By Gaurav Banka, Chief Risk Officer, Aviva India Life Insurance
Trust has always been the foundation of insurance. Customers share some of their most personal information and commit to long-term financial decisions with the expectation that insurers will act fairly when it matters most. As artificial intelligence reshapes how risks are assessed and decisions are made, that expectation remains unchanged. What is changing, however, is the need for stronger oversight of the systems driving those decisions. In the age of AI, risk governance is no longer just about managing downside risk; it is about ensuring that innovation remains worthy of customer trust.
Consider a situation where an AI model identifies two customers with similar risk profiles but recommends different outcomes based on patterns that are not immediately evident to business users. The technology may have worked exactly as designed, but if the rationale behind the decision cannot be understood, challenged or explained in simple terms to a customer, the risk extends beyond the model itself. It becomes a question of fairness, accountability and ultimately trust. This is why responsible AI cannot be viewed as a technology initiative alone. It is an enterprise-wide governance responsibility that requires business, risk, compliance, legal, data and technology teams to engage from the outset. Risk must enter the conversation when an AI use case is being designed, not when it is presented for approval.
Governance added at the end acts as a checkpoint; governance embedded from the start becomes an enabler.
Explainability is central to this approach. When an AI-supported decision influences a customer’s eligibility, premium, claim settlement or service experience, insurers should be able to clearly articulate how that outcome was reached. Customers may not need to understand the underlying algorithm, but they do expect decisions that affect them to be transparent and defensible. Complexity within a model cannot become an excuse for opacity outside it. In insurance, a “black box” is not merely a technical concern; it is a trust risk. While AI can identify patterns at a scale and speed beyond human capability, accountability cannot be delegated to an algorithm. The responsibility for every decision continues to rest with the institution that deploys it.
Human oversight becomes particularly important where AI-driven decisions have a direct impact on customers. The objective is not to place a person mechanically in every process to ensure that judgement, empathy and responsibility remain present where they matter the most. However, as AI tends to learn from historical data, which carries the risk of bias, objectivity must be monitored continuously through a human-in-the-loop approach. Without the right controls, technology may reproduce or amplify that bias with greater consistency and reach.
Looking Beyond the Algorithm
Effective AI governance requires insurers to focus not just on how a model performs, but on the outcomes it creates. As AI enables more granular segmentation and personalised offerings, organisations must ensure that personalisation does not inadvertently become exclusion. For instance, a model may identify customer characteristics that correlate with higher risk and recommend differentiated treatment. While technically sound, the more important question is whether that distinction is appropriate, defensible and aligned with customer interests.
The same principle applies to data and ecosystem governance. As insurers increasingly collaborate with cloud providers, analytics partners and fintechs, accountability cannot become fragmented. A customer interacts with one insurer, not a network of vendors, and the ultimate accountability must always reside with the insurer.
Building Resilience into AI
Risk governance must also account for resilience. AI models can drift, data pipelines can fail, third-party services can be disrupted and cyber threats continue to evolve. A claims model, for example, may perform effectively under normal conditions but produce unexpected outcomes when faced with changing customer behaviour or unfamiliar claim patterns. For insurers, the challenge is not only to test whether systems perform under normal conditions, but also how they respond under stress. Often, the greatest risk is not the one that has been identified, but the dependency or scenario that has yet to be be considered.
Yet resilience is not built through frameworks alone. It depends equally on a culture where employees feel empowered to question assumptions, raise concerns and view risk as part of everyday decision-making. The role of the risk function is not to slow innovation, but to help the business make better-informed choices.
As AI continues to reshape insurance, the organisations that earn lasting trust will not necessarily be those that adopt it the fastest, but those that adopt it responsibly with trust and explainability as the foundation. Effective risk governance provides the confidence to innovate at scale while ensuring that transparency, fairness and accountability remain at the centre of every decision. In an industry built on promises, that trust remains the most valuable asset of all.