Autonomy is becoming one of financial services; most widely used, and inconsistently defined, terms. It gets conflated with automation, confused with AI, or presented as a future where banks and insurers run themselves.
Srinivasan Seshadri of HCLTech argues autonomy in financial services is not automation, and that trust must be engineered into every AI-driven action.
At its most honest, autonomy in financial services is a shift in the relationship between people and systems. From human-led operations supported by technology, to technology-led operations guided by human judgement. That changes what people inside these institutions spend their time on, and everything downstream, from cost structures and speed to market, to customer experience and competitive positioning.
The institutions getting this right aren’t chasing a destination. They’re progressing along a spectrum. And the ones making real headway share something in common: they start with the outcome and work backwards.
Start with the customer experience
Here’s a useful question for any institution evaluating its own progress: what does your customer actually experience?
A payment clears without intervention. A potential overdraft is identified before it occurs. A mortgage application progresses without repeated requests for updates.
The desired outcome is service that’s invisible when everything is fine, but becomes present and personal when it isn’t. AI can help deliver that personalisation at scale, not as a luxury for high-net-worth clients, but as a standard for every customer, without inflating cost-to-serve.
Most institutions have built the front door for this. What’s missing is the operation behind it that can actually deliver on what the interface promises.
Embed intelligence within clear guardrails
Anticipating customer needs requires more than just faster execution. Systems must be able to meaningfully surface the right offer at the right time, flag a potential issue before it escalates, or adjust a risk threshold in real time, within guardrails and with traceability.
Embedding intelligence across the value chain can support underwriting, surveillance, claims handling, advisory, and product recommendations, not just enabling, but constantly learning, adapting, and acting.
This is the difference between automation and autonomy. Automation does what you told it to. Autonomy does what the situation requires, within the boundaries you’ve set.
And for financial services institutions operating under regulatory scrutiny, those boundaries matter more than the intelligence itself.
Engineer trust into the system
For most financial institutions, the focus falls on capability rather than trust.
Trust must be supported by structural properties: decisions should be explainable, traceable, auditable and reversible. The data informing them must be secure, while outcomes must be monitored for accuracy, accountability and bias.
In financial services every autonomous action needs to be compliant, defensible and reversible. That’s not a constraint on autonomy. It’s the precondition for it.
Trust must therefore be treated as part of the architecture, not as an assurance exercise added later. Governance, security and explainability should be designed into systems and workflows from the outset. People must also retain the ability to intervene, override or redirect decisions when the stakes require it.
Connect information and orchestrate work
Intelligent systems depend on clean, connected, timely information. Yet data silos, legacy technology and fragmented integrations continue to limit what most institutions can achieve. When the architecture is composable and governed, new capabilities plug in without rework. Data moves where decisions need it. Partners connect cleanly. Regulatory requirements in new jurisdictions don’t require starting from scratch. Without this foundation, AI can act only on the partial information available to it, creating blind spots and risk.
The same principle applies to workflows. Many operations still rely on people to pass files, chase approvals, check exceptions and reconcile information across systems. Every handoff adds time, cost and potential error.
Self-steering workflows can handle routine volumes across onboarding, identity checks, reconciliations, disputes, payment exceptions and claims. People then intervene by exception rather than by default, focusing on complex risk decisions, sensitive customer conversations and strategic judgement.
Turning autonomy into measurable value
Commercially, autonomy should deliver two outcomes: lower structural costs and faster growth. Better orchestration can reduce manual work and prevent efficiency gains from disappearing between transformation programmes. Connected, governed infrastructure can also make it easier to test products, enter markets and scale successful ideas.
Progress should be incremental and measurable. Institutions should identify where autonomy already works, determine where the next step will create the greatest value, and measure improvements in cost, speed, risk and customer experience.
Autonomy will remain a buzzword unless it becomes defined by specific, measurable, operational terms. The institutions that get this right won’t be the ones that moved fastest. They’ll be the ones that moved with the most clarity, maintaining trust, compliance and control at every stage.
Facts Only
* Autonomy in financial services is a widely used but inconsistently defined term.
* Srinivasan Seshadri argues that autonomy in financial services is not automation.
* Autonomy is framed as a shift from human-led operations to technology-led operations guided by human judgment.
* Institutions progressing in this area start with the desired outcome and work backward.
* A useful question for evaluation is assessing what the customer actually experiences.
* Desired outcomes include invisible service when functioning normally, becoming present when issues arise.
* AI can deliver personalization at scale without inflating cost-to-serve.
* Systems must surface appropriate offers and flag issues within defined guardrails with traceability.
* Autonomy requires decisions to be explainable, traceable, auditable, and reversible to build trust.
* Intelligent systems require clean, connected, timely information; data silos limit achievement.
* Self-steering workflows can handle routine volumes, allowing people to focus on exceptions and complex decisions.
* Commercial autonomy should result in lower structural costs and faster growth.
Executive Summary
Full Take
The narrative positions autonomy not as a technological endpoint but as a change in the fundamental relationship between human agents and automated systems within regulated environments. The core tension lies between raw capability (automation) and responsible action (autonomy), suggesting that for financial services, defining agency is less about computational power and more about enforceable systemic trust. The progression model—starting with customer experience and working backward—is a pragmatic strategy that attempts to bridge abstract technological potential with concrete business goals. This implies that the greatest barrier to adoption is not the ability of AI systems to act, but the institutional willingness to delegate consequential judgment while maintaining accountability.
The emphasis on engineering trust into the architecture points toward a necessary shift in governance philosophy: accountability cannot be an afterthought bolted onto autonomous systems; it must be inherent to their design. The critique of data silos and fragmented integration highlights that achieving true autonomy is architecturally dependent, suggesting that technical solutions alone are insufficient without a composable, governed foundation. This resonates with broader systemic challenges where emergent capabilities risk outpacing regulatory and ethical oversight. The implication for human agency is that autonomy must enhance human focus by automating the routine, allowing human expertise to be deployed precisely where context, ethics, and judgment are most critical—the exceptions and strategic risks.
What frameworks exist for operationalizing "trust" as a measurable architectural property? If systems are designed to be explainable and reversible, this implies an acceptance of bounded error rather than perfect prediction. The challenge remains in translating the philosophical demand for trust into auditable regulatory compliance that satisfies both the institution's need for speed and the public's need for security. What specific metrics bridge the gap between operational efficiency gains and the psychological assurance required by regulators?
Sentinel — Human
The text functions as a well-structured argument defining 'autonomy' in finance through a series of linked principles concerning trust, architecture, and workflow orchestration, exhibiting high human analytical depth.