Banking and Financial Services (BFS) has moved from asking what Artificial Intelligence (AI) can do, to asking where it can create measurable enterprise value. Early Generative AI (Gen AI) adoption was largely exploratory, with financial institutions testing use cases while navigating model risk, data, governance and explainability concerns. The next phase is different: The priority is to move from isolated AI tools and pilots to Agentic AI workflows that can operate across systems, functions and decision points — with the Human-in-the-Loop (HITL) controls required for a regulated environment.

A recent WNS live poll of BFS leaders points to an uneven transition: Exactly one-third of respondents were experimenting through ad hoc trials, another third were running limited departmental pilots and a final third had achieved full automation in select processes. Two-thirds therefore remained in what the poll characterized as a transition zone.

The implication is not that financial institutions lack AI capability. It is that scaling AI requires a different operating model — one that treats workflow re-design, orchestration, governance and human accountability as part of the transformation rather than as layers added after the technology is selected.

The Barriers to Scale
Why Agentic AI Is Hard to Operationalize

Financial services workflows often reflect the constraints of legacy platforms, fragmented data and processes built around sequential human handoffs. Following decades of mergers and acquisitions, many institutions now operate across complex technology estates, making it difficult to move information cleanly from one stage of a journey to the next. Where a workflow spans multiple systems, simply inserting an AI agent into one step can improve local efficiency without removing the underlying friction.

Regulation raises the bar further. Financial institutions need governance that can establish accountability, monitor AI use and provide appropriate transparency and explainability for material decisions. The Financial Stability Board's 2026 guidance¹ emphasizes organization-wide AI governance, lifecycle risk management and explainability, while the Bank for International Standards (BIS)² has stressed that human responsibility and accountability must remain central to AI-enabled financial decision-making.

This makes agent sprawl more than a technology-management issue. When business units adopt AI agents independently, organizations can lose visibility into what those agents are doing, what data they can access and how decisions move across an end-to-end process.

The strategic response is not to eliminate experimentation, but to create a governed path from experimentation to production.

Where to Start
Re-design the Workflow, Not Just the Task

A useful starting point is the workflow rather than the individual task. The strongest candidates for Agentic transformation are processes that are long-running, document-intensive and dependent on multiple systems or teams. In financial services, these often occur at the handoff between client-facing teams and middle- or back-office operations.

← Swipe →

Potential use cases for this include:

← Swipe →

The strategic shift is to ask what the ideal agent-supported workflow should look like if the institution were designing it today — not how to automate each step of a process inherited from the past. That distinction separates task automation from operating-model transformation.

Re-defining ROI in the Agentic Era

Traditional automation programs often emphasized measures such as Full-Time Equivalent (FTE) reduction, cost per transaction and processing speed. Those metrics remain relevant but do not capture the full value of an Agentic operating model. A broader ROI framework should connect operational performance to enterprise outcomes.

← Swipe →

Competitive advantage does not rely simply on how much labor and costs an institution can save. It lies in what additional value can be created when scarce expert capacity is re-directed from administrative coordination toward judgment, relationships and higher-value work.

A Case Study
Automating the Investment Lifecycle

A global investment management firm faced operational friction owing to fragmented, manual workflows across spreadsheets, legacy systems and e-mail. Core activities such as investor onboarding, record maintenance, approvals and reporting were time-consuming and lacked end-to-end visibility.

WNS-Vuram implemented an Appian-powered automation solution across more than 15 applications, streamlining workflows, approvals and investor record management. The solution enabled:

← Swipe →

This real-world example illustrates an important distinction:
The value came from connecting multiple activities into a coherent workflow, rather than automating a single task in isolation.

Keeping Humans in the Loop
Match Autonomy to Risk

Agentic automation should not be treated as an all-or-nothing decision. A more practical model is adjustable autonomy: Agents can begin by gathering information, preparing recommendations or executing low-risk actions while human professionals validate material decisions. As evidence accumulates, organizations can expand autonomy where controls and performance support it.

For high-impact decisions such as commercial lending, agents may be well suited to gather, synthesize and extract information from large documentation sets. Final credit decisions, however, should remain subject to the institution's risk framework, applicable controls and accountable human judgment. This is consistent with emerging guidance that emphasizes human responsibility, explainability and appropriate oversight in financial AI.

← Swipe →

Technology fit matters too. Large Language Models (LLM) and Agentic systems are strong at language-heavy work such as document interpretation, information retrieval and conversational interaction. Deterministic calculations, credit-scoring logic and rule-based compliance controls may require other technologies. The target architecture is therefore not an LLM-only stack, but a hybrid model in which Agentic intelligence orchestrates specialized tools.

← Swipe →

The Question Every BFS Leader Needs to Ask

The next phase of AI adoption in financial services will be defined less by the number of pilots launched than by the number of workflows fundamentally re-designed. The strategic question is no longer simply which AI tool to deploy. It is where Agentic intelligence can safely remove friction across the enterprise — and where human expertise must remain decisive.

Organizations that can answer that question with a clear operating model, measurable value framework and auditable governance structure will be better positioned to move from AI experimentation to sustained enterprise value.

About the Author

Leigh Varnham

Chief Growth Officer – UK and EU, WNS-Vuram

Leigh Varnham leads WNS-Vuram’s end-to-end growth strategy across the UK and EU, overseeing business acquisition, account expansion and strategic client relationships. He drives sales leadership, pipeline development and go-to-market execution, working closely with clients and partners to scale hyperautomation, low-code and digital transformation solutions for measurable business outcomes.

FAQs

References

1. https://www.fsb.org/2026/06/sound-practices-for-responsible-adoption-of-artificial-intelligence-ai-consultation-report/

2. https://www.bis.org/speeches/20260416-ai-finance-what-can-change-what-must-never-change