Generative AI and the Future of Underwriting

The insurance sector in New Zealand is on the cusp of a technological transformation that goes beyond automation and into the realm of creative intelligence. As traditional risk models evolve, one of the most disruptive forces shaping underwriting today is Generative Artificial Intelligence (Generative AI). No longer limited to structured data analysis, this subset of AI is redefining how insurers create, evaluate, and personalise insurance policies—ushering in a new age of hyper-specific risk profiling and policy drafting.

While conventional underwriting has always aimed to balance accuracy with efficiency, generative AI introduces a fundamentally new approach: it creates. Trained on massive data corpora—claims data, policy documents, economic patterns, behavioural metrics—generative AI tools can now simulate scenarios, compose coherent documents, predict risks, and even converse intelligently with customers. For underwriting, this means greater speed, nuance, and adaptability.

Historically, policy drafting was a rule-bound process. Standardised templates were used across customer segments, with only minor alterations based on age, income, or property value. While this allowed for consistency, it often lacked the flexibility to account for nuanced risk factors and emerging needs.

Generative AI changes that. By analysing an applicant’s unique data—such as lifestyle habits, travel patterns, business exposure, and even geolocation—it can generate custom policy language in real time. For example, a generative model might craft a property insurance clause specifically tailored to a homeowner in a flood-prone coastal suburb, incorporating satellite imagery, rainfall forecasts, and historical claims data.

For small businesses, this is particularly transformative. Rather than purchasing generic commercial insurance bundles, they can now receive dynamically generated policies that reflect their size, industry, supply chain complexity, and customer footprint—driving fairer pricing and more comprehensive coverage.

Traditional underwriting models tend to cluster risk using broad demographic indicators—age brackets, occupation categories, or zip codes. However, these generalisations often fail to capture the complexities of individual behaviour or real-world exposure.

Generative AI excels in contextual risk profiling. By integrating disparate data streams—such as smart home sensor data, vehicle telematics, wearable healthy metrics, and even environmental trends—it builds a more comprehensive and dynamic profile of the applicant.

This holistic understanding of risk means policies can now adapt in real time. For instance, if a driver consistently maintains safe driving habits, their generative AI-supported policy could automatically lower premiums at renewal. Conversely, increased exposure to health risks or erratic driving could trigger a risk reassessment—done not through manual review, but by the model itself.

Such a system encourages safer behaviour, enables real-time underwriting, and allows insurers to be more proactive in their risk management strategies.

One of the lesser known but highly impactful features of generative AI is its capacity to simulate scenarios. For underwriters, this offers powerful tools for stress-testing policies and predicting future risk.

Imagine being able to ask, “What would happen to this customer’s risk profile if a severe drought hit their region?”, or “How would an economic downturn affect the claim probability of small business owners in hospitality?” Generative AI models can generate hypothetical data based on existing patterns and historical correlations, offering actionable insights.

This predictive ability helps insurers refine their portfolio strategies, set aside appropriate reserves, and remain agile in responding to emerging threats—whether environmental, social, or economic.

The underwriting process, even in its semi-digitised form, has long been a bottleneck in the customer acquisition journey. It involves data gathering, verification, analysis, policy generation, review, and approval—all of which take time and coordination.

With generative AI, the turnaround time is dramatically reduced. The system can collect applicant data from integrated sources (e.g., digital forms, financial records, and connected devices), evaluate the risk profile using predictive models, and generate a draft policy—often within minutes.

For customers, this means faster access to coverage. For insurers, it means reduced administrative costs, fewer manual errors, and more conversions at the point of sale.

A common concern with advanced AI technologies is the potential displacement of human roles. In the context of generative AI and underwriting, however, the trend is toward augmentation, not automation.

While AI models can draft policy language, score risks, and simulate events, it’s still the human underwriter who provides contextual oversight. Especially in high-stakes or ambiguous cases—such as insuring unique artworks, cross-border logistics businesses, or high-net-worth individuals—human discretion remains essential.

Moreover, the role of the underwriter is evolving from one of manual processor to strategic analyst. With AI taking over the grunt work, professionals are free to focus on exception management, regulatory compliance, product innovation, and client advisory.

Generative AI, while powerful, is not without risks. The black-box nature of many AI models can make it difficult to trace how a particular policy clause was drafted or why a certain risk score was assigned.

To address this, insurers must invest in explainable AI frameworks—models that provide clear reasoning for their outputs. This is crucial for maintaining customer trust and ensuring compliance with local regulations around fairness and transparency in financial services.

Bias in training data is another concern. If generative models are trained on biased or incomplete datasets, they may unknowingly reinforce discriminatory outcomes. For instance, if historical underwriting data reflects systemic disparities, AI-generated policies could unfairly penalise certain groups.

Ensuring ethical data governance, ongoing audits, and human oversight are therefore essential as generative AI becomes more deeply embedded in underwriting workflows.

Interestingly, generative AI isn’t just useful in customer-facing applications. It’s also being used to draft compliance documents, adapt policy language to meet local regulatory nuances, and flag outdated clauses in legacy contracts.

By cross-referencing evolving regulatory frameworks with internal documentation, generative models can suggest updates or insert disclosures that keep the insurer aligned with the law—saving compliance teams hours of manual work.

This not only reduces legal exposure but also ensures that customers receive policies that are up-to-date, relevant, and easy to understand.