Indian Banks Turn to Observability as AI Moves From Promise to Production
By Smruthi Nadig ·
As AI moves from experimentation into production, Indian financial institutions are finding that deploying an AI system is only the beginning. Banks and insurers increasingly need visibility into how applications, infrastructure, models and AI agents behave once they are operating at scale.
Philippe Deblois, Global Vice President, Solutions Engineering at Dynatrace, tells AIM the growing complexity of AI environments is making observability more important than traditional application monitoring. “With AI comes a whole lot more complexity,” he says, adding that the scope of systems involved can become “exponential” as organisations become more AI-native.
The challenge becomes particularly pronounced when autonomous agents interact with one another and with enterprise applications. Traditional monitoring may show whether an application is functioning, but it may not explain what an AI agent is doing, why it took a particular action or how that action affected a wider transaction.
For financial institutions, where a failed transaction, inaccurate recommendation or service outage can directly affect customers and revenue, that visibility is becoming an operational requirement.
Deblois says observability in India has moved beyond being a “nice to have”. It is increasingly becoming a C-suite initiative because organisations cannot manage systems effectively without the underlying visibility needed to support trustworthy automation.
The financial sector is already among the leading adopters. Deblois says seven of India’s largest banks and insurance companies are Dynatrace customers. He cites Shriram Capital as an example, where Dynatrace’s unified observability platform is used across their hybrid cloud environment for automated root-cause analysis, faster issue resolution and improved productivity.
AI Brings a New Layer of Operational Risk
For enterprises, monitoring AI requires more than measuring uptime and latency. AI systems are probabilistic, meaning their outputs can vary and may be difficult to predict or explain.
That creates a different operational challenge for banks and insurers. An AI workload can be technically available while still producing inaccurate outputs, consuming excessive resources or delivering an experience that fails to meet business expectations.
Deblois says enterprises therefore need real-time visibility into agent behaviour while also measuring the business impact of their AI investments. Observability can help organisations determine whether AI workloads are actually generating value or simply replacing existing processes with AI without producing meaningful returns.
The metrics extend beyond traditional IT performance. Deblois points to revenue growth, customer experience, productivity, cost optimisation and model accuracy as indicators that need to be connected to observability.
Amol P Bisarya, Regional Sales Manager at IT software company Motadata, described the shift in enterprise technology in a post on LinkedIn: “Observability ‘today’ is what ‘Cloud’ was 10 years ago.” He argues that observability is moving from a boardroom buzzword to a business necessity as enterprises seek greater visibility, intelligence and speed across their IT environments.
Cost is another emerging concern. AI workloads can consume significant computing resources and generate costs through model and token usage. As enterprises move from pilots to large-scale deployment, understanding latency, token consumption and operational expenditure becomes increasingly important.
RBI’s Resilience Push Adds Regulatory Pressure
The regulatory environment is making this visibility more consequential. RBI’s Master Direction on Information Technology Governance, Risk, Controls and Assurance Practices, issued in 2023, strengthened expectations around IT governance, risk management, controls, business continuity and disaster recovery for regulated entities. The framework also places greater emphasis on monitoring, assurance and technology risk management.
The RBI’s IT Outsourcing Directions, meanwhile, require regulated entities to maintain oversight of outsourced IT services, including risk assessment, monitoring of service providers, business continuity and exit arrangements. This is increasingly relevant as banks depend on cloud platforms, fintech partners and other third-party technology providers.
The regulatory framework has also evolved for NBFCs. In November 2025, the RBI issued its Managing Risks in Outsourcing Directions, consolidating outsourcing requirements, including for IT services. For covered NBFCs, the framework emphasises ongoing vendor oversight, performance, availability, incident response and resilience rather than treating due diligence as a one-time exercise, according to Vinod Kothari Consultants.
Deblois says AI deployments also require governance around privacy and security. Organisations need to ensure that personal information is not leaked and that security policies are enforced as AI workloads move into production.
Bringing AI, Infrastructure and Business Data Together
Dynatrace’s approach is built around Grail, its AI data lakehouse, which brings metrics, traces, logs, events, security events and AI telemetry into a single storage architecture. Deblois says this allows organisations to add context to telemetry and connect technical signals with business insights.
This becomes particularly important as banks modernise technology stacks comprising legacy systems, private infrastructure, public cloud and newer AI workloads. Treating AI as a separate experimental environment can create blind spots once those workloads enter production.
Deblois says organisations need to move beyond controlled AI environments and develop an end-to-end view across hybrid and multi-cloud infrastructure.
Security is also moving closer to observability. While the two functions remain largely separate within enterprises today, Deblois expects them to converge because both are concerned with ensuring that production systems are secure, governed and trustworthy. Combining runtime vulnerability information with observability data, he says, can help organisations understand what is running in production and assess the potential “blast radius” of a risk.
Dynatrace Expands AI Observability With Arize
Dynatrace is also strengthening its position in AI observability through its proposed acquisition of Arize AI. Announced on August 17, the transaction values Arize at $915 million in cash and stock, including approximately $815 million in cash and replacement equity awards. The deal is expected to close later this quarter or early in Dynatrace’s third quarter, subject to regulatory review and customary conditions.
Arize provides AI observability and LLM evaluation capabilities designed to help teams detect issues, troubleshoot them and measure AI output quality across traditional machine learning and generative AI applications. The combined platform is intended to cover the AI lifecycle from experimentation and deployment readiness through runtime evaluation and observability.
Dynatrace says AI observability is one of the fastest-growing categories in the market and projects it will exceed $10 billion by 2030. CEO Rick McConnell says the acquisition advances the company’s AI observability leadership, accelerates its roadmap and expands its reach among developers.
The company is also building its telemetry foundation. Deblois says Dynatrace recently acquired Bindplane, a telemetry pipeline provider, to help customers ingest data in a cost-effective and vendor-neutral way.
For Indian financial institutions scaling AI, the direction is clear: simply knowing whether systems are online will no longer be enough. Observability increasingly needs to explain what AI is doing, how it affects the wider technology environment, what it costs and whether it is producing the outcomes the business expects.
Deblois sees the next generation moving away from teams constantly watching dashboards towards “headless observability”, where insights are delivered directly into developers' and operators' workflows through IDEs, ServiceNow, MCP and command-line interfaces.
For a financial sector built around always-on digital services and increasingly AI-driven operations, observability is consequently becoming not just a monitoring function, but a foundation for resilience, governance and trusted automation.