Sriram Naganathan, President & CTO, HDFC ERGO General Insurance, on how AI is moving beyond automation to reshape risk assessment, claims, customer experience and the way insurers build technology.
For much of the digital transformation era, insurance companies measured the success of technology through familiar metrics—uptime, speed, efficiency and the digitisation of processes. Technology was largely viewed as an enabler sitting alongside the core business.
That equation is changing.
As artificial intelligence, machine learning, generative AI and agentic systems move into underwriting, pricing, claims, fraud detection and customer service, technology is increasingly becoming part of the business model itself.
At HDFC ERGO General Insurance, Sriram Naganathan, President & CTO, describes this shift as technology becoming a “co-architect of the business.”
“For years, technology in insurance was measured by metrics such as uptime, speed and efficiency—all of which typically reflected its role as a support function. Today, however, technology is becoming a strategic function,” says Naganathan.
The reason is straightforward: insurance is fundamentally a data and risk business. The ability to understand that data increasingly determines how an insurer assesses risk, prices products, designs customer propositions and delivers service.
The transformation, therefore, is not simply about putting existing processes online. It is about using data and intelligence to change how those processes work.
The shift from digital to intelligent
HDFC ERGO’s technology journey illustrates that progression.
Sriram says the organisation has been using traditional AI and machine learning for several years across underwriting support, pricing, fraud detection and claims assessment. It is now extending those lessons to generative AI and agentic AI.
The scale of digital adoption provides the foundation for this next phase. According to Sriram, 95% of policies are issued digitally, while around 88% of service interactions happen digitally, with nearly 10% of those interactions being AI-led.
That marks a subtle but important change in strategy.
The first phase was about making the insurance journey digital. The emerging phase is about making that digital journey intelligent.
One example is HDFC ERGO’s Here app ecosystem, which has recorded more than 1.2 crore downloads, including majority downloads by users who are not customers. An example is its Know Your Policy capability, which uses GenAI to answer specific queries in simplified language to queries from customers on their policies. Customers can ask questions about their policies in natural language and receive explanations of what is covered, what is payable and what is not.
Policy documents can be technically precise but difficult for customers to interpret. GenAI creates an opportunity to make the information layer of insurance more accessible without necessarily changing the underlying policy.
AI should assist decisions, not eliminate people
As AI moves into higher-stakes decisions, the question becomes less about what the technology can do and more about where it should be allowed to act independently.
Sriram’s approach is clear: AI should help humans make decisions.
The technology can analyse underlying data, extract knowledge and make intelligence available to servicing, claims and sales teams. Human professionals can then use that intelligence to improve the decisions they make.
That distinction becomes particularly important in claims.
For low-value and low-complexity claims, HDFC ERGO is using AI and automation to handle more of the process in automated fashion. For higher-value claims, where emotional and contextual considerations become more significant, AI operates as a co-pilot for experienced claims professionals rather than replacing them.
This creates a model of graduated automation: automate what is predictable, augment what requires expertise, and retain human intervention where context and sensitivity matter.
Claims as the ultimate AI test
Claims represent perhaps the clearest test of whether AI is actually improving insurance.
The objective is not merely to settle a claim faster. It is to make the process transparent, predictable and frictionless for genuine customers while identifying fraudulent cases more precisely.
Sriram points out that the presence of a small number of fraudulent claims should not make the experience more difficult for the larger population of genuine customers. HDFC ERGO therefore uses data and AI to identify suspicious cases and move them into a separate investigation route, while allowing legitimate claims to progress with minimal friction.
The approach is already visible in motor insurance.
For more than six years, the company has used traditional AI models to analyse images and videos for motor claims. A customer can receive a link through WhatsApp that activates an AI-enabled camera experience on a smartphone. Guided photographs are captured and submitted, after which an AI model assesses the damage and supports the surveyor in responding more quickly.
The company is now exploring GenAI-powered assessment as well.
In health insurance, the focus is similarly on using AI agents for repetitive, lower-level activities while retaining human involvement for sensitive interactions and complex claims.
The direction is clear: AI is being positioned not simply as a claims-processing tool, but as an orchestration layer around the claims experience.
The data problem has changed
Interestingly, the biggest data challenge is no longer necessarily the absence of data.
HDFC ERGO has accumulated two decades of information spanning underwriting, claims and customer experience. Sriram says the bigger challenge today is turning that information into meaningful and actionable intelligence.
The organisation has therefore moved towards a modern data platform designed to connect signals from its underlying data and use them across underwriting, servicing, claims and distribution.
This has also enabled more granular behavioural cohorts.
The result is an operational scale that would be difficult to manage manually. Sriram reveals that HDFC ERGO facilitates close to 30 customer interactions per minute and issues around 82 policies every minute.
But as AI becomes more influential in business decisions, another issue rises in importance: trust.
“Ultimately, the insurance business rests on one very important word: trust,” Sriram says, highlighting explainability and governance as critical components of AI adoption.
For an insurer, this is not a theoretical concern. When an algorithm influences pricing, underwriting, fraud detection or claims, the organisation needs to understand why a particular outcome was produced and have appropriate controls around it.
Modernisation cannot become an IT exercise
AI adoption also forces insurers to rethink how they approach technology debt.
For Sriram, modernisation should not be pursued simply for the sake of upgrading technology. Every technology investment needs a strong business rationale.
“We don’t treat innovation and modernisation as competing priorities,” he says. The underlying platforms should enable the organisation to innovate faster and allow technology to remain a sustainable partner to the business.
This philosophy has also influenced HDFC ERGO’s approach to software.
Over the past year and a half, the organisation has deliberately shifted from primarily buying software and SaaS packages towards building more of its own capabilities. Sriram says the company’s accumulated data and insurance-specific insights represent a form of institutional knowledge that can create greater value when embedded into internally developed assets.
The company is building engineering capabilities and developing its own AI models while using underlying technologies from large technology providers.
That does not mean everything needs to be built internally.
Rather, the distinction is between commoditised technology and capabilities that represent genuine business differentiation.
The rise of smaller, specialised models
The same thinking is beginning to influence the AI stack itself.
Sriram expects insurers and other enterprises to become more selective about where large language models are used, with smaller language models potentially providing greater efficiency for specific enterprise workloads.
This could represent an important evolution in enterprise AI.
The early phase of GenAI adoption was dominated by experimentation with large, general-purpose models. The next phase is likely to involve a more nuanced architecture—using the appropriate model for the appropriate task, with considerations around cost, efficiency, security, latency and data sensitivity.
For insurance companies handling highly sensitive customer and risk information, that distinction could be particularly relevant.
When AI agents start making decisions
The move from AI-assisted workflows to agentic AI introduces another layer of complexity.
At HDFC ERGO, AI agents are already being piloted for low-value retail health and motor claims, including document verification, image-based assessment and routine adjudication. Sriram says that the early results have been encouraging, with potential benefits including faster settlements for genuine customers, lower friction and continued fraud controls.
But the biggest risk, in his view, is not necessarily the technology itself.
It is governance.
As agents become capable of executing multi-step processes and taking decisions with greater autonomy, insurers need clearly defined boundaries around what an agent can decide and when it must escalate to a human. Data quality, accountability and explainability become equally important.
This is where AI governance starts looking less like a separate compliance function and more like an operating discipline.
From digitising processes to anticipating customers
The ultimate distinction between a digital insurer and an intelligent insurer, according to Sriram, lies in what happens after processes have been digitised.
“Digitisation puts your existing processes online and enables self-service. AI, by contrast, can help transform the entire organisation into an intelligent enterprise.”
The difference could be visible in something as ordinary as a renewal reminder.
Instead of repeatedly sending the same generic notification, an intelligent system could understand a customer’s preferred channel, behaviour and the time at which they typically complete a renewal—and use that context to determine when and how to engage them.
This represents a shift from schedule-based automation to context-based decisions.
It is also where the technology transformation becomes visible to the customer.
The future is not autonomous insurance—it is augmented insurance
Sriram’s vision ultimately points towards an insurance enterprise where AI operates throughout the business, but not necessarily without humans.
The future architecture could combine specialised models, AI agents, customer data platforms, cloud infrastructure and internally developed engineering capabilities, all governed by strong security and accountability frameworks.
Yet the human layer remains essential.
For claims involving accidents or hospitalisation, for example, technology can improve communication and speed, but empathy remains difficult to automate.
That may be the most important distinction as insurance enters its next technology cycle.
The objective is not to remove humans from the insurance process. It is to remove unnecessary friction from the process while giving human professionals better intelligence when their judgement matters most.
As Sriram puts it when asked to choose between AI and human intuition: “AI-enabled human intution.”
For an industry built on uncertainty, that could be the more meaningful promise of AI—not eliminating risk or replacing judgement, but giving insurers the intelligence to understand risk better, respond faster and make the insurance experience more contextual for the customer.