Underwriting is the original act of insurance. Before claims, before premiums, before policies, there is the fundamental question on which the entire industry rests: how risky is this person, this property, this business, and what should we charge to take that risk on?
For centuries, the answer to that question came from human judgement. An underwriter would gather what data they could, apply experience and intuition, consult actuarial tables, and make a call. It was a craft as much as a science, and like all crafts dependent on individual human judgement, it was inconsistent, slow, limited by the data a single person could process, and vulnerable to the biases and blind spots that no human entirely escapes.
That era is ending, and it is not coming back. Historically, underwriting relied on manual methods dependent on the expertise and judgement of underwriters, labour-intensive evaluations based on limited data sources. The introduction of AI marks a significant milestone in this journey, offering a way to not only automate tasks but augment human capabilities through enhanced data analysis. The shift is not a refinement of the old model. It is a replacement of its fundamental constraints.
The Structural Problem With Manual Underwriting
To understand why AI underwriting is irreversible, you have to understand what was actually wrong with the manual model, not its surface inefficiency, but its structural limitations.
The first limitation was data bandwidth. A human underwriter can hold only so much information in mind, cross-reference only so many sources, and detect only the correlations that experience has taught them to look for. Traditional underwriting relies on historical data and human judgement, which can result in inconsistencies or overlooked risk factors. AI, by contrast, uses predictive analytics and machine learning to identify hidden correlations, detect subtle patterns, and continuously refine its assessments based on new information. The human cannot see the subtle, non-obvious patterns buried in thousands of data points. The machine can.
The second limitation was consistency. Two underwriters assessing the same application could reach different conclusions. The same underwriter on a Friday afternoon could decide differently than on a Monday morning. This variance is not a moral failing, it is an inherent property of human judgement at scale. And in underwriting, inconsistency translates directly into mispriced risk, which translates into either lost business or losses on the book.
The third limitation was speed. Manual underwriting was slow because it required a human to assemble, read, and evaluate information sequentially. AI has reduced the average underwriting decision time from three to five days to 12.4 minutes for standard policies while maintaining a 99.3% accuracy rate in risk assessment, according to a 2025 technical analysis. For complex policies, AI has reduced processing times by 31% while improving risk assessment accuracy by 43%.
Phaneesh Murthy has frequently made a point that applies directly here: when a process is simultaneously slow, inconsistent, and constrained by the limits of individual human cognition, incremental improvement is the wrong goal. The process needs to be reconceived around a fundamentally different capability. Underwriting was exactly such a process, and AI is exactly such a capability.
The Data Revolution Beneath AI Underwriting
The most consequential change AI brings to underwriting is not the algorithm. It is the sheer breadth of data the algorithm can incorporate, far beyond anything a manual process could ever consult.
Insurers are now empowered to leverage real-time data sourced from wearable devices, Internet of Things sensors, and other digital signals, processing massive datasets to assess risk with unprecedented accuracy. The application form, once the primary input to an underwriting decision, becomes just one source among many.
AI extracts data from public records and historical claims to prefill applications, while machine-learning models analyse historical claims, environmental data, and behavioural patterns to identify risks. For property and mortgage underwriting, AI can research a property’s location, the housing market, nearby sales, and weather data, and study images or video to quantify property features and condition hazards.
This breadth fundamentally changes the granularity of risk assessment. AI-powered risk assessment enables hyper-personalised pricing, shifting insurers from broad demographic segments to behaviour-based, real-time risk profiles. The implications of that shift deserve emphasis. The traditional model priced risk by putting people into broad buckets, age bands, postcode tiers, occupation categories, and charging everyone in the bucket roughly the same. This was crude by necessity. The data and tools to do better did not exist.
Behaviour-based underwriting prices the individual, not the bucket. A careful driver no longer subsidises a reckless one simply because they share a demographic profile. A health-conscious individual is no longer priced as if they shared the risk profile of a sedentary peer. This is not just more profitable for insurers, it is, arguably, more fair, because it ties price to actual risk rather than to membership in a statistical category.
Intelligent Automation: The Underwriter Augmented, Not Erased
It would be easy to read the speed and accuracy gains as a story about replacing underwriters. That reading misunderstands what is actually happening.
The most effective AI underwriting systems are not replacing human underwriters wholesale. They are automating the routine and augmenting the complex. Machine learning programs process applications immediately while predictive analytics spot risks and pricing opportunities automatically, automation that cuts operating costs, reduces human error, and lets underwriting teams handle far more applications without hiring more people. Advanced algorithms can process routine applications in minutes rather than days, while automated workflows route complex cases to appropriate specialists.
This division of labour is the heart of the matter. The straightforward applications, the ones that previously consumed the bulk of underwriting capacity while requiring little genuine judgement, are decided automatically. The complex, ambiguous, high-stakes cases are routed to human specialists who now have the time and the comprehensive data analysis to make better decisions.
This is precisely why hundreds of companies now rely on AI-based underwriting as a second set of eyes to catch details that might otherwise be missed, exemplified by tools like Allianz’s BRIAN, a generative AI underwriter guidance system. The framing of “a second set of eyes” is telling. The AI is not the decision-maker of last resort. It is the tireless analyst that surfaces what the human might miss and handles what the human need not touch.
Phaneesh Murthy’s consistent guidance on automation applies here with particular force: the goal is never automation for its own sake. It is the redeployment of scarce human judgement to the decisions where judgement actually adds value. An underwriting operation that automates the routine and concentrates its experts on the genuinely difficult is not a smaller operation. It is a more capable one.
Predictive Analytics: From Assessing Risk to Anticipating It
The deepest transformation AI brings to underwriting is the shift from assessing risk as a static snapshot to anticipating how it will evolve.
Predictive analytics is the difference between reading a history book and reading a weather forecast, turning raw data into foresight, using statistical science, machine learning, and AI to uncover hidden patterns rather than only looking at the past to identify what went wrong. A manual underwriting decision was, by nature, a point-in-time judgement. The risk was assessed at the moment of application and rarely revisited until renewal.
AI underwriting is dynamic. Machine-learning algorithms continually update outputs using the latest changes to the claimant’s life and the broader market, powering risk models through predictive analytics, dynamic risk scoring, and scenario simulations. The risk profile is not frozen at application. It evolves as new data arrives, allowing insurers to adjust, intervene, and price with a precision that a static model could never achieve.
This anticipatory capability extends beyond pricing into proactive risk management. Predictive analytics groups customers by lifestyle, spending habits, and risk exposure, letting insurers develop targeted products, boost cross-selling, and lower loss rates, staying ahead by rolling out preventive measures, sending timely notifications, or adjusting policies to anticipate claim surges or high-risk conditions.
The strategic value of this is significant. Customer retention improves when churn signals are identified early, allowing proactive outreach before policyholders switch providers. The same predictive engine that prices risk can detect when a valuable customer is at risk of leaving, turning the underwriting data asset into a retention asset.
The Governance Imperative: Why This Cannot Be a Black Box
For all its power, AI underwriting introduces a category of risk that manual underwriting never had to confront at the same scale: the risk of opaque, unfair, or non-compliant automated decisions affecting millions of people.
This is not a reason to retreat from AI underwriting. It is a reason to build it responsibly. Insurers must ensure compliance with regulations such as data protection laws, and implement explainable AI that provides clear insights into decision-making, because explainability enhances accountability and builds confidence in automated assessments.
The explainability requirement is not a bureaucratic inconvenience. It is foundational. Success depends on data quality, model explainability, and fairness, ensuring compliance, regulatory transparency, and long-term customer trust. An underwriting model that declines an applicant or charges a higher premium must be able to articulate why, in terms that satisfy a regulator and that the affected customer can understand. A black-box model that performs beautifully on accuracy but cannot explain its decisions is not a sophisticated asset, it is a liability waiting to surface as a regulatory action or a public trust crisis.
The fairness dimension is equally critical. Models trained on historical data inherit the biases embedded in that history. An underwriting model that learns from decades of decisions made under different assumptions can perpetuate and even amplify discrimination that the insurer would never sanction explicitly. Detecting and correcting for this requires deliberate, ongoing effort, fairness audits, bias monitoring, and a governance structure that treats these as continuous obligations rather than one-time checks.
Phaneesh Murthy has consistently held that the organisations that lead in deploying powerful technology are those that earn the right to operate it, through transparency, governance, and demonstrated fairness. In underwriting, where automated decisions directly shape who gets coverage and at what price, this principle is not optional. It is the precondition for sustainable AI adoption.
What This Means for the Industry
The trajectory is unmistakable. By 2025, AI was expected to become a standard tool in 90% of insurance companies, driving automation beyond claims into policy administration and customer service, enabling deeper insights into customer behaviour and risk patterns, and supporting personalisation at scale. The question facing insurers is no longer whether to adopt AI underwriting, but how quickly and how well.
The competitive implications are stark. An insurer underwriting with AI prices risk more accurately, decides faster, processes more volume at lower cost, and detects fraud and adverse selection that a manual process would miss. An insurer still underwriting manually is, against that competitor, structurally disadvantaged on every dimension that matters, cost, speed, accuracy, and the loss ratio that ultimately determines profitability.
This does not mean the transition is simple. Predictive analytics strengthens digital transformation by enabling touchless underwriting, automated claims triage, and scalable decision intelligence, but success depends on data quality, model explainability, and fairness. The insurers that succeed are those that invest in the data foundation, build the governance infrastructure, and manage the organisational change of moving experienced underwriters from routine processing to high-value judgement.
The Irreversible Shift
There is a reason this article is titled around the word “never.” Some technological shifts are reversible, adopted, found wanting, abandoned. AI underwriting is not one of them.
The reason is simple economics combined with simple capability. Once an insurer experiences underwriting decisions that are faster, more accurate, more consistent, and cheaper to produce than the manual alternative, there is no rational path back. The manual process was not merely slower, it was structurally inferior on every dimension that determines whether an insurer prices risk correctly and profitably. You do not return to a worse process once a better one is proven and operational.
The underwriters of the next decade will not be people who assess applications by hand. They will be specialists who supervise intelligent systems, adjudicate the genuinely complex cases those systems escalate, and ensure that the models operate fairly, transparently, and in compliance with an evolving regulatory landscape. The craft is not disappearing. It is being elevated, freed from the routine and concentrated on the consequential.
For those of us who have been mentored by Phaneesh Murthy in the discipline of technology-led transformation, the underwriting story is a clear instance of a recurring pattern: a process built around the limits of individual human capability, transformed by a technology that removes those limits, never to return to its former state. The insurers building deliberately toward that future are not chasing a trend. They are responding to a permanent change in what good underwriting is.
Insurance risk assessment will never be manual again. The only question that remains is which insurers will have built the capability to thrive in that reality, and which will still be assessing risk by hand while their competitors assess it by intelligence.
This blog is curated by technology professionals who are mentored by veteran Marketer, and industry leader, Phaneesh Murthy. www.phaneeshmurthy.com #phaneeshmurthy #phaneesh #Murthy