The insurance industry is undergoing a pivotal transformation, yet true AI autonomy remains out of reach. While a few leaders have successfully scaled AI, most insurers are still constrained by legacy infrastructure, regulatory caution and immature governance frameworks. Even among advanced adopters, fully delegating operations to autonomous AI agents is not yet feasible.

This paper explores two strategic paths forward: the long-term pursuit of artificial general intelligence (AGI) and the immediate application of reinforcement learning (RL). We introduce a novel reinforcement-switch framework which combines continuous learning with proactive human-AI control transfers to enable accountable autonomy. This model ensures resilience in dynamic environments by embedding trust, reversibility, and oversight into AI operations. It represents a fail-forward approach to engineering safe, scalable autonomy in insurance.

Get White Paper

Sponsored by Cognizant & Venbrook

About the Authors

Dr. Venkatesh Upadrista leads global transformation for the BFSI-IOA vertical at Cognizant. In this role, he is responsible for driving AI-led transformation across the unit, ensuring customer success and enhancing the delivery of modern business operations within the financial services and insurance sectors.

Justin Slaten is Chief Information Officer at Venbrook Group, LLC. He is an accomplished technology executive with over 25 years of management experience. He is recognized for his innovative approach to product development, business process improvement, and scaling teams and operations. His leadership consistently drives enhanced productivity and sustainable growth.

We were recently having a discussion with insurance industry peers, bemoaning a vendor announcement of significantly lower pricing for a popular service in a fiercely competitive segment of the collision repair industry. As we thought about it more broadly, we agreed that the announcement was just the latest in a long series of strategic decisions that reflect the serious threat of a problem we call a "race to the bottom."

The term "race to the bottom" describes a competitive situation where companies lower prices to gain an advantage. The origins of this process are often thought to be a result of globalization and government attempts to bolster their economies. Race to the bottom tactics most often spark a destructive cycle in which the ultimate outcome is negative for all involved.

The property/casualty insurance ecosystem is one of the largest U.S. economic segments and touches hundreds of millions of consumers and businesses. It is a victim of this race to the bottom in many serious ways – some well-recognized but even more that are less visible and rarely discussed candidly. The recent roller coaster cycles of steep rate increases, underwriting losses now turning into gains and corresponding efforts to contain operating costs also have downstream implications to scores of industry service and solution providers.

Property & Casualty Insurance Ecosystem

Prior to 2021, both auto and homeowner insurance lines experienced this very competitive scenario. Advertising budgets swelled, with leading carriers like Progressive spending at $1.95 billion and GEICO at $1.13 billion in 2020. "Switch and save" marketing was everywhere with telematics, usage-based-insurance offers to "bundle and save" luring price-sensitive consumers.

More glaring and lasting was the ramp up to today's runaway homeowner insurance premiums. Weather exposure and inflation attracted much-deserved attention, obscuring what was developing in plain sight during the same period. Insurance to value (ITV) was rapidly eroding as housing prices shot up, followed by steep repair/rebuild costs from 2021 to the present. Rates were simply not keeping pace and were already well behind due to years of suppressed rates: California's Prop 103 contained rates, only for there to be a massive reckoning following the Los Angeles wildfires.

Combined, this has been one of the most disruptive insurance cycles in decades, with highly competitive pressures serving as the "race to bottom" catalyst.

CATALYSTS AND IMPLICATIONS

Private equity and consolidation: The services side of the property/casualty insurance industry – particularly personal insurance claims services – was historically as fragmented a marketplace as any, characterized by many thousands of smaller family-owned businesses. These markets (collision repair, auto salvage, independent appraisers and adjusters, repair and home renovation contractors, software providers and numerous others) were ripe for consolidation, a process that can yield large gains over time but requires immense amounts of investment capital and expertise. Many large private equity funds with mountains of "dry powder" (free cash) entered these spaces, which has triggered many aspects of this race to the bottom. And this is still unfolding.

Reduced profitability: Lower premiums may constrain carriers' ability to adequately cover claims and other operational costs and, in the worst cases, lead to insolvency.

Lower underwriting standards: Less stringent underwriting and risk assessment standards can lead to adverse selection, leaving carriers overweight with high-risk policies and further threatening profitability.

Brand and reputational damage: If carriers prioritize low costs over customer experience, brand loyalty declines. A reputation for poor claims service or missing coverage can make it difficult to attract and retain profitable customers, even at lower prices.

Innovation stagnation: A focus on price competition can stifle innovation in products and services in the name of cost savings and perpetuates product and service commoditization.

The Ecosystem

Effect on ecosystem partners: The majority of insurer operating and claim costs are external, and as carriers seek to contain and reduce these costs, they tend to impose their leverage on their business partners throughout the ecosystem.

Pressure on compensation and commissions: The race to the bottom can affect how insurers compensate independent agents and brokers. This can pressure reps to recommend the cheapest policy, even if it is not the best fit for the customer's needs.

Disruption of the traditional role of agents: The shift toward lower cost digital-first information models threatens the traditional agent's role as a trusted advisor, potentially leading to their replacement by apps or automated systems.

Claim costs: Pressure to reduce loss adjustment expenses (LAE) and cutting full-time equivalent (FTE) labor has been continuing. The corresponding effect (but hard to quantify with precision) on loss costs is now catching up, and job cuts are turning into job replacements and additions of adjusters. After a decade of hammering LAE downward, containing indemnity payouts is now center stage.

The Customers

Protection gaps: Customers may be lured by low premiums without understanding that their policy offers less coverage than they need. This leaves them exposed and vulnerable in the event of a large claim.

Higher out-of-pocket costs: When claims are denied or coverage is insufficient, customers often face unexpectedly high out-of-pocket expenses. This is particularly evident in healthcare, where medical debt is a major issue.

Poor claims experience: To cut costs, some insurers may invest less in their claims departments, leading to protracted, difficult, and adversarial claims processes. This can cause significant stress for customers during an already difficult time.

Eroding trust: The focus on aggressive pricing and cost-cutting undermines customer trust in the insurance industry as a whole. Many customers already view insurance as a necessary evil, and a negative experience can reinforce this sentiment.

Counterarguments and Debate

Not everyone agrees that a race to the bottom is an inevitable consequence of competition. Some economists and commentators argue that it is a myth, especially in the context of globalization. A 2023 article from the Cato Institute argues that global trade has generally led to improved working conditions and higher wages, with most foreign direct investment still flowing to relatively wealthy countries.

But proponents of free trade suggest that companies and investors are ultimately more interested in a skilled workforce and stable institutions, which are more common in countries with higher standards.

However, other research has found empirical evidence supporting the theory, particularly regarding the effects of globalization on labor standards. This divergence of evidence means the debate over the existence and effect of a "race to the bottom" is continuing.

Avoid the Bottom

In the property/casualty industry, these conditions may be dismissed as underwriting cycles as famously illustrated in Paul Ingrey's "Insurance Clock." (See, Insurance Industry's Mixed Signals.) On the one hand, you have new, high premiums, returning profits and growth appetite on the rise. On the other hand are uncertain and fluid economic conditions, weather exposure and threats from legal abuse trends signaling caution. Which will prevail is the question du jour, and has the new race to the bottom already begun?

We believe that this race to the bottom is avoidable even while mitigating its effect on earnings. Scale clearly affords efficiencies, but managing how to wield this leverage without affecting the ultimate quality of outcomes is paramount. Price competition is a positive market dynamic that benefits customers, but irrational and cannibalizing pricing serves nobody. Keeping safety and consumer satisfaction at the top of our strategic priorities is paramount and if not implemented voluntarily will only invite unwanted outside regulation.

As always, common sense is mandatory as we move forward.

AI Disclosure: AI tools were used for some content generation in conjunction with our own subject matter expertise. The authors take full responsibility for the accuracy of the final content.

At a time when highway deaths in the U.S. are finally in a sustained decline again, after a surge related to COVID recklessness and distracted driving, car makers seem to be set on a dangerous path. Wanting to show off their AI chops, they are rolling out and emphasizing autonomy-lite capabilities that are convenient in the best of all possible worlds but that could well increase accidents and fatalities.

Basically, many car companies seem to be heading in the Tesla direction, bragging about their cars' autonomous capabilities while telling drivers they need to have their hands on the wheel and be constantly alert — a combination that just doesn't work. Once you tell people their cars can do the driving, they let their cars do the driving, and the sort of autonomy that exists this side of Waymo and a few other providers of fully autonomous vehicles simply isn't reliable enough yet. 

I've become what's called a 410-er, and I think insurers should be, too.

410 — unrelated to 420, a coinage celebrating cannabis that seems to amuse Elon Musk — refers to three of the six levels of autonomy: levels 4, 1, and 0, which I believe are the safe levels at the moment. Car makers, meanwhile, are focusing on providing levels 2 and 3 to the mass market, levels that I believe are fraught with danger. 

Level 0, as you can imagine, involves no autonomy. Level 1 is referred to as driver-assist — your car keeps you in the center of the lane if you're starting to drift and lets your cruise control maintain a safe distance from cars around you. Both provide little enough assistance that the driver stays fully engaged.

Level 4 is fully autonomous within a certain area and under certain conditions. As long as you stay within, say, a well-mapped area in good weather in daytime, you could be asleep in the back of the car. This level is safe, too — once the technology is proven, of course.

Combine those three levels, all of them helpful without overpromising to the driver, and you get to 410. You could also add level 5, which is fully autonomous anywhere at any time, but we're not there yet, even with Waymo, et al. I think we'll get to a 5410 paradigm, just not yet.

The problem is that car companies want to brag about levels 2 and 3, which can lull drivers into danger. Level 3 is especially seductive, because it promises "conditional driving automation": The system handles all aspects of driving under certain conditions, but the driver must be available and able to take over.

Google, Waymo's parent, tried a Level 3 approach years ago and quickly gave up. The technology was so good that Google employees who volunteered as test subjects soon zoned out and started checking their phones, playing video games or whatever — but the car wasn't totally reliable. And when the car told the driver to reengage, many seconds passed before they could stop whatever they were doing, size up the issue and act. When you're traveling at 75mph, 10 seconds equals almost 400 yards of distance traveled while the driver is taking control of a situation. 

Tesla is finding out how dangerous Level 3 autonomy can be. It's faced numerous lawsuits over fatal accidents that occurred while drivers had engaged what Tesla calls Full Self-Driving (FSD) but which is really a Level 3 system. Tesla has mostly escaped liability by arguing that it warned drivers repeatedly that they were responsible for the car's actions, but it did recently lose a $240 million wrongful death judgment that should serve as a warning to Tesla, to other auto makers offering Level 3 systems and to auto insurers.

Tesla, seemingly unchastened, recently told its car owners that if they felt drowsy they should engage FSD — even though a drowsy driver would take even longer to reengage if told to by the AI driving the car. Other car companies are promising what are generally referred to as Level 2+ systems, and Telemetry says more than half of new cars will be equipped with such "hands free" systems by 2028. 

Car companies will surely promote these systems. Everybody wants to be seen as being at the cutting-edge of technology, and AI can be a real selling point these days. Autonomy is exciting.

But I hope cooler heads will prevail. 

"Customers really love these hands-free systems, especially on longer drives, but God is in the details, and... the worst of these systems may result in preventable injury or worse," Telemetry says.

Traffic deaths in the U.S. had been declining steadily for decades, reaching a low of roughly 35,400 in 2014, but increased steadily as smartphones tempted drivers with distractions and then surged as COVID somehow made drivers more reckless. Deaths peaked at approximately 47,000 in 2021. They dropped slowly from there, falling back to 44,700 last year, and declined a gratifying 8.2% in the first half of this year. 

Let's keep the progress going and push back on efforts to promote Level 2+ autonomy as anything more than an occasional convenience. Level 2+ and Level 3 are wildly impressive technology — impressive enough to be truly dangerous.

410, 410, 410....

Cheers,

Paul

P.S. What happens when a driverless car commits a traffic violation? Who gets the ticket?

That was the issue facing police officers when a Waymo vehicle made an illegal U-turn right in front of them in San Bruno, CA, just south of San Francisco, a week ago. They pulled the car over — Waymo cars pull off the side of the road when a police car turns on its emergency lights — but when they approached the car they found... no one in it.  

Under California law, tickets can't be issued to driverless vehicles until next summer — “Our citation books don’t have a box for "robot,” the San Bruno Police Department noted. Even once tickets can be issued, they carry no penalty. 

Expect the law to change — and expect further oddities as autonomous vehicles become more common. 

At the core of the trust formula, the reliability, honesty, and competence of insurance planners are crucial for building trust. Yet how insurance companies can quantify these factors objectively remains a major challenge. By establishing trust scores and trust levels, improvements in customer relations, sales performance, satisfaction, and competitiveness can be expected.

The purpose of this article is to present the basic reasoning and exploration of “making trust AI-driven.” In a human–machine collaboration environment, this provide support for humans; in a machine-based environment, it serves as guideline and regulation for AI. This is my fascinating learning journey together with GPT-4 — a process that moved from nothing to something, through initial exploration, iterative dialogue, and joint construction, transforming abstract concepts into digital form and refining a thinking framework for solving complex problems. Although the conclusions here have reference value, I suggest readers focus more on the process: to imagine innovative solutions, and to consider how to collaborate with artificial intelligence by building new thinking frameworks to solve their own problems.

What is trust? It is a highly abstract concept, yet so close to our daily lives. Why do we need trust? No one would deny that all transactions are built on trust, especially in insurance. This is not only because insurance is an intangible product, but also because its transactions involve a time gap — unlike “cash on delivery.” Therefore, trust is particularly important for insurance.

Insurance relies on “sales,” and primarily on “face-to-face sales.” The current insurance market is still dominated by company agents. Even in multi-channel sales, apart from pure online sales and telemarketing, bancassurance and broker channels are also based mainly on “people + face-to-face interaction.” Today’s difficulties and transformation challenges in the insurance industry may have their stage-specific inevitabilities, but technological trends still bring a glimmer of hope. Personally, I believe the primary problem in the insurance industry today comes from the sales side, with lack of trust being the core issue. Under the same thinking framework and system structure, merely relying on technological empowerment to rebuild consumer trust in agents and insurance companies is, frankly speaking, difficult. In the past we have shared many articles discussing the importance of building trust. Especially with the advent of the AI era, when human–machine collaboration has become a trend and machines are gradually replacing humans in many areas, how should we face such a scenario? How should we “effectively manage” trust? This has always been a question on my mind.

Therefore, how to make the abstract concept of trust more concrete — not limited to conceptual analysis and written description, but able to be digitalized, structured, and even AI-driven — is the focus of this article. Using insurance planning as the scenario, the views on trust and the final constructed “trust framework” in this text were developed through strategic, dialogue-based prompting with OpenAI’s GPT-4, combined with existing theories and logical reasoning. Although these views and results are based on current research and theory, their combination and application in different scenarios still carry originality and innovation. I encourage readers to think critically about these viewpoints, and to consider how they might apply to more diverse real-world situations. Furthermore, this highly experimental attempt does not fully cover all relevant knowledge and needs and still requires further exploration and validation by professionals.

This article includes two types of visual inserts to distinguish between human prompts and AI responses:

• Gray background with black text represents the prompts or instructions I provided to the AI.
• Blue background with blue text represents the AI’s responses.

These visual cues are designed to help readers follow the flow of interaction and better understand how the trust framework was co-developed through iterative dialogue.

To support this exploration, the article includes 12 visual inserts that document the process of building a trust framework through iterative dialogue with AI. These visuals are not standalone illustrations—they form a clear and progressive logic path for understanding how trust can be made AI-driven. From conceptual exploration to quantification, from system design to scenario-based application, each figure contributes to a modular and adaptable trust framework. Together, they demonstrate how AI can not only execute trust logic, but also participate in shaping, adjusting, and contextualizing it across different environments.

Below is an overview of the entire process of “building the trust framework.”

Figure 1. A Nine-Step Framework for Building Trust in AI Systems

As illustrated in Figure 1, the nine steps outline the process of constructing the entire trust framework.

I. Iterative Exploration Process

1.1 From vague to concrete

For a long time, our understanding of the question “How can trust be made AI-driven?” was vague, let alone having a concrete answer — and even today, the results cannot truly be called a final “answer.” Our knowledge of trust came only from theories and concepts found in books, along with subjective impressions shaped by personal life experiences. There was no consistency in understanding, much less any generalization. Because trust is deeply entangled with human relationships and subjective perceptions, we once assumed that digitizing trust was an impossible task. Unknowingly, we even “assumed” viewpoints such as: trust can only be described in words; trust must be built through process and details; trust cannot be quantified objectively.

When I came across Helen Keller’s description of her understanding of colors (Note 1), my rigid thinking was jolted. Later one day, I accidentally saw a written expression of “beauty” and “art” (see prompt illustration below). Suddenly, inspiration struck — could we try transferring the problem, and “use beauty to map the presentation of trust”? Perhaps this could allow AI to concretize the difficult and abstract concept of trust. Once it is concretized, subsequent digitalization and structuring would be relatively easier.

As my dialogue with GPT-4 continued, the questions and answers gradually became more concrete and in-depth. It was an iterative process. Each round of dialogue built on the previous one, where human and machine together clarified and refined the problem, jointly exploring and constructing an understanding of the concept of “trust.” Views emerged, expanded, and improved in the process.

Looking back now, many prompting strategies introduced in the past — such as Chain of Thought (COT), OPRO, knowledge generation, step-back prompting, etc. — together with advancements in AI capability, such as emotional intelligence, all transformed seemingly impossible barriers into goals that could be achieved.

1.2 Starting Point: Building a Thinking Framework

To obtain a concrete result for a virtual concept (such as trust), the process begins with abstraction, and abstraction starts from extracting the key information of the virtual concept.

This entire exploration journey began when I entered the following prompt of “from beauty to trust” (Note 2):

Figure 2. Prompt for Using the Concept of Beauty to Explore and Define Trust

After receiving the above question, based on the way of thinking I provided, GPT-4 demonstrated impressive reasoning ability in concretizing the concept of “trust.” Its process of “making trust concrete” was as follows:

Figure 3. Concretizing Trust Through Definition, Metaphor, and Social Relevance

From the above response, some key pieces of information already emerged, mentioning many possible keywords for trust elements, such as reliability, honesty, competence, time, interaction, emotion, expectation, dependency, scenario, etc. Although still not clear enough, I felt — it seemed I had asked the right question!

Next, two things needed to be done with these keywords: condense them again to abstract form, and then reason based on abstraction. When the experiment ended, I realized these two steps were consistent with the “Step-Back Prompting” methodology I had just learned.

In handling the first step, my thought was: GPT-4 has a vast knowledge base and letting it select the most essential keywords would certainly be more objective than if I did it myself. What I needed to do was to tell it how to abstract. Thus, I gave GPT-4 the task of condensing the keywords, while my purpose for abstraction was to build a “trust thinking framework.” I needed a method that not only represented this virtual concept of trust, but also could, through reasoning, later achieve digitalization and quantification. I thought, perhaps a “formula” could solve this tricky problem.

Although I had no confidence in how GPT-4 would respond, I asked the next question, right after its previous output (noting that context continuity is very important for GPT-4): “Can trust be transformed into a formula?”

GPT-4’s response was surprising! It really output a formula. At the same time, it explained that expressing trust as a formula is highly simplified and abstract, and that trust is in fact a complex human emotion and psychological state that cannot be fully described by a formula. Instead, the formula serves as a thinking tool to help us understand and analyze the different components of trust. I understood its point, but with a formula, the second step of reasoning became possible. Below is GPT-4’s output of the “trust formula.” I personally think the elements it identified are key, theoretically supported, comprehensive, and logically consistent (Note 3). I present the “trust formula” in the following original form:

Figure 4. Translating Trust Into a Quantifiable Formula

Overcoming the first hurdle, we have the trust formula：

Trust = (Reliability × Honesty × Competence) × Time / Interaction

Among these, the most critical elements are reliability, honesty, and competence, while time and interaction act as positive or negative forces influencing trust.

1.3 Process: Reasoning to Find the Answer

Having only a trust formula could not solve the problem; we also needed a method to “quantify trust.” To attempt digitalization, GPT-4 suggested a simplified method: assigning a quantitative value to each factor in the formula.

With this quantification method to overcome the barrier of digitalization, we need a “password,” which would not only indicate subjective trust levels but could also evaluate the value of trust, including reviewing our coming action plans and building a user trust system.

Inspired by Ant Group’s “Sesame Credit Score,” I temporarily called this password “Trust Score” (Note 4). Returning to the example above, the calculated trust score was 840.

II. Building a User Trust System

2.1 Final Confirmation and Key Prompts for Building a User Trust System

My next question was: “If we want to form a Trust Score based on the trust formula, and each user is an independent individual, how do I know my Trust Score in the user’s mind? And can this Trust Score be objective and accurate enough to serve as an indicator for continuous improvement in the future?”

This question was important because I was clarifying to GPT-4 who was evaluating whom, in preparation for applying it to the insurance planning scenario. And because in real business contexts, we need a quantified understanding of the trust level users place in us. These must be objective and accurate enough to provide a reliable foundation for guiding business strategies and improvement plans.

Since the application scenario has not yet been clearly defined, GPT-4’s response lacks specificity. However, its comprehensiveness and completeness still offer valuable reference points. Regarding the “objectivity of Trust Scores,” here is the simplified version of its response:

Figure 6. Operational Framework for Evaluating and Strengthening User Trust

At this point, we had roughly mastered abstraction of the virtual concept of trust (through transfer and mapping), the method for extracting key information (the trust formula), and the feasibility of objectively quantifying trust (the Trust Score). Next, with the goal of “completing the understanding of user needs, matching appropriate insurance products, and completing transactions,” I described a specific scenario to GPT-4 and proposed three action requirements: “Building a user trust system,” “Trust Score grading and standard operating procedures,” and “Operational guidelines for obtaining the Trust Score.” Please see the following "Three Action Prompts for Building User Trust":

Figure 7. Prompt for Constructing a User Trust System in Insurance Planning

2.2 Three Actions for Building a User Trust System

Below is my summary of GPT-4’s “Three Actions for Building a User Trust System,” omitting details, as reference for constructing a trust system and managing trust. Readers are welcome to try any methods mentioned here or copy the prompts and adjust them for their own scenarios.

Figure 8. AI-Generated Framework for Building a User Trust System

Action 1: Build a User Trust System.

• Reliability and honesty: evaluated through self-assessment, customer feedback, or direct communication with customers to understand perceptions and satisfaction.
• Competence: improved by regular training and assessments.
• Interaction: record not only frequency but also quality and effectiveness.
• Time: emphasize the importance of building long-term relationships.
• Trust Score: calculate periodically (e.g., monthly or quarterly) to monitor changes in trust levels. Based on results, identify areas needing improvement and set strategies accordingly.

Action 2: Trust Score Levels.

• Different levels correspond to different actions and goals.
• Example: Level 1 (0–200 points) — Low trust.
  • Recommended actions: strengthen basic communication skills, improve service transparency, ensure accuracy of information.
  • Goal: build basic trust and resolve misunderstandings.

Action 2 (continued): Standard Operating Procedures.

• Regular evaluation: assess each customer’s Trust Score to determine their level.
• Personalized strategy: design and implement service strategies based on trust levels.
• Continuous tracking: track effectiveness and adjust as needed.

Action 3: Guidelines for Obtaining the Trust Score.

• Positive interactions: ensure each interaction is constructive, focused on customer needs, providing professional advice and solutions.
• Avoid negative interactions: prevent delays, inaccurate information, or unsuitable suggestions.
• Post-service updates: collect feedback through surveys, especially on reliability, honesty, and competence.
• Regular updates: update scores based on feedback and interactions.
• Implementation:
  • Employee training on how to build and maintain trust in every interaction.
  • Monitoring quality of interactions and adjusting based on customer feedback.

III. Ensuring Objectivity of Subjective Evaluation and Notes on AI-Driven Implementation

To fully convey the integrity, transparency, and fairness of the methodological validation process to readers, this chapter presents excerpts of GPT-4's responses.

3.1 How to Ensure Objectivity of Subjective Evaluation

When considering the Trust Score and evaluation methods, readers may already notice that some approaches may carry subjective bias. Therefore, regarding “how to ensure the objectivity of subjective evaluation,” below is an excerpt from GPT-4’s response:

Figure 9. Strategies for Enhancing Objectivity in Subjective Trust Evaluations

We already know AI systems have advantages in handling data and ensuring consistency, but they have limitations in building human emotional trust and understanding complex interpersonal interactions. Therefore, if we want to apply the trust system to AI planners, some key adjustments and special considerations are required.

3.2 Notes on AI-Driven Implementation

To untie the AI knot, we must ask who tied it. GPT-4 suggests that when applying a trust model to AI planners, several key dimensions must be considered: 

• Demonstrating competence.
• Transparency and honesty (helping users understand how AI generates suggestions).
• Adaptability and personalization (learning and adapting based on user history and feedback).
• Context awareness (AI must understand and adapt to different user contexts and needs).

By integrating these theoretical elements, we can create a more complete and in-depth trust framework that applies not only to human service providers but also to AI systems. The following excerpts on “AI-Driven Trust” are drawn directly from GPT-4’s responses:

Figure 10. Foundational Components for Building Trust in AI Systems

In addition, GPT-4 offered further suggestions: continuous user education, enhancing user participation (channels for feedback and suggestions), use of social proof (customer recommendations and positive experiences), emphasizing personalized services, use of technology (CRM and data analysis tools), transparent communication (sharing both positive and negative feedback), flexibility and adaptability (customized services and solutions), risk management and compliance (clearly explaining risks and uncertainties), and continuous evaluation and improvement (keeping track of industry trends).

So far, our “trust framework” is roughly complete. It is built on existing theoretical reasoning and logical deduction, with strong internal consistency and theoretical foundation. This framework integrates multiple key factors in building trust, such as reliability, honesty, competence, interaction, and time — all widely recognized in trust theories and research.

IV. Other Suggestions

The above trust framework was developed in the scenario of insurance planning. But whether the framework can stand the test of generalization is still a concern. Every industry and scenario have its own unique characteristics and needs, which may affect how trust is built and maintained. Therefore, while we can provide a general framework, it may need to be adjusted and customized for specific situations.

For example, in healthcare, trust relies more on professional knowledge and privacy protection; in retail, it depends more on product quality and customer service. This is why an effective trust framework must have adaptability and flexibility to fit different environments. Industry knowledge, culture, and expectations are also indispensable in building trust frameworks.

4.1 Other Insurance Scenarios

Different insurance scenarios require different trust-building strategies. For example, in claims and underwriting, GPT-4 developed adjusted formulas to fit the unique needs of these contexts. Transparency, fairness, expertise, and effective communication remain essential elements of trust.

Figure 11. Scenario-Based Adjustments to the Trust Formula in Insurance

Different scenarios in the insurance industry may require different trust-building strategies, but transparency, fairness, expertise and effective communication are always key elements in building trust.

4.2 Other Industry Scenarios

To adapt to different industries, GPT-4 also created examples for healthcare and retail. In healthcare, trust is more tied to expertise, patient care, and communication transparency. In retail, trust emphasizes product quality, service, and price transparency.

Figure 12. Industry-Specific Trust Formulas for Healthcare and Retail

Beyond insurance, healthcare, and retail, other industries such as education, financial services, technology, and online services — as well as areas like corporate management and customer relationship management — can apply similar methods. By adjusting elements, weights, and sequence, each industry can find the most suitable trust formula. For some industries, honesty and transparency may matter more; for others, professional competence and long-term service quality may be emphasized. By understanding industry-specific needs, these factors can be effectively adapted for better trust-building.

V. Conclusion

By constructing a trust system through self-assessment, customer feedback, or direct communication, planners can enhance their reliability and honesty. Companies can use customer Trust Scores and levels to design standardized procedures and personalized strategies, driving decisions with data. Whether as a sales management tool or a way to improve satisfaction and loyalty, this should be helpful. Therefore, building a trust system can not only improve customer relationships and sales performance but also bring long-term benefits to companies — including brand building, competitiveness, employee and customer satisfaction.

In this article, we explored the application of the trust concept in insurance through AI. We discussed methods for building a user trust system, emphasized the importance of ensuring objectivity in subjective evaluations, and highlighted factors to consider when integrating AI. We also explained how to transform trust from an abstract concept into concrete, digitalized practices, and how to apply these principles in different contexts. The key elements of trust include reliability, honesty, competence, interaction, and time.

A major challenge is the objectivity of reliability, honesty, and competence. Although we provided a basic formula, real-world application requires flexibility to fit different situations. Trust-building is a dynamic process that demands ongoing effort and maintenance. This means regular evaluation and adjustment to reflect industry trends and changing customer needs. From examples in different industries, we can see that trust has both universality and specificity. Thus, the trust framework should be seen as a flexible, adaptive tool, not a rigid rule.

In this exploration journey, GPT-4, trained on massive text data, displayed strong contextual understanding, creative thinking, logical reasoning, and interactive dialogue ability — becoming an ideal partner. GPT-4 could track discussion context, adjust responses accordingly, and integrate new viewpoints throughout the process. Its creativity in handling complex problems was impressive, and its interactive dialogue proved that conversational prompting could match structured prompting in problem-solving ability.

This experiment was only a starting point and process for logical reasoning and theoretical validation. Looking ahead, AI’s role in building and maintaining trust in insurance will become increasingly important. With technological progress, we can foresee a smarter insurance ecosystem, where AI not only improves efficiency but also strengthens customer trust through precise data analysis and personalized service. In the future, AI will become humans’ best assistant, understanding customer needs more deeply, predicting market trends, and even supporting complex risk assessments and decision-making. This will change how insurance products are developed and marketed, and will redefine the relationship between insurance companies and customers.

Applying these theories to practice means insurance companies must continuously adapt to new technologies while ensuring ethical and legal compliance. In practice, companies should focus on cultivating employees’ understanding and use of AI, while strengthening communication with customers to ensure technology truly meets their needs and expectations. Companies must also take measures to protect customer data security and privacy — both are critical for maintaining and enhancing trust. Through these efforts, AI will not only be a tool but also a major driving force for making insurance more efficient, transparent, and customer-centered.

Trust is fragile, but it is also essential in all human relationships. In today’s fast-changing world, trust is not only the trigger for completing transactions but also the bond for maintaining long-term relationships. How to make AI a key part of building and maintaining trust is a question every decision-maker and manager must consider.

The insurance industry today urgently needs to find a way out of its current challenges. I hope this experiment can serve as a starting point—testing the waters and offering something that may inspire further exploration.

Further Reading

AI Reshapes Trust in the Insurance Industry. Insurance Thought Leadership (ITL). This article was also selected for promotion by the International Insurance Society, IIS.

References and Notes

1. Helen Keller’s reflections on her perception of color: “For me, there are also delicate colors. I have a palette of my own. I will try to explain what I mean: pink reminds me of a baby’s cheek, or a gentle southern breeze. Lilac was my teacher’s favorite color, and it reminds me of faces I have loved and kissed. To me, there are two kinds of red—one is the warm blood of a healthy body; the other is the red of hell and hatred. I prefer the first because it is full of vitality. Likewise, brown comes in two forms: one is the rich, friendly brown of living soil; the other is dark brown, like a worm-eaten tree trunk or a withered hand. Orange gives me a sense of joy and cheerfulness, partly because it is bright, and partly because it gets along well with many other colors. Yellow means richness to me—I think of the sun pouring down, full of life and hope. Green means vitality. The smell brought by warm sunlight reminds me of red; the smell brought by coolness reminds me of green.”
2. Source currently unverified. If any copyright concerns arise, please contact the author.
3. This formula integrates several key factors in trust-building—such as reliability, honesty, competence, interaction, and time—which are widely recognized across various trust theories and research. For example:
   1. Mayer, Davis, and Schoorman’s trust model identifies three core components: ability (skills and knowledge in a specific domain), benevolence (the belief that the trustee has good intentions), and integrity (adherence to principles and values).
   2. Rogers’ extended trust model builds on Mayer’s framework by adding contextual factors (how environment and situation affect trust) and historical interaction (how past experiences shape current trust levels).
   3. Other relevant theories include social exchange theory and psychological models of trust.
4. “Sesame Credit Score” evaluates users from the merchant’s perspective; “Trust Score” evaluates merchants from the user’s perspective. Though their starting points differ, their nature is similar.

I can remember back in 2013 when healthcare reform was passed, wellness programs were springing up everywhere claiming to reduce health insurance costs.  In theory it sounded great. Healthier employees meant lower insurance claims and showed employees the company valued them and wanted to keep them around. Many companies enrolled their employees into these programs expecting to see health insurance premiums reduced. Flash forward a few years, and many of these companies have dropped their wellness programs. Why? Because the hoped-for results didn’t happen. Instead, there was a. lot of initial interest, but engagement waned over time, and the promised results were not achieved.

Companies that dropped wellness programs discovered their programs weren’t living up to their promises because installing a program was not as easy as recruiting employees and waiting for results. It would take some work!

In today’s uncertain economy, employers are seeking ways to reduce costs, and, after payroll, employee benefits are usually a business’ highest expense. It may be the right time to revisit wellness and see if it could offer cost reductions.

For the companies that had good results, corporate wellness programs were able to offer significant financial advantages that leverage IRS-allowed reductions. Structured properly, these programs can generate a reduction in payroll tax liability while promoting healthy behaviors.

Structuring a Successful Wellness Program:

A successful wellness program is comprehensive, data-driven, and tailored to your organization’s unique needs. Here’s a structured approach based on best practices and expert recommendations:

1. Understand employee health needs, stressors, and preferences [1] [2]
2. Appoint an executive sponsor and form a cross-functional wellness committee (HR, health & safety, line managers, employee volunteers) to oversee the program [3].
3. Develop a holistic, inclusive plan that addresses multiple dimensions of wellness: physical, mental, social, financial, nutritional, environmental, spiritual, and intellectual [4] [5].
4. Set clear, measurable goals and objectives (e.g., increase participation rates, reduce stress, lower healthcare costs) [6] [7]. Establish a governance structure to manage and evaluate the program.
5. Make programming accessible and engaging, offering a mix of easily accessible activities and interventions that resonate with employees, such as fitness classes, mental health resources, health education, and flexible work options [1] [2] [5] [6]. Use a user-friendly platform for resources, progress tracking, and rewards [6].
6. Pilot, communicate, and iterate -Launch a pilot program in one department or location to test effectiveness [7]. Gather feedback through surveys and participation data; adjust programming as needed [7]. Communicate regularly
7. and clearly about program benefits, activities, and successes to maintain engagement [4] [6]
8. Measure and evolve - Monitor participation, health outcomes, and organizational metrics to assess impact[2] [4] [7].Use data to refine and expand the program, ensuring it continues to meet evolving employee needs[2] [7].

To maximize benefit usage, it is important to make sure the program is structured to keep the employee engaged. No single incentive works for all employees. A mix of monetary, benefits-based, and recognition incentives, tailored to employee preferences, is most effective.

Benefit -- Wellness Program Impact

Engagement -- 3x more likely to be engaged, higher job satisfaction, motivation and loyalty

Turnover/Burnout -- Lower turnover, less burnout, improved retention

Team Dynamics/Culture -- Better collaboration morale and inclusivity

Overview of Tax Incentives for Wellness Programs (see appendix 1)

By strategically implementing and managing wellness programs-while taking full advantage of available tax credits, deductions, and payroll tax reductions-employers may create a healthier, more engaged, and loyal workforce, all while improving their bottom line. There are three types of financial benefits for employers that create and implement wellness programs: Direct Tax Credits and Deductions, Payroll Tax Incentives, and Premium Deductions. Here is an overview of each:

Direct Tax Credits and Deductions: Recent initiatives allow businesses to claim tax reduction or deductions for investing in employee wellness. For example, some programs offer up to $650 per employee as a tax incentive, directly reducing the cost of implementing wellness benefits [9] [11]. Eligible benefits include [12]:

1. Health screenings and assessments
2. Fitness equipment and facilities
3. Preventive care initiatives
4. Mental health resources
5. Nutrition workshops and fitness challenges

Payroll Tax Incentives: Programs like the Preventative Health Initiative (PHI) enable companies to reduce payroll tax liabilities, freeing funds to reinvest in employee health and wellness without increasing out-of-pocket costs [11].

Premium Deductions: Employers can deduct the portion of health plan premiums that fund wellness programs, further lowering their taxable income [10].

Review of Tax Regulations for Employer-Sponsored Wellness Programs

This review assesses the accuracy of tax regulation statements regarding employer-sponsored wellness programs, focusing on the deductibility of expenses, tax treatment of incentives, and compliance with relevant sections of the Internal Revenue Code (IRC).

1. Tax Deductibility for Employers

• Ordinary Business Expense Deduction: Employers may deduct the costs of implementing and maintaining wellness programs as ordinary and necessary business expenses. This includes expenses such as health screenings, fitness equipment, wellness coordinators’ salaries, and educational materials [16] [15].
• Premium Deductions: Employers can deduct the portion of health plan premiums that fund wellness programs, further reducing taxable income [15] [16].
• Payroll Tax Incentives:  Some programs, like the Preventative Health Initiative (PHI), allow employers to reduce payroll tax liabilities, making wellness investments more cost-effective [13].

2. Tax Treatment of Wellness Incentives for Employees

• Tax-Free Benefits: Certain wellness benefits can be provided tax-free to employees if they qualify as medical care under IRC Sections 105 and 106. These include:
• On-site fitness facilities.
• Health screenings and preventive care.
• Employer contributions to HSAs, FSAs, or HRAs (if compliant with plan rules and nondiscrimination requirements)[16] [15].
• Taxable Benefits:** Cash incentives, gift cards, gym membership reimbursements (unless prescribed for a diagnosed disease), and other non-medical rewards are generally considered taxable income to employees. These are subject to payroll and income tax withholding unless they qualify as a “de minimis” fringe benefit or meet specific medical care criteria [15] [16].
• Wellness Incentives and Double Dipping: If employees pay wellness premiums via pre-tax salary reductions and later receive reimbursements for those premiums, the reimbursement is generally included in gross income, as this constitutes a “double-dipping” arrangement [15].

3. Individual Taxpayer Deductions (Section 213)

• Medical Expense Deduction: Individuals may deduct unreimbursed medical expenses exceeding 7.5% of adjusted gross income (AGI). Qualified expenses must be for the diagnosis, cure, mitigation, treatment, or prevention of disease, or for affecting any structure or function of the body [17] [18] [19] [13].
• General Wellness Exclusion: Expenses that are merely beneficial to general health (e.g., general gym memberships, nutrition counseling not tied to a diagnosed disease) are not deductible. Only those prescribed for a specific medical condition qualify [17] [18] [19] [13].
• Examples of Deductible Expenses: Physical exams, smoking cessation programs, prescribed nutritional counseling, and gym memberships if prescribed for a diagnosed disease [18].

4. Section 105 and Section 125 Plans

• Section 105 (HRAs): Allows employers to reimburse employees for medical expenses, including premiums, on a tax-free basis. This applies to a wide range of medical, dental, and vision expenses [13].
• Section 125 (Cafeteria Plans): Enables employees to pay for qualified benefits (including wellness-related health insurance premiums and FSAs) with pre-tax dollars. Plans must be nondiscriminatory and properly documented [13].

Key points from IRS guidance:

• Wellness Incentive Payments Are Taxable: If a wellness program provides cash or cash-equivalent benefits (such as gift cards or direct payments) to employees for participating in wellness activities, and those payments are not strictly reimbursements for qualified medical expenses, the IRS considers those benefits taxable wages. Employers must withhold federal income and employment taxes (FICA, FUTA) on these amounts [20] [21] [23] [24]
• No Double Tax Advantage: The IRS has specifically cautioned against programs that attempt to provide both pre-tax premium payments and tax-free benefit payments under fixed indemnity or similar plans, as this would constitute “double dipping.” If premiums are paid pre-tax and benefits are not strictly for unreimbursed medical expenses, the benefits are taxable [20] [24].
• Qualified Medical Expenses Only: Only wellness programs that reimburse employees for actual medical care expenses (as defined by IRC Section 213(d)) can provide tax-free benefits. Payments for general wellness, nutrition, or personal health expenses that do not qualify under Section 213(d) are not excludable from income.  Neither do gym and fitness equipment. [21].

Appendix 1

Appendix 2

Appendix 3

Wellness Platforms

Note: Suitability depends on an organization's specific needs, such as the importance of customization, holistic wellness, ease of use, or integration with existing systems

Appendix 4

Pros and Cons of Wellness Incentives

Appendix 5

The IRS has ruled the following expenses qualify for tax-free reimbursement:

• Acupuncture (excluding remedies and treatments prescribed by acupuncturist)
• Alcoholism treatment
• Ambulance
• Artificial limbs/teeth
• Chiropractors
• Christian Science practitioner’s fees
• Contact lenses and solutions
• Co-payments
• Costs for physical or mental illness confinement
• Crutches
• Deductibles
• Dental fees
• Dentures
• Diagnostic fees
• Dietary supplements with doctor’s le]er of medical necessity
• Drug and medical supplies (i.e. syringes, needles, etc.)
• Eyeglasses prescribed by your doctor
• Eye examination fees
• Eye surgery (cataracts, LASIK, etc.)
• Hearing devices and batteries
• Hospital bills
• Insulin
• Laboratory fees
• Laser eye surgery
• Nutritional counseling, weight-loss programs, and gym memberships if the purpose is to treat a specific disease diagnosed by a physician
• Obstetrical expenses
• Oral surgery
• Orthodontic fees
• Orthopedic devices
• Over-the-counter medication
• Oxygen
• Physician fees
• Prescribed medicines
• Psychiatric care
• Psychologist’s fees
• Routine physicals and other non-diagnostic services or treatments
• Smoking-cessation programs
• Smoking-cessation over-the-counter drugs
• Surgical fees
• Vitamins with doctor’s letter of medical necessity
• Weight-loss programs with doctor’s letter of medical necessity
• Weight-loss over-the-counter drugs with doctor’s letter of medical necessity
• Wheelchair
• X-rays

Appendix 6

Key points from IRS guidance:

• Wellness Incentive Payments Are Taxable: If a wellness program provides cash or cash-equivalent benefits (such as gift cards or direct payments) to employees for participating in wellness activities, and those payments are not strictly reimbursements for qualified medical expenses, the IRS considers those benefits taxable wages. Employers must withhold federal income and employment taxes (FICA, FUTA) on these amounts [20] [21] [22] [23].
• No Double Tax Advantage: The IRS has specifically cautioned against programs that attempt to provide both pre-tax premium payments and tax-free benefit payments under fixed indemnity or similar plans, as this would constitute “double dipping.” If premiums are paid pre-tax and benefits are not strictly for unreimbursed medical expenses, the benefits are taxable [20] [23][24]
• Qualified Medical Expenses Only: Only wellness programs that reimburse employees for actual medical care expenses (as defined by IRC Section 213(d)) can provide tax-free benefits. Payments for general wellness, nutrition, or personal health expenses that do not qualify under Section 213(d) are not excludable from income [21] [24].

Paul Carroll

The insurance industry has been operating under projections that approximately 400,000 professionals will retire soon, meaning that an awful lot of knowledge as well as an awful lot of people will be walking out the door. What can we do?

Sharon Emek

The insurance industry is challenged because we don't have enough young people still coming into the industry. There's now become sort of a bias by companies against young people. I hear it from my clients all the time. Young people don't have the same work ethic. They're nine to five. They don't put in the extra effort. Within a year, they want a promotion and a raise. Then they move on.

On the other hand, with the older generation, first of all, we're living longer. People have to work beyond 65. So retirement is a conversation the industry needs to have.

If somebody is turning 65 and they say "I'm done with all the stress," it's not that they really retire and do nothing. A lot of people retire and end up being a greeter at Walmart. I see them at retail stores, see them at grocery stores because they still want to work. They just don't want the kind of stress and travel time required for a full-time job. They want to be near home.

So the industry needs to think about this looming huge Boomer population that is going to retire—they don't really want to retire fully from work.

Some will because they have medical issues, but many of them want to do something less challenging while still being productive. Because people cost more medically as they age, the industry sort of lets them go instead, but we need to figure out ways to keep them in a different capacity, maybe with lower salary.

Young people can teach older workers about technology while gaining insights from them. One group is technologically savvy while the other has all that institutional knowledge, wisdom, and work ethic.

Paul Carroll

Where do you think holding on to older talent could have the most benefit?

Sharon Emek

Brokers are probably being affected more than any other sector, because a lot of the people in brokers' offices are Boomers—especially women who never went to college and went to work when they were 18 or 19 years old at a local agency.

And brokers don't have HR people helping with all this stuff.

Paul Carroll

The move these days is to get people back in the office. How does that trend fit with what you’re describing?

Sharon Emek

Well, I've been doing this for 15 years and have placed thousands of people at brokers, carriers, reinsurers, MGAs, MGUs. While a young person may need to be in an office—they're on a career path, they need relationships, they need to be mentored—a highly experienced older person does not. And everyone is familiar enough with the technology these days to work remotely.

Older workers have an amazing work ethic. They're happy to feel that they're still engaged. They're happy that you're allowing them to have a flexible life. They can work from home and save all that commuting time. They’re no longer on a career path, so they’re not stressing themselves or my clients about promotions and raises and all that stuff.

Instead of offshoring, having a retiree is such a better option because they could do double the work because they know what they're doing. And they're less expensive because they're retired; they know they're not going to get the same money as before.

I call them the "pre-tired." They’re in their "pre-tirement."

Paul Carroll

I love the concept in theory but am curious about the details. How do you structure employment for your workers?

Sharon Emek

You can't do part-time less than 20 hours a week. It's not enough time. So the minimum is typically 20 hours. Nobody wants to just work 10 hours.

60% of our people are working roughly 25 hours, and 40% work 35 to 40 hours a week. Some people feel like 35 to 40 hours is nothing compared with what they were doing. They say, "Oh, I live outside of Atlanta and was commuting an hour and a half each way in the car. That's three hours a day."

We pay them by the hour, but this isn’t like a consulting gig. We tell clients they need to treat our people just like they would any full-time employees. Our people have to become part of the team to do the work they need to do.

We give people the option of having benefits if they need them. They're not independent contractors.

Paul Carroll

How long do people keep working?

Sharon Emek

Our average age right now is 64, but I have people in their eighties who are still working, as well as people who are in their late sixties or seventies. So it depends.

One of my managers is in her mid-eighties. She's terrific, and she's not ready to retire yet. She still goes to rock concerts. Another woman who's almost 80 is a roller skate champion and still competes.

Paul Carroll

My mother was still playing tennis three times a week at age 87 and, at 92, was getting answers that all the contestants in a round of the Jeopardy! Champions tournament didn’t know, so I hear you.

Any final thoughts?

Sharon Emek

The industry is still going to need young people, and we need to have a multigenerational workforce. Many companies are having some challenging experiences with young people, but they have to move on from that and figure out how to attract the right young people with the right culture.

Getting them mentored the right way is critical because companies need to bring them up through the ranks. Eventually, our retiring population will fully retire, and the next generation isn't as big a cohort. We simply don't have enough young people coming into the workforce.

And AI is not going to replace everything. There will always be roles that require human judgment and expertise.

But the “pre-tired” can help a lot as part of the mix.

Paul Carroll

This is great, Sharon. Thanks so much.

About Sharon Emek

Sharon Emek, Ph.D., CIC, is chairman and CEO at WAHVE, a talent agency founded with a vision to address the approaching Baby Boomer retirement and growing need for experienced talent in the insurance industry. She is a frequent speaker on the challenges that employers and “vintage” professionals are facing today.

As an advocate for innovation who has been chronicling the power of digitization for nearly four decades, I sometimes get impatient with the pace of change in insurance. But I take comfort from the latest global survey by our friends and colleagues at the International Insurance Society. 

In particular, I'm encouraged that 66% of those surveyed said artificial intelligence is their top technology priority, up from 17% just four years ago. I'd say the figure should be well north of 90%, but the latest results still show a remarkable increase in awareness of the possibilities presented by AI.

The emphasis on AI dovetails with other survey results on issues such as operational efficiency and talent, where AI can solve a lot of problems. More broadly, the survey supports my belief — or, at least, hope — that insurance can be a leader in adopting generative AI. Insurance obviously has major regulatory issues to deal with and owes stability to its customers, but, despite those constraints, insurance is a perfect fit for AI. At its core, it is a purely digital business, based on the masses of data that AI is perfectly designed to collect, analyze and act on. 

The tendency is to look at technology revolutions as Big Bangs. For instance, we all learned in high school that James Watt's steam engine produced the Industrial Revolution. True enough, but what we (at least, I) missed is that Watt unveiled his key improvement in 1776 while the economic gains didn't start showing up until the 1820s, some 45 years later. A lot had to happen in the interim, including technology improvements, legal innovations and the invention of the factory. The same will be true with AI: We aren't going straight from invention to a change in life as we know it. But we've learned a lot over the past two-plus centuries and won't need 45 years this time. A lot of the reason is the sort of focus that the IIS survey of insurance executives reflects. So, kudos.

Let's look at some of the other interesting results.

The other super-high priority in the IIS survey was inflation. It was the top economic priority for the fourth consecutive year, with 63% of respondents listing it among their top three issues.  

Concerns over an aging workforce almost doubled over the past year, as the industry faces a much-publicized wave of retirements. I think AI will play a major role in addressing the issue by capturing the knowledge that in years past would have simply walked out the door, by facilitating part-time work by those who would have otherwise retired, and even by attracting some younger, tech-hungry talent. 

The survey found: "Beyond AI, the broader emphasis on technological advancement has also experienced significant growth. Forty-one percent of respondents now view technological advancement as a top social and environmental priority, continuing its rise from only 12% in 2021." AI will be key on this issue, too, as it will for operational efficiencies, which for the second year in a row top the list of operational priorities. 

I encourage you to not only read the full version of the report available to the public but to also consider attending the IIS's Global Insurance Forum, which will explore the key issues in the report. This year, it's being held Oct. 26 and 27 near Zurich.

At the GIF, we at ITL will join our colleagues at the IIS and our sponsor, Lloyd's, in handing out the Global Innovation Awards. The finalists are:

Life/Health/Retirement Innovator of the Year

• Irish Life: AI Reasoning Assistant for Claims
  A multimodal AI solution that automates document processing, accelerates claims assessments, and enhances accuracy, efficiency, and customer satisfaction during critical moments.
   
• RGA: MedScreen+: AI Solution That Strengthens Digital Underwriting
  An AI- and OCR-powered underwriting solution that digitizes and standardizes health records, enabling faster, more accurate, and customer-friendly life insurance decisions while bridging the gap between centuries-old practices and the modern era.

Property/Casualty Innovator of the Year

• Gallagher: Global Data & Technology Capabilities
  A data and analytics platform that simplifies complex industry insights into rates, losses, and limits, empowering clients, brokers, and partners with timely intelligence to make confident, informed decisions.
   
• Tawuniya: End-to-End Motor Claims Transformation
  A fully digital motor claims platform that streamlines the entire journey from accident reporting to settlement, delivering faster processing, greater transparency, and higher customer satisfaction for over 1 million users annually.

Insurtech Predict & Prevent Innovator of the Year

• AXA: Transforming Insurance at a Time of Polycrisis 

  A one-stop, AI-powered ecosystem that unites real-time catastrophe intelligence, cyber defense, risk tools, and training to help organizations predict, prevent, and respond to today’s connected crises. 

• Tawuniya:  Vitality & Drive: A Lifestyle-Based Ecosystem
  Leveraging gamified insurance, health apps, and telematics to engage 500,000-plus users in safer driving and healthier living, cutting accidents by over 25% and reducing claims while building resilience.

They're all deserving. I'll be fascinated to see who wins.

Cheers,

Paul

The reason everyone seems to have a scam story these days isn't due to an increase in reporting; it's because scams have become a near-universal experience. As underscored by a significant increase in the past five years of both reported incidents to the FBI and mainstream media coverage, scams are more frequent, more costly, and more difficult to discern than ever before.

The State of Scams USA 2025 report, conducted by the Global Anti-Scam Alliance (GASA) and sponsored by Iris Powered by Generali, showed that 77% of American consumers encounter scams on a daily basis, with over 70% indicating they had been scammed in the last 12 months. The report also found that one in five Americans lost money to a scam in the same time frame, with an average of over a thousand dollars lost per person and over $64 billion stolen in total.

With scams and fraud on the rise, consumers have turned to institutions and communication platforms for help. Almost three in four (74%) respondents who had experienced a scam reported it to an authority or company for assistance. That is consistent with the findings of the Iris 2025 Identity and Cybersecurity Concerns survey (“ICC”) conducted in April, which found that most consumers reach out directly to companies that have been part of a data breach. However, over half the time, nothing is done – with 57% of reported incidents having no discernible action taken. Even worse, of the 82% of U.S. consumers who reported scams to payment services or financial institutions, less than half (44%) were able to partially recover money in the end, and 38% received nothing back at all.

This gap between consumer action and institutional response feeds a dangerous sense of futility: if reporting scams doesn’t lead to meaningful outcomes, why report at all? This mentality can allow scammers to gain the upper hand. Americans need an ally in the fight to defend themselves against scammers, and they’re expecting financial and communications platforms to step up.

Digital Platforms Top the List for Scammer Channels

By and large, scammers are targeting consumers digitally. Most consumers reported encountering scams via SMS messenger, followed closely by emails and phone calls. Americans reported that 82% of scam attempts occurred on platforms with direct messaging capabilities, including social media, instant messengers, online marketplaces, and even digital ads.

In terms of platforms, Gmail ranked highest in reported instances at 45%, followed closely by Facebook at 41%. TikTok, Snapchat, and X (Twitter) ranked notably lower, but consumers tended to take the longest to recognize that they were being scammed on those platforms.

Consumers are offered little recourse through the platforms themselves. Recent reports indicate that large social or digital communications platforms can take weeks to act when scams are reported. This lack of urgency contributes to an erosion of consumer trust.

Most Lose Money Through Debit Cards and PayPal

Debit cards were the most common method used by scammers, accounting for 30% of reported losses to fraud, followed by PayPal at 25% and credit card payments at 23%. When fraud occurred, most consumers discovered it themselves: 66% discovered it on their own, while only 14% were alerted by their bank or financial services provider.

Americans who were affected overwhelmingly reported the fraud to banks or payment services, with 82% reaching out for support once they realized they had been scammed. But again, this ultimately had mediocre returns for consumers. Likewise, according to Iris' ICC survey, 46% of Americans say their first call would be to their bank after receiving a notification of a data breach, making it their top choice.

These patterns make it clear that consumers view banks and payment platforms as their frontline defense. But when response and recovery prove insufficient, trust is eroded.

Consumers Blame Commercial Organizations – But U.S. Laws Don't

Consumer protection authorities are contacted only 12% of the time, compared with banks or payments services at 25%.

While one in three Americans believe that commercial organizations should be responsible for protecting consumers, U.S. laws and regulations don't agree. For instance, authorized user payments, such as those through platforms like Zelle or Venmo, have no legal requirement for banks to reimburse customers. Additionally, newer scams like imposter scams or AI/deepfake scams are not covered by older FTC regulations and U.S. laws, leading to confusion and denials from banks to reimburse victims.

Where Third-Party Identity Protection Services Fill the Gap

Consumers need stronger support in the fight to protect their identities – and wallets – online. Yet major commercial organizations and digital communications platforms are failing to provide adequate protection.

Iris' ICC survey found that most consumers want a comprehensive, all-in-one solution and are willing to pay for it. While just three in 10 Americans indicated they follow all recommended data protection practices, close to eight in 10 said they would likely use identity protection features if they were integrated into an app they already use, with banks and credit card providers being one of their top picks to purchase from.

Third-party identity protection solutions help close critical gaps. Tools that monitor for compromised data on the dark web, help to spot scams, and offer expert fraud recovery services aren't new but are increasingly sought after. These services not only accelerate resolution by managing outreach to banks and authorities but also help ease the emotional toll of falling victim to scams – a cost that's often overlooked. Additionally, they often take critical steps on the consumer's behalf to prevent further damage.

Consumers want accountability from today's institutions, but they also want protection and peace of mind. There are bills currently under review by the U.S. Congress – like the Protecting Consumers from Payment Scams Act – that are aimed at addressing accountability gaps with banks and payment providers. But responsible businesses shouldn't wait for laws to catch up with the rising threats; they should show up today for their customers and be the ally they need by offering protection.

Not only is it the right thing to do – but it's a powerful investment in customer loyalty and trust.

Walk through any insurtech conference today and you'll hear "agentic AI" mentioned at every turn. Every vendor booth promises autonomous systems that can think, act, and learn. But when you examine these solutions more closely, many turn out to be large language model (LLM) implementations with intelligent automation added. These are valuable advances, certainly, but not the autonomous agents they claim to be.

This terminology inflation creates a fundamental problem. When insurance executives hear every vendor claiming to have "agentic AI," the market becomes so cluttered that companies that invest in building these new capabilities get lost among rebranded automations.

Defining Agentic AI

Part of the problem is that there's no uniform definition of what makes AI "agentic." Different experts emphasize different aspects: Some focus on autonomous decision-making, others on learning capabilities, and still others on goal-directed behavior. But while the exact boundaries remain fuzzy, we can certainly identify what agentic AI is not.

It's not just a chatbot with a fancy prompt. It's not a series of LLMs strung together with if-then logic. And it's definitely not traditional automation rebranded as AI.

One proposal is that true agency requires at least three core capabilities:

1. Tool usage - the ability to navigate and interact with different systems
2. Memory - maintaining context and learning from past interactions
3. Real-time adaptation - adjusting approach based on results when something unexpected happens

The coding assistants like Cursor and Claude Code offer a useful reference point. These tools represent the current state of the art in AI, and most industry observers would comfortably label them as "agentic." If these are our benchmarks for genuine agency, the gap with most "agentic AI" solutions in insurance becomes clear.

This distinction matters because it reveals a spectrum. On one end, you have simple automation following predetermined paths. On the other end, you have fully autonomous systems that set their own objectives and continuously evolve. Most of what's being called "agentic" in insurance today sits firmly at the automation end, despite the marketing claims.

Current Market Examples

The evidence for this is everywhere. Take one major claims administrator's recent announcement of their "agentic AI" solution. Dig deeper, and it's a bundle of voice bots, intelligent document processing, and some alerting.

Another prominent vendor markets six different "AI agents" as part of their agentic platform. Remove the marketing speak, and you find a data layer with LLMs for document routing, a chatbot that accesses internal data, and template generation with compliance checks. These are often solid implementations that deliver real value, but they're a far cry from being truly "agentic."

The Market Distortion

The pressure to appear cutting-edge creates an arms race of terminology. When every vendor feels compelled to claim "agentic AI" to stay competitive, an insurtech that invested heavily in genuine foundations—tool usage, memory, and real-time adaptation—gets lumped in with one that simply added "agent" to their chatbot's name.

This creates an unfortunate dynamic. Insurance executives face an impossible task: evaluating solutions when every vendor uses the same terminology for vastly different capabilities. Even sophisticated buyers struggle to identify which systems will grow into true agency as AI matures versus those that are essentially dead ends with fancy names.

When implementations fall short of vendor promises, it naturally reinforces skepticism about AI investments. The insurtechs building for the future get caught in this backlash, making meaningful transformation even more challenging. Everyone loses: Buyers miss out on genuinely transformative technology, real innovators struggle to differentiate themselves, and the industry's digital evolution slows to a crawl.

The Path Forward

The insurance industry doesn't need to claim false sophistication. Current AI applications can provide tremendous value. Intelligent document processing saves countless hours. Well-designed chatbots genuinely improve customer experience. Predictive analytics enhances decision-making in measurable ways. These are powerful tools that augment human capabilities. The industry benefits when we accurately describe what these tools accomplish and match them to appropriate use cases.

For those evaluating solutions, start with a more fundamental question: Do you actually need agentic AI? If your goal is to reduce document processing time by 80%, intelligent automation might be exactly what you need. If you want to improve first-call resolution rates, a well-designed LLM-powered chatbot could be the perfect solution. These aren't agentic, but they solve real problems with proven technology available today.

Reserve the search for true agentic capabilities for problems that actually require them: complex claims that need dynamic investigation across multiple systems, underwriting decisions that must adapt to unique scenarios in real-time, or fraud detection that needs to evolve its approach as schemes change. For these use cases, ask the hard questions: Can this system actually use tools to solve problems? Does it maintain context across interactions? Can it adapt when things go wrong?

As agentic AI capabilities mature, they will transform how we handle claims, assess risk, and serve customers. But we'll only realize that potential if we're honest about where we are today and deliberate about where we're investing for tomorrow.

As buyers and builders, we all have a role in maintaining clarity about what AI can actually accomplish. This ensures that success goes to companies building on real capabilities rather than marketing claims, while preserving confidence in AI's genuine transformative potential.