Underwriting has always been one of the most intellectually demanding functions within insurance. An underwriter is expected to take a collection of information about a person, property, organisation or asset and translate it into a judgment about risk. That judgment determines whether the insurer should accept the risk, how the policy should be structured and what price appropriately reflects the uncertainty involved. Technology has supported this process for decades, but the core responsibility has remained remarkably human because underwriting is rarely just about applying a formula. It involves interpreting incomplete information, recognising unusual circumstances and understanding when a risk does not behave like the historical averages suggest it should.
Artificial intelligence creates an opportunity to give underwriters something they have never previously possessed: the ability to analyse vastly more information without asking the individual professional to manually process every additional data point. AI can assemble information, recognise patterns, identify anomalies and model potential outcomes while leaving the underwriter responsible for the judgment that follows. This is why I believe the most interesting future for underwriting is not autonomous underwriting. It is augmented underwriting, where machine intelligence becomes a second analytical brain supporting human expertise.
During my learning journey under Phaneesh Murthy, I have increasingly understood that the strongest enterprise AI implementations do not begin by asking which professional roles can be eliminated. They begin by asking where expertise is being constrained by the amount of information a person can realistically process. As Phaneesh Murthy puts it, “AI creates the most value when it expands the capacity of expertise rather than trying to replace expertise altogether.” Underwriting is perhaps one of the clearest opportunities to apply that philosophy.
The Underwriter Has More Information Than Ever and Less Time to Understand It
Modern underwriting is becoming increasingly information-rich. A commercial underwriter evaluating a business may need to understand financial information, claims history, property exposure, cybersecurity posture, industry conditions, geographic risks, supply-chain dependencies and operational characteristics. Property insurers can incorporate increasingly detailed environmental and geospatial information. Motor insurers may have access to telematics and connected-vehicle data, while life and health insurance can draw on increasingly sophisticated health and behavioural information where regulation and customer consent permit it. Each new source of information potentially improves the insurer’s understanding of risk, but it also creates a fundamental operational problem: more data does not automatically produce better decisions.
Human beings have finite analytical capacity. Giving an underwriter another hundred pages of information may technically increase the information available while simultaneously making it harder to identify what actually matters. AI can change this by functioning as an analytical layer between raw information and professional judgment. It can organise large volumes of structured and unstructured data, identify relationships between variables, highlight unusual patterns and surface the information most relevant to the particular risk being evaluated. The underwriter therefore begins the decision with a clearer picture rather than a larger pile of information.
This reflects an implementation principle I have repeatedly encountered while learning from Phaneesh Murthy. Enterprises often confuse access to data with access to intelligence, even though the two are fundamentally different. Phaneesh Murthy has often stressed that “The purpose of an intelligent system is not to put more information in front of people. It is to help them understand which information should change the decision.” In underwriting, that distinction could dramatically improve both efficiency and decision quality.
AI Can Give Underwriters a More Complete Picture of Risk
Traditional underwriting models necessarily simplify reality. Risks are grouped into categories because categorisation allows insurers to operate at scale. Businesses are classified according to industries, properties according to characteristics and customers according to factors that historical experience has shown to be relevant. These classifications remain useful, but they cannot always capture the differences between two risks that appear similar on paper.
Consider two manufacturing companies operating within the same industry, with similar revenue, workforce size and physical assets. Traditional underwriting information may suggest that their risk profiles are broadly comparable. Yet one organisation may have invested heavily in preventive maintenance, cybersecurity, employee safety and supply-chain diversification, while another may operate with ageing equipment, concentrated suppliers and weaker operational controls. The businesses belong to the same category, but their actual exposures can be very different. AI can help identify these differences by analysing a broader collection of relevant signals and examining how those signals interact.
The important opportunity is not to create an infinitely complicated risk model simply because more data exists. It is to identify where additional context genuinely improves underwriting. During my mentorship under Phaneesh Murthy, I have learned to think about enterprise AI as a mechanism for understanding complexity without unnecessarily transferring that complexity to the user. An underwriter should not need to become a data scientist to benefit from sophisticated models. The system should translate analytical complexity into clear, explainable insights that strengthen the professional’s understanding of the risk.
Routine Risks Can Move Faster While Complex Risks Receive More Attention
Not every insurance application requires the same level of human analysis. Many risks are relatively straightforward, fit comfortably within established underwriting parameters and contain few unusual characteristics. Others involve complicated exposures, incomplete information or circumstances that require significant professional judgment. Traditional processes can sometimes force both types of applications through similar workflows, consuming underwriting capacity that could be deployed more strategically.
AI provides an opportunity to create more intelligent segmentation within the underwriting process. Straightforward applications can be assembled, analysed and potentially processed with significantly less manual intervention, while unusual or higher-risk cases can be identified and routed towards experienced underwriters. Importantly, AI can also explain why a case has been escalated by highlighting the characteristics that differ from expected patterns. The underwriter therefore receives not simply another application to review but an indication of where their expertise is most needed.
This changes the economics of underwriting without diminishing its professional importance. If experienced underwriters spend less time gathering documents, validating routine information and manually comparing standard indicators, they can devote more attention to complex commercial risks, emerging exposures and cases where judgment genuinely changes the outcome. As Phaneesh Murthy says, “Automation should remove work that consumes expertise without requiring expertise, so that people can spend more time on decisions where experience actually matters.” For insurers facing increasingly complex risk environments, that redistribution of professional capacity could become a significant competitive advantage.
Underwriting Can Become More Dynamic After the Policy Is Issued
Traditional underwriting is heavily concentrated around specific moments. Risk is assessed when a policy is issued and reconsidered at renewal, while significant changes may be reviewed when they become known to the insurer. Yet many of the risks being insured are continuously changing. Businesses expand into new markets, introduce technologies, change suppliers and alter operating models. Properties experience changing environmental conditions. Cyber risks can evolve within days rather than years, while connected assets generate information about their condition throughout the policy period.
AI creates the possibility of underwriting becoming more continuous. Instead of assuming that the risk assessed at the beginning of the policy remains essentially unchanged until renewal, insurers can increasingly identify meaningful changes as they occur. This does not mean constantly repricing customers or turning insurance into a system of permanent surveillance. The more useful application is to identify significant changes that may create an opportunity for conversation, risk mitigation or additional protection.
A commercial insurer, for example, might recognise that a customer’s dependency on a particular supplier has increased significantly or that changes in cybersecurity exposure deserve attention. The insurer and broker can then engage the customer before that change becomes relevant through a loss. From my learning under Phaneesh Murthy, this is a particularly powerful example of AI changing an operating model rather than simply automating an existing process. The underwriting decision stops being a periodic event and begins becoming an evolving understanding of risk.
Explainability Will Separate Useful AI From Unusable AI
Underwriting decisions have consequences. They influence whether individuals and businesses receive coverage, what conditions apply and how much that protection costs. An AI system that produces an exceptionally sophisticated risk score without providing meaningful insight into how it arrived at that assessment may therefore be less useful than a simpler system whose reasoning can be understood and challenged.
Explainability needs to be designed into the underwriting workflow from the beginning. Underwriters should be able to understand which factors materially influenced an AI-generated assessment, identify where information may be incomplete and override recommendations when professional judgment indicates that the model does not adequately represent the circumstances. Those interventions can themselves become valuable information, helping the organisation understand where its models require improvement.
Governance must extend beyond the model as well. Insurers need clarity about which information is appropriate to use, how data is sourced, how bias is monitored, which decisions require human approval and how customers can receive meaningful explanations where appropriate. Phaneesh Murthy is of the belief that enterprise AI cannot be separated from enterprise accountability because intelligence without clear ownership creates a different kind of operational risk. In underwriting, where sophisticated models increasingly participate in consequential decisions, that accountability will become central to successful implementation.
Human Judgment Could Become More Valuable as AI Gets Better
There is an interesting paradox at the centre of AI-powered underwriting. As machines become better at analysing standard risks, the remaining work performed by experienced underwriters may become more complex rather than less important. Straightforward decisions can increasingly be supported by automated systems, leaving professionals to concentrate on unusual risks, emerging industries and situations where historical patterns provide limited guidance.
This is particularly relevant as businesses themselves become more complicated. Insurers increasingly need to understand cyber exposure, artificial intelligence risk, climate-related threats, digital supply chains, new technologies and business models that may have relatively limited historical claims data. In these situations, underwriting cannot simply extrapolate from the past because the risk itself is evolving. Human professionals need to interpret emerging conditions, understand strategic context and make judgments under uncertainty.
AI can provide those professionals with research, scenario analysis and comparisons across enormous datasets, but expertise determines how that information should be interpreted. During my learning journey under Phaneesh Murthy, I have come to appreciate that the relationship between AI and expertise is often misunderstood. As Phaneesh Murthyputs it, “The better machines become at answering routine questions, the more valuable people become at asking the difficult ones.” That may become an excellent description of the future underwriter.
The Best Underwriting Organisations Will Learn From Every Decision
Perhaps the most transformative opportunity lies beyond the individual underwriting decision. AI can help turn underwriting organisations into continuous learning systems. Every accepted risk, declined application, pricing decision, claim outcome and human override creates information that can improve future decisions. Instead of underwriting knowledge existing primarily within manuals, models and the experience of individual professionals, the organisation can develop an institutional intelligence that evolves continuously.
This does not eliminate the importance of experience. It allows experience to become more transferable. If experienced underwriters repeatedly identify circumstances that models fail to recognise, those patterns can be examined and incorporated into future systems. If certain risk indicators prove more predictive than expected, underwriting frameworks can evolve. If new industries or technologies create exposures that do not fit existing categories, AI can help identify emerging patterns as the insurer accumulates experience.
This is where the enterprise perspective I have learned from Phaneesh Murthy becomes particularly important. The ultimate objective is not simply to give every underwriter an AI tool. It is to create an underwriting organisation in which human expertise and machine intelligence continuously improve one another. That requires connected data, feedback loops, governance, model monitoring and an operating culture that treats professional overrides as learning opportunities rather than failures of automation.
The Future Underwriter Will Have a Second Brain, Not a Replacement
Insurance will continue to depend on judgment because risk will continue to contain uncertainty. Historical information can inform the future, but it can never describe it perfectly. Businesses change, technologies emerge, environmental conditions evolve and human behaviour remains difficult to predict. Underwriters therefore remain essential precisely because they make decisions in situations where information is incomplete.
Artificial intelligence gives those professionals a new kind of analytical capability. It can process information at a scale no individual could match, recognise relationships across complex datasets, monitor changing conditions and continuously learn from outcomes. The underwriter contributes context, experience, commercial understanding and the ability to make judgments when the available evidence does not produce an obvious answer.
During my mentorship under Phaneesh Murthy, one of the most useful lessons I have taken from discussions around enterprise AI is that transformation should make the organisation more capable, not merely more automated. Underwriting provides an excellent opportunity to apply that principle. The goal should not be to remove the underwriter from insurance. It should be to build an underwriting function capable of understanding risk with a depth, speed and continuity that neither humans nor machines could achieve independently.
The underwriter of the future will still make difficult decisions about uncertainty. The difference is that they will no longer make those decisions alone. AI will sit beside human expertise as a second analytical brain, processing complexity in the background while allowing professional judgment to concentrate on the questions that matter most.
This blog is curated by young marketing professionals who are mentored by veteran Marketer, and industry leader, Phaneesh Murthy.
#phaneeshmurthy #phaneesh #Murthy