, she shared a common language that had been developed over time. The disconnect in this case was more an issue of control and resource-management. Framing her challenge in this way allowed her to carve it out from the whole and address it separately, drawing on workflow and process-mapping to illustrate anticipated efficiency benefits and systems maps along with specifications to justify vendor selection. In other words, these models along with a common language created a space where all parties could compare their understanding and negotiate points of difference. Her challenge with finance, however, was more complex.

The challenges that the finance leadership team presented stemmed from their position that they saw the re-platform as an opportunity to build new capability (specifically, to support incentive planning for distribution partners and forecasting) though they had limited awareness of the cost, time, and complexity involved in introducing non-standard features. Rather than litigating this issue with the whole team, Alison mapped a quick analogy that drew on the fundamental conditions of the proposed platform onto an example of upgrading a car's engine from a V6 to a V8 (a process thoroughly distant from both of their fields). This example helped her to illustrate the complexity of the existing transformation even before considering upgrades and additional enhancements. The example was sufficiently detailed and connected to the core problem that the finance team felt they were working through a meaningful problem and it helped to provide a clear example of the trade-offs involved before returning to the re-platforming issue at hand and finalizing the scope.

Transformation projects as proving grounds

Transformation projects, even if they're an added responsibility to an already busy schedule, are an opportunity to demonstrate your capacity to engage a wide range of stakeholders and align their efforts around a shared task. Leaders who understand why people interpret the same situation differently, and who have the tools to bridge that divide, are better positioned to turn that live audition into an appointment. The ability to model - to represent a complex idea in a form that others can see, explore, and act on - is one of the most underrated capabilities an insurance operations leader can develop.