Looking for “rights” and
“wrongs” in the AI future
[O]ur inner critic often keeps us in one place, feeling both stagnant
and frustrated, but also safe and predictable. It’s a relationship
worth exploring—and challenging—if that feels right for you.
By Alan Demers
Artificial Intelligence (AI) now permeates business and daily life as a tool, a substitute for humans, or a subject of experimentation. This landscape is fraught with contradiction, making it difficult to untangle hype from reality. The argument over whether AI will replace human jobs or create new ones illustrates just one area of ambiguity. Billions of dollars of AI investments from energy production, data centers, and the likes of Anthropic’s revenue curve speak volumes about demand and adoption.
Sizing up the state of AI in property/casualty insurance is similarly challenging, with many debating if AI is a tool or if it is transformational. Perhaps the proper answer is “both.” More nuanced answers involve when, how much, and to what degree, all of which are (or should be) tied to return on investment (ROI). Either way, AI tools apply across the entire insurance value lifecycle, including risk management and combating fraud.
Deploying agentic AI to operate as an insurance agent, however, is even more debatable, despite its use in today’s insurance sales process. Some go so far as to tout an AI quote-and-bind.
Tool or transformation
It’s safe to say that AI enthusiasm within the insurance industry has outpaced measurable uplift at present. This isn’t about predicting the future or AI disappointing but rather a reflection of how things stand today in the early years of AI in insurance. Conflicting signals about successes and setbacks only distort the picture.
The scope of AI usage, specifically in fraud fighting and risk management, is quite extensive. From finding to binding insurance and the attendant underwriting functions of risk transfer, to avoiding, detecting, and investigating risk and fraud, these are fertile grounds. Since AI is purpose-built to analyze large volumes of data, there is good reason for optimism when it comes to accuracy, speed, and pattern identification alone.
Fraud: A moving target
Fraud detection has improved with the introduction of so-called real-time technology, such as Shift and FRISS tools. Alerts are created to signal potential issues or “red flags” that agents, underwriters, and claim adjusters can act upon. Clearspeed was developed to analyze voice and score risk for early action. Senzing provides “entity resolution,” critical in today’s world in order to have confidence in data, relationships, and simply to know who is who. These are just a few examples on top of numerous tools and capabilities provided by insurance ecosystem incumbents like LexisNexis, Verisk, and many more.
Insurance fraud continues to be a moving target, especially when technology advances, and bad actors become early adopters ahead of insurers. Cybercrime, AI hacking, and the manipulation of records, images, and information are problematic and, on top of traditional fraud types, leave insurers in a constant state of catch-up.
Obstacles in fighting fraud
The promise of AI is to identify risks that humans miss—with greater accuracy and efficiency. After all, the impacts are significant, including uncollected premiums, overpaid claims, and unprofitable business. Inconsistency among agents, adjusters, and others, either through overlooking or ignoring warning signs, often results in low volumes and missed investigation referrals.
However, the end goal cannot be solely about increased alert volume. False-positive alerts divert time and resources, and large volumes are impractical to investigate. This is particularly true among straight-through-processing underwriting and claim models competing in the market for speed and efficiency.
There are additional constraints when it comes to investigative resources, not to mention reliance on law enforcement and the judicial system when it comes to tackling organized fraud. Detection is one thing, but reaching desired outcomes—like claim denial, premium collection, and restitution—is completely different. Plus, it is highly resource dependent and time consuming and presents potential legal and reputational brand risks.
Either way, in the future AI must fight fraud by concentrating on early avoidance while separating and doing more to support honest customers.
Risk management appeal
AI for risk management has a much broader appeal and is already in play. Opportune use cases involve gathering, analyzing, and sorting data to better evaluate, price, and accept risks. It is equally exciting to consider AI applied to a whole portfolio and an entire line of business for the benefit of future-forward learning and better risk management. At least this is the promise of AI—“connecting the dots” better, faster, and more cheaply than humans can do.
This enthusiasm is reflected in industry projections and surveys. The AI insurance market is expected to grow from over $8 billion to some $59.5 billion over the next eight years. Numerous carrier/CEO surveys suggest increased AI spending and broader use. Meanwhile, financial investments and bullish AI announcements are abundant. To cite a few:
- Chubb suggests a 20% workforce reduction in three to four years due to AI growth;
- GEICO established an AI hub in Palo Alto, California, to transform insurance operations;
- Nationwide committed $1.5 billion to accelerate technology and AI;
- Allstate’s “ALLIE” (Allstate Large Language Intelligent Ecosystem) handles 400,000 customer conversations monthly; and
- Travelers recently announced an AI Claim Assistant using OpenAI to file claims conveniently, efficiently, and effectively.
AI insurance headwinds
There are other voices among the chorus of AI promoters. Regulators are unclear regarding how insurers are using and planning to use AI, seeking clarity while pushing the “NAIC Model Bulletin” to promote standards. Carrier insiders express frustration with internal AI governance and legal barriers that, with the best of intentions, slow or impede progress in data privacy and security.
Investors are hampered by sorting out true AI providers from those who merely claim to be such experts. AI solution providers face off with insurers that are buying and building AI, leading to shorter contract terms as future switching is anticipated, unlike most other “rinse and repeat” enterprise technology agreements.
Agents and customers are among the most impacted when it comes to selling and servicing insurance. Perhaps AI-modeled quoting is an extension of making it easier to shop and compare, first made available online directly. Yet customer behavior indicates a lasting need for human advice and confidence delivered by human agents, that seems to not be going away anytime soon.
AI risk management vision
The risk management discipline relies on volumes of data to identify risk and drive action to quantify, underwrite, price, and service insurance protection. A wider view can include loss control, loss prevention, and portfolio management. At present, AI is helping insurers crunch data faster, with expectations of quicker and better decisions.
Many small business insurers are using some forms of straight-through underwriting based on rules engines. AI can connect to these rules engines to boost accuracy while consuming many of the surrounding manual tasks. Faster decisions are intended to write more business with less expense, making the cost/benefit crystal clear, good for both agents and their customers.
However, risk acceptance/rejection decisions are anchored in carriers’ unique risk and market “appetite,” which are shaped by multiple variables spanning loss experience, cost of capital, market share, and competitive forces, to name a few. Likewise, risk appetite tends to change over time as carriers ebb and flow between growth and profitability priorities. Reinsurance, carrier surplus, and probable maximum loss (PML, or the amount a portfolio may incur from a catastrophe) are also part of the appetite mix.
It’s also safe to say that human experience, knowledge, and feeling are equally blended into such decision-making, making it nearly impossible to decode at any point in time. Such practices only fuel “black box” pricing criticism at a time when AI regulations call for greater transparency. Thus, risk appetite serves as a virtual throttle for growth, and any visions of transformative AI risk management must consider how this throttle works.
AI hurdles/looking ahead
Over the last decade or so, much of the industry has upgraded its core policy administration, billing, and claims systems. Yet, traditional workflows remain. In fact, many of these system implementations were highly customized to retrofit existing practices.
These dated workflows extend to organizational structures, processes, job duties, and data management, inadvertently serving as barriers to the introduction of new technology. Meanwhile, the conventional wisdom is that AI functions depend on the right data in the right amounts.
In roles like agency, underwriting, and claims, perhaps 50% or more of the workload is administrative, from researching and gathering to entering information, updating, documenting, and so on. These routine tasks are forecast to be eventually replaced by agentic AI.
Consequently, new job roles are envisioned to concentrate on relationships and highly critical work. This implies a human resources overhaul, including upskilling, hiring, development and, for starters, heavy doses of change management.
The voices of customers and agents are paramount for adoption and cost/benefit assessment. AI adoption will hinge on principles that carriers intend to achieve. Easy to do business with, and customized solutions, efficiency, and convenience are at the top of the list. Given the negative public perception of AI, it will take significant creativity and promotion to win over these stakeholders.
“AI” has breathed new life into the prospects of most information and technology providers—both big and small—attracting carriers seeking solutions and some simply suffering from FOMO (fear of missing out).
To sum it up, insurance AI transformation has a long way to go, not because of the technology itself, but because of the state of readiness encompassing legacy systems, dated workflows/processes, data preparation, risk appetite pressures, operational and organizational change management needs, along with the people side—employees, customers, agents.
No doubt AI holds the promise of unlocking tremendous value. All of which requires heavy lifting to commensurately change an industry well known for its demonstrated resilience and measured tempo.
The author
Alan Demers, president of InsurTech Consulting, LLC, provides strategic development, subject matter expertise and advisory services. A former executive with more than 35 years of P&C insurance claim experience, Alan collaborates in the forefront of insurtech, partnering with insurance leaders, insurtech start-ups, solution providers and investors to modernize insurance.
As vice president, Claims Innovation at Nationwide, Alan conceptualized a vision and roadmap to build Claims of the Future, focused on automating and digitizing claims experiences, progressing from inception through prototype testing. He served as a founding member of the Corporate Innovation Council and played a key leadership role in establishing goals, best practices and building an innovative culture at Nationwide.
Alan is an accomplished executive leader and has worked for two Fortune-100 insurance companies in a number of corporate, national and regional leadership roles among personal, commercial, non-standard and specialty lines claims. Alan served as head of claims for Nationwide’s Commercial Agribusiness and Non-Standard Claims. Prior roles include, field vice president, regional claims officer, and national catastrophe director. Alan began his career with Aetna Casualty & Surety as a claim adjuster, manager and advancing to corporate claim consultant, prior to joining Nationwide.
