In the previous entry of the series, Bartek Antoniak shared his market-wide observations from the perspective of a professional software engineering and consulting firm. In the second article in the series, Peter Ratcliffe takes a deep dive into how AI has already reshaped the insurance industry, particularly the underwriting process.
My business career has spanned nearly twenty-five years, allowing me to witness significant shifts in traditional financial markets and how they operate. I have seen the rise of open-source, the development and implementation of algorithmic systems, and, of course, Data Science and Machine Learning.
This journey has been very interesting. It has enabled me to analyse and understand how the businesses I have partnered with have adapted to and embraced digitalisation.
One particularly intriguing area of the market is the insurance sector.
Insurance for loans and cargo is one of the oldest financial instruments in the world, and it originated with Babylonian bottomry contracts. These contracts provided maritime insurance for hardworking sailors who invested significant amounts of money in transporting cargo across the seas. Understandably, the process for creating these agreements was much less advanced compared to today’s standards. The underwriters of the day, “contract backers,” still had to grapple with making the right decisions while balancing risk and reward.
Many of the challenges from those early contracts still echo through the insurance industry today. While the term 'underwriter' didn’t enter everyday use until the 17th century, these early contract backers were already grappling with the same core issues of risk, trust, and enforcement just like the modern-day underwriter. The key factors to success for these early contracts depended on the ability to model risk as accurately as possible based on the available data, manual effort, and intuition, offering fair pricing to be competitive, and developing trust with customers for renewals, alongside managing bogus claims.
Sound familiar?
Even now, in the 21st century, the dynamics of this industry have not really changed.
Insurance is, at its core, a powerful and important industry to be part of — because it offers customers something deeply valuable: “security”. It’s a promise that when things go wrong, someone has your back. That is a big deal!
Today, data is absolutely king, and in this data-rich world, we hope that the often-overlooked underwriter can perform their job more effectively and achieve a higher level of accuracy. Unfortunately, this is not yet the case. Insurance underwriters still cannot access quality data at any velocity; they are often bogged down by manual and laborious tasks and frequently have to rely solely on their own judgment to make the best decisions. There is now more data and so more opportunities, but there are still blockers.
In summary, I want to emphasise that the issues surrounding insurance have not yet been fully resolved. However, I am confident that solutions will come very soon. The technological advancements I have observed over the years are nothing compared to our significant progress in just the last six months…. It's an exciting time to be alive!
This recent surge in innovation, particularly in areas like data analytics, artificial intelligence, and automation, is rapidly transforming how insurance products are developed, priced, and delivered. The industry is moving towards more personalised and dynamic models, leveraging real-time data to create highly customised policies that better reflect individual risks and customer needs. This change promises to address existing inefficiencies and opens avenues for new product offerings and a more equitable and responsive insurance landscape.
In this article, I will focus on how operating models must evolve alongside new technology, specifically a flavour of AI called “Agentic AI”. I also plan to provide insight into some historical constraints affecting operating models and highlight specific macro & micro-level operational changes that may emerge in the near future, specifically related to Agentic AI and how humans take part in this journey.
We are all on the brink of witnessing significant “systemic” change to business and the markets in which they operate.
This is primarily driven by the exponential promise of AI, particularly Agentic AI. The emergence of AI represents a significant and exciting shift for many of us, offering new possibilities that go beyond traditional approaches. It's a fascinating time to be the Human in the Loop!
Human-in-loop (HITL) is an industry standard term supporting the idea that humans will stay actively involved in the decision-making process, working alongside AI systems to build, guide, correct, govern, or approve AI actions.
Right now, humans are actively unlocking the full potential of their data, which has enabled a pivot to more intelligent systems that can mimic and act like their human creators.
The journey over recent decades has been marked by stops and starts, with data as the central theme driving this paradigm and technology finally catching up. Markets, businesses, and governments initially focused on collecting and managing data — tackling issues of quality, veracity, and volume as the first port of call.
By the 2000s, this had evolved into the Big Data explosion, a period that laid the groundwork for the rise of modern data science. Emerging data scientists began to explore new techniques. The use of mature machine learning algorithms, deep learning architectures, and neural networks unlocked an unprecedented expansion of data-focused capability.
The tech advancements mentioned above have really helped shape the kind of AI we see today. At this point, AI is everywhere and being used to tackle just about every problem you can think of. There’s a wide range of ways people go about building AI and figuring out what problems it can help to solve.
A primary motivation for companies to adopt AI is to enhance efficiency, particularly in sectors like manufacturing, where key improvements are (e.g., predictive maintenance, supply chain optimisation) and medical research (e.g., drug discovery, diagnostic imaging). Let's just say AI is changing the game completely!
I would venture to say that all industries could benefit from AI, but the insurance industry is a prime candidate for this innovation, especially when considering the use case for Agentic AI.
Why?
Over time, numerous businesses have tried to replicate an underwriter's decision-making in a computer, with varying degrees of success. Ultimately, underwriting involves subtle human-level processes; it is highly intuitive, and decisions often rely on an underwriter's instinct or years of market experience. Therefore, certain elements of decision-making and actioning should likely remain under human control. However, that being said, automation and capabilities that can be unlocked through Agentic AI, as this maps quite well to this particular problem space, especially so when thinking about gathering data, or we are thinking about automatons being set up to manage some very laborious tasks so that highly skilled underwriters can do other things more effectively. … It's a perfect fit.
Agentic AI enables a workplace to be enhanced by robotic teams called“Agents”. These robotic teams can interact with humans and interact among themselves. Thus, creating a toolkit from which business staff can flourish and benefit. That sounds like a great place to be!
Within the Underwriting domain, this Agentic system will be built with the underwriters' needs in mind and targeted primarily to take over the repetitive and time-consuming tasks. However, these Agents can also take on some of the more complex tasks, which are usually harder for people to manage on their own. The UX would be designed and mapped by the builders of the Agentic system with the support of the underwriting teams, in order to cover many processes across the Underwriting lifecycle.
For example, Agentic AI can automate the submission process and also look through large amounts of data at the same time to pick up on patterns or trends that most Humans wouldn’t notice, such as unusual behaviour. This can be really useful in areas like fraud detection or system monitoring
On top of that, the Agentic AI can be trained to provide personalised recommendations to help guide human decision-making, whether that’s suggesting the next best action in a workflow or offering ideas based on someone’s pre-programmed preferences. These kinds of features are what make Agentic AI especially valuable, because the systems, once guided by the underwriter, are not just completing simple tasks; they are helping people make better choices and save time by handling the processes that are either too overwhelming or too detailed to figure out by a human alone.
The promise is not only in the automation itself, but also in the time that is created, which could be reutilised and repurposed for other, more business-critical things.
On the micro level, Agentic AI systems are expected to really change how underwriting works, primarily from a mechanics perspective. With some added functionality as the Agentic AI evolves, this is a significant win for the insurance business. The process will still be familiar to the underwriters and their team; however, it will be optimised in various ways.
Below, I provide a brief summary of changes that may occur, “though not exhaustive”.
As a whole, insurance operations are also going to be entirely transformed by Agentic systems. IT teams, operational teams, product teams, and leadership in areas like risk and compliance will all start to see the benefits of having a more connected and intelligent system. Identity and knowledge integration will feel a lot more seamless, and the systems will be way more tightly built together than what’s in place right now. Agentic AI will basically sit at the core of the business, using all the data already available and pulling in new data from the market to create even more valuable insights. The possibilities are enormous, and the boost in efficiency will be clear across every human doing the real work.
In summary, this shift allows underwriters and ops specialists to focus more on high-value work, like making smart, informed business calls.
Today, some outliers are reshaping how insurance operates.
We've seen real momentum in both commercial and personal lines, from disruptors like Lemonade to established players like Aviva. Those deploying modern tech to deliver the same product have paid dividends! The real transformation lies not in the UI or chatbot or flashy website, but in the mechanics of the products themselves and in the advancement of analytical and automation capability, which enables both speed and precision. The heavy lifting and much of this unseen effort is happening behind the scenes. These endeavours cover many bases, from domain-specific innovation to changes in how businesses engage with the broader marketplace.
Underwriting, in particular, is a domain that is deeply systemised, spreadsheet-driven, and steeped in legacy processes and procedures that are key to the required process. Slowly but surely, we are witnessing a shift from the old ways and the birth of new approaches.
Naturally, there are reasons why people cling to old tools like Excel: familiarity and inertia. But that’s no longer good enough. The pressure to modernise is no longer optional. It’s market-driven and driven by disruption.
The birth of the Digital -Native Insurance model and its effects on the status quo
As mentioned earlier, the disruption happening across insurance is affecting the entire industry. Long-time leaders in the space are now starting to see this as a real chance to shift the market in their favour, and they’re using that pressure to modernise, to push for much-needed changes. At the same time, a new group of fully digital businesses is starting to take shape. These companies aren’t held back by old systems or outdated processes like the more traditional insurers are.
Instead of trying to fix the past, these digital-first companies are building strong, modern tech platforms that can actually compete with or even outperform those of the bigger, more established players. And the changes they’re bringing are making a real impact. This new model is forcing legacy companies to take a hard look at how they operate, rethink how they reach customers, and face the fact that everything is moving a lot faster now. But it’s not just affecting a few companies here and there. What’s happening is reshaping the entire industry from top to bottom.
For example, the influence of the digital native approach is impacting M&A activity, and investment conversations now involve quite familiar questions. Do we/you use AI? Not in theory, but in practice. How does it support our/your decision-making? How does it reduce risk for you /us? How does it help our/your staff do better … and make/save money?
For digital-native start-ups, these questions are simple to answer; however, more traditional businesses have to cope with limitations of current business models and outdated technology, making the claim harder to prove.
In the battle between Balance book vs Balance book + Innovation (including AI) I am sure I know who the winner shall be!
M&A activity, for example, has traditionally been driven by scale, cost synergies, portfolio diversification, and distribution power. However, we now see more value attributed to an M&A target’s ability to change and innovate post-acquisition to increase returns within the investment horizon.
Of course, innovation alone doesn’t make a business valuable. It’s the outcome of that innovation that drives profit. Whether you're building a defensible technology/IP moat, pursuing innovations that drive strong premium growth and improve loss and expense ratios, or simply leveraging advancements in AI and automation to be faster, smarter, and leaner “the value lies in execution, not just intention” Investors are no longer just buying today's balance sheet ”they're also betting on tomorrow’s operating model”.
This raises the question of where companies should start. It’s essential to recognise that implementing new technology is insufficient; investing in and developing a modern operating model must occur in lockstep because technology, people, and processes are truly intertwined as part of the future business model.
The Operating Model (OM) serves as a fundamental framework that outlines how people, functions, processes, technology, and data work together to create value for the business. This encompasses various functional areas, including customer interactions, partner management, IT, risk management, operations, and analytical and transactional activities.
It is well understood that regularly testing different approaches to the OM is critical; however, this practice is often neglected, leading to gaps that market disruptors can exploit. It's important to understand that changes in OM can either enhance or negatively impact a business. Therefore, managing these changes effectively and supporting this process on an enterprise scale is crucial.
This requires strong change management, well-defined processes, and technology to meet this need, such as a flavour of AI and its use. Additionally, it necessitates thoughtful consideration of the marketplace in which you operate, the demographics of the customers you serve, and, importantly, how your organisation's employees can adapt to new technology and new ways of working within your new OM.
OM’s have consistently evolved throughout history, from the industrial to the internet era. However, the current pace of change is exponential, representing an unprecedented evolutionary step. The integration of technology and business strategy is now commonplace, and properly harnessing this synergy is crucial for business growth, sales, and maintenance.
Significant challenges in the insurance sector have prompted a reevaluation of traditional business operations. This digital-native approach requires deeper customer—and user-focused centricity. Lemonade, a disruptive force in the insurance industry, is an excellent example. Of course, a company that has like Lemonade has had its stuttering moments; however, it is a good example all the same!
A great example is Lemonade, a truly disruptive force in the insurance industry
Traditional Insurance
Lemonade
Profit from unused premiums
→ Takes a flat fee; leftover funds go to charity
Human-driven underwriting
→ AI-powered underwriting and claims
Long claims process
→ AI bot reviews/approves small claims instantly
Customer service-centric
→ App-first, self-serve digital experience
The key point here is that in a single year, Lemonade saw an increase in premium revenue, rising from $794 million to $944 million. They also improved their loss ratios from 83% to 73%, with claim costs dropping by 26% year-over-year in the fourth quarter. Additionally, headcount remained almost flat during this period. This demonstrates a story of efficient growth, driven by smart underwriting, operational discipline, and effective use of digitisation and data analytics, including artificial intelligence and OM.
Lemonade has achieved this because it dared to challenge and disrupt the most traditional of markets, with a key focus on innovating the OM and the business model as a whole.
“In an earnings call”
Lemonade’s co-founder and CEO said, “It’s (Lemonade) has been built for AI since day one.” He emphasised that without integrating AI deeply, where the tech has access to comprehensive, entwined data, and stated it would be tough to extract the kind of “deep insights” that Lemonade has structured its business upon.
Here are the key areas of focus identified for achieving their goals:
Strategic Reassessment: They reassessed traditional operational strategies to identify fundamental necessary changes.
Market Understanding & Brand Building: They recognised the importance of understanding demographics, addressing the lack of trust, and developing a strong brand presence within the insurance market.
Technological Modernisation: They aimed to apply modern technology and practices across the entire enterprise.
Core Adoption of Advanced Technologies: They focused on adopting automation, Data Science, and AI as core components of their operations.
Behavioural Economics Integration: They emphasised applying Behavioural Economics-Driven Design
Future-Ready Operating Model (OM) Development: They were committed to building an operating model that is prepared for future challenges and opportunities. Lemonade truly stands out
Let's all just accept that the status quo was challenged yesterday
By bringing in new and advanced tech, the OM can move past old limitations, make work smoother, improve decisions, create a space where new ideas are always coming up, and enable businesses to compete.
At a macro level, the focus extends beyond mere efficiency gains. It involves transforming market dynamics and introducing novel value propositions that challenge conventional approaches.
Businesses need to realise that building a new and innovative operating model means using and applying technologies in a much more creative and far more deeply integrated fashion. This isn’t just about adding tools; it means reworking processes and systems that might’ve seemed solid and stable before to develop the capabilities that are now expected in today's insurance market.
Take the impact of Agentic AI as an example. It’s already changing how Insurance processes work by augmenting a manual and strained process with much-needed Automation. This, in turn, creates much-needed space for underwriters, Insurance ops staff, and business leaders to focus on more important things for the business.
Once solutions like Agentic AI become part of everyday workflows, they really shift people's roles away from traditional responsibilities and turn highly manual tasks into something quite different. Every part of the business systems and processes changes in some way, and the staff involved will feel that change in one way or another. In most cases, the experience will be a positive one!
Indeed, on the Macro level, these changes will be vast and generally enable more cohesion across separated domains, eventually breaking down longstanding silos.
Consequently, key business functions will progressively become AI-enabled, with each area undergoing distinct transformations.
Here are several examples that illustrate these changes at the macro level
In summary, businesses need to fully commit to using innovative technologies like AI to build a truly innovative OM. This isn’t just about adding new tools; it’s about rethinking and reworking even the most established processes to meet today’s need for speed and flexibility. As AI becomes a core part of work, it reshapes roles, systems, and decision-making across the board.
The Importance of a true Human-AI Partnership “Human-in-the-Loop (HITL)”
On a broader level, companies are training leaders and staff to adapt to these changes and adjusting their strategies to make AI a central part of their operations. The result is a more connected, AI-enabled way of working that’s changing the foundation of the operating model itself.
This transition involves a move from traditional, manual processes to a more analytical and technology-centric approach, where human expertise is supported by artificial intelligence. Critical workflow requires reevaluation of existing skill sets, experience, and know-how within the Insurance profession to support an AI-driven future.
This is a pivotal moment for the insurance sector, grappling with a well-documented talent drain and an outdated image that hinders attraction. Embracing new technology is crucial for drawing in and retaining top-tier talent, as human capital remains our most invaluable asset.
Below, I provide a few examples for consideration, highlighting key areas, gaps where skills will need to evolve, and high-level indicators of how to do this. Not exhaustive, of course!
The purpose of this change and this exercise in general is to ensure that humans who work for your business are empowered to get the most out of the AI that is supporting them, putting their best foot forward to enable your staff members, employees, and even board members to truly trust AI while also understanding the potential, risks, and inner workings of AI.
Despite widespread concerns about automation replacing human jobs, a balance exists between AI and human involvement. The "human-in-the-loop" approach ensures that AI is developed, governed, and managed by humans, thereby retaining a level of control for both employees and the wider business.
There is a fine balance, of course, because some elements of work processes could be fully automated without being overseen by humans. However, key elements will always need human interaction or intervention. When we consider risks, ethics, and controls, a solid framework needs to be in place to support ongoing AI initiatives, which will inevitably be led by humans.
Ergo HITL is seen by many as the most effective way to gain the full value from Agentic AI transformation and to ensure a balance remains.
Also, while Agentic AI can automate and optimise at unprecedented speed, there are clear limits to what it can achieve on its own, and those limits are most visible where context, judgement, and accountability matter most. This is where humans excel.
For example:
Ethical complexity: Real-world scenarios often present subtle ethical dilemmas that AI alone may misinterpret.
Data limitations: Agentic AI models are trained on historical data, which can skew interpretation and limit adaptability to novel situations.
Pattern recognition gaps: While Agentic AI excels at identifying patterns, rare edge cases and ambiguous problems can still elude agents.
Bias mitigation: AI can inadvertently reinforce biases present in training data unless actively monitored and corrected.
Trust and accountability: Maintaining human oversight ensures that critical decisions remain transparent, explainable, and attributable.
HITL adds a feedback and oversight layer that helps protect business outcomes. It gives organisations the tools to step in, guide, and correct when needed, so the business can:
Responsibly use AI without slowing things down
Combine human judgment with fast machine decisions to improve how choices get made
Stay compliant and keep trust with stakeholders
At the end of the day, HITL does not hold anything back. It’s This approach is vital for Agentic AI to work well in complex situations where the stakes are incredibly high!
The implementation of HITL can take different forms depending on task complexity, risk appetite, and operating model maturity. Below, I share brief details of model types. These are not mutually exclusive; they are interlinked and often coexist throughout the process of developing a new AI-enabled operating model.
Network Security Monitoring: Policyholder & Underwriting Data Protection Monitoring
Large-Scale Data Processing: (Mostly Routine)High-Volume Policy & Claims Processing Automation
Alert Fatigue: Too many false positives can lead to human supervisors ignoring critical alerts.
Clear Escalation Paths: Well-defined protocols for human intervention are essential.
Interpretability: Humans need to understand why the AI made a certain decision to effectively intervene.
In this model, the AI works on its own, but people keep an eye on how it is performing and get alerts if something unusual happens, if it makes a mistake, or if it takes a high-risk action.
Human-as-Validator/Reviewer
Medical Diagnoses: Automated Medical Assessment for Health & Life Claims
AI's "Confidence Score": The AI should be able to assess its own confidence in a decision and hand off when low.
Seamless Handover: Smooth transitions from AI to human are crucial for user experience.
Learning from Exceptions: How are these exceptions documented and used to improve the AI?
The AI takes care of routine tasks, and people step in only when it runs into something it cannot handle or that is outside its set limits.
No matter which HITL model is used “more”, there are a few key things that are important for it to work well:
Clear role definition: Set clear lines between what the AI handles and what people handle in underwriting, claims, fraud checks, and customer service, with responsibilities documented and easy to audit
Continuous Feedback Loop: Establish a system for capturing human corrections, insights, and contextual information. This feedback should be used to continuously improve the AI's accuracy and decision-making. Additionally, a clear escalation process for AI agents should be implemented to hand off complex or ambiguous cases or instances of error to human oversight.
Explainable outputs: Make sure AI decisions can be clearly understood by underwriters, claims teams, and compliance, so they can see the reasoning behind each recommendation
Focused training: Give teams the skills to use AI tools well, understand where they have limits, and manage the ethical issues that come up in insurance work
Ethical design: Build in bias checks, fairness testing, and safeguards to prevent harm, protect customers, and maintain trust from the start
Step-by-step rollout: Begin with safe, high-volume tasks like document sorting or early fraud checks, then grow AI’s role as trust and performance improve
Compliance first: Keep AI processes in line with all insurance regulations, privacy laws, and industry standards, with compliance built into everyday workflows
In summary, Agentic AI isn’t meant to replace humans altogether. It’s really about helping people do more by taking care of the repetitive and laborious process or by helping humans discover additional insights, so we humans can spend more time on the creative, complex, and essential work.