For decades, financial close has run on a familiar rhythm: a frantic scramble at month-end, followed by weeks of reconciliation, review, and reporting. That calendar-driven model is now being reshaped by agentic AI, which is turning close from a periodic fire drill into a continuously managed process. Instead of discovering discrepancies after the fact, finance teams can now monitor transactions, flag anomalies, and route exceptions in real time, all while retaining human oversight for material decisions. As regulatory scrutiny intensifies and CFOs face pressure to deliver faster insight without adding headcount, the shift raises important questions around governance, data quality, and organisational trust.
In this conversation with ET Edge CIO&Leader, Madhukar Uniyal, senior director, solutions engineering, Applications, Oracle India, unpacks how embedded AI agents are redefining the finance function and what leaders must do to prepare for it.
Senior Director, Solutions Engineering, Applications
Oracle India
CIO&Leader: Financial close has traditionally been viewed as a back-office, calendar-driven exercise. How is agentic AI fundamentally changing that perception, and what’s pushing finance leaders to rethink the process now rather than five years ago?
Madhukar Uniyal: We’re seeing the financial close shift from a monthly fire drill to a continuously managed process. This comes from our practical experience working with the CFO and the Office of Finance.
The close is shifting from a periodic, calendar-driven event to a continuous, exception-driven process. Especially with agentic applications and AI agents on the way, they can monitor transactions, reconcile accounts, identify anomalies, and coordinate follow-ups throughout the month. So, if you ask me, finance is no longer waiting until period-end to discover issues.
Today’s urgency comes from greater reporting complexity, tighter security, and pressure to deliver faster insight without adding headcount. From Oracle’s standpoint, we offer Cloud EPM, which supports continuous, agent-assisted close, near-real-time anomaly detection, and ultimately reduces period-end disruptions.
To add a fact: Oracle is one of the fastest in the S&P 500 to close the books.
This is happening because we now have the technology to process it faster. It’s shifting from a traditional, back-office, calendar-driven exercise to a more agentic, AI-led process, pushing finance leaders to rethink how they work.
CIO&Leader: As autonomous agents take over reconciliations and routine transactional work, how does the day-to-day role of a finance professional change? What new skills or mindsets will finance teams need to develop?
Madhukar Uniyal: When we talk about autonomous agents, I want to highlight that agents and this AI aren’t removing financial judgment. It’s creating more capacity for it.
From a finance professional standpoint, they spend less time, you know, doing the rudimentary work that we always used to hear: matching transactions, chasing approvals, preparing routine reconciliations, and so on. The role shifts toward reviewing exceptions, applying judgment, doing more business, interpreting business drivers, and advising the organisation. So, we are seeing a fundamental shift in what finance professionals do.
The most important new skills will be data literacy, critical thinking, and AI supervision; for example, it’s an important skill we’re seeing develop. And the confidence to challenge an agent’s recommendation, asking questions so it auto-tunes itself.
From Oracle’s perspective, our stated direction is to automate high-volume transaction work.
The finance team can focus more on, you know, things like analysis, judgment, and, importantly, improving the business partnership.
CIO&Leader: With regulatory scrutiny intensifying globally and in India specifically, how does embedding AI agents into the close process strengthen rather than complicate governance and audit trails?
Madhukar Uniyal: Regulatory scrutiny in India is increasing, and laws are taking effect as guidelines reach the finance office.
The key is to embed agents inside governed finance workflows. Rather than operating as uncontrolled external tools, agents in our application and the Office of the CFO we work with operate within existing roles, permissions, approval hierarchies, policies, workflows, and audit trails. This lets the organisation record what an agent detected, recommended, acted on, and who approved it.
The full trail shows what the agent was intended to do within the guardrails, whether it did so within those boundaries, whether it had the required permission, and whether humans were in the loop.
In India, regulators emphasise stronger governance, fraud detection, responsible AI, transparency, and accountability. We also saw the RBI issue stated guidelines.
A well-governed agent can make the controlled environment more visible, not less.
CIO&Leader: Can you walk us through a real-world scenario of how an embedded AI agent identifies an anomaly during close, and what happens next? Is it fully autonomous, or is there a human-in-the-loop checkpoint?
Madhukar Uniyal: Imagine an account that normally receives invoices between 5 lakh and 20 lakh. If a 2- or 3-crore invoice appears near quarter-end from a new supplier, the agent can flag it as unusual by comparing it with past purchase orders, receipts, historical patterns, and policy thresholds.
It can review tax treatment, approval rules, and other tasks that would be a task in itself for a human. It can then flag the transaction, explain why it is unusual, gather supporting evidence, and route it to the appropriate controller.
With a human-in-the-loop supervisor, it can do all of this faster, while maintaining the same audit rate.
The autonomy level is configurable. Most organisations should start with human supervision, with a human in the middle for material or high-risk items, while allowing agents to resolve routine, low exceptions automatically.
We describe a progression from human-in-the-loop to human-in-the-lead, and eventually greater autonomy for routine execution.
That was one example, but there could be many more where agents manage anomaly detection, corrections, and rules effectively.
CIO&Leader: Faster close cycles are often assumed to come at the cost of accuracy. How does Oracle’s agentic AI approach address this perceived trade-off?
Madhukar Uniyal: Faster close doesn’t mean lower accuracy. A few years ago, I might have said more manual work in a shorter window could compromise accuracy. But not today, as agentic AI changes the equation.
They reconcile continuously, apply consistent rules, and escalate exceptions early.
Speed comes from removing manual work and bottlenecks, not from reduced control.
Oracle says embedded AI supports automated reconciliation journals and anomaly detection, enabling a faster, more accurate close.
The goal isn’t to close faster by checking less. It is to close faster because checking happens continuously.
CIO&Leader: How are Oracle Fusion Cloud Applications architected to embed these AI agents natively, as opposed to bolting AI on top of existing ERP workflows? What does that mean for customers on legacy systems considering modernisation?
Madhukar Uniyal: We have a clear message about Fusion applications: built-in, not bolt-on. Oracle decided not to give AI a separate name outside our core application, so the AI sits right within the workflow.
The Fusion advantage is that AI is embedded within the application and transactional workflow. The agents have governed access to enterprise data, workflows, policies, approvals, permissions, and audit mechanisms users already rely on in our Fusion applications suite.
For legacy customers, modernisation is more than a technical upgrade. It is a move towards a unified data model and process automation, where AI can act with full context.
If you look at the Oracle architecture and messaging, we are positioned because we have the fundamental layer: Oracle Cloud Infrastructure, on which our Fusion application runs. Between OCI and the Fusion application are agentic workflows, AI Studio, and agentic applications.
It’s like an orchestra playing a symphony. Unlike hiring a drummer from somewhere else, we are hiring a guitarist from elsewhere and teaching them to play together.
CIO&Leader: Agentic AI is only as good as the data it acts on. What data governance or data quality prerequisites should CFOs put in place before deploying autonomous agents in their close process?
That’s true. I agree that it is as good as the data. I often raise this in discussions among Oracle experts, CFOs, and financial controllers.
We allude to five fundamentals: Number one, consistent charts of accounts and master data. Number two, clear ownership of critical data. Number three, standardised closed policies. Four, strong access control. And number five, active monitoring of data quality.
These are the five fundamentals that we would advise CFOs to focus on. An agent should have a defined threshold, an approved source system, and very clear rules for when human review is mandatory.
Introducing AI will solve the data problem and the fragmentation. No, not at all.
It will not solve fragmented definitions or weak process discipline. A unified, trusted data foundation gives the agent the context required to produce a reliable outcome.
CIO&Leader: What has Oracle observed as the biggest barrier to adoption among Indian enterprises: technology readiness, organisational trust in autonomous decision-making, or something else?
Madhukar Uniyal: This answer reflects my experience with conglomerates and financial enterprises in India, one of our core industries.
Technology is often not the hardest part; it usually comes later. The biggest barrier is organisational trust.
Financial leaders want clarity on accountability, explainability, data security, and what happens when the agent is wrong. There is also a process issue: automating an inconsistent or poorly controlled close only scales the inconsistency.
The most successful approach is phased adoption: start with recommendation and low-risk automation. That is the biggest adoption barrier we have seen.
Start with recommendation and low-risk automation. Then measure performance, preserve human checkpoints, and expand autonomy as confidence grows.
RBI’s recent work on responsible AI stresses transparency, accountability, governance, and human oversight. The adoption challenge is less about whether the AI works and more about whether the organisation trusts the operating model around it.
CIO&Leader: How should a CFO measure ROI on agentic AI investment in the close process? Is it purely time-to-close, or are there other KPIs (error rates, headcount reallocation, audit costs) that matter more?
Since we are talking about the CFO, this question is about how CFOs measure ROI. No one else in the organisation would look at ROI as granularly as the Office of the CFO and the CFO themselves.
Time to close matters, 100%, but it is only one dimension. CFOs also track reconciliation automation rates, manual journal volumes, exception resolution time, late adjustments, and error and rework rates.
Auditing effort matters. Control failures are another important ROI measure.
The percentage of finance capacity redirected towards analysis and decision-support tasks is another ROI driver. The strongest ROI combines productivity, risk reduction, employee capacity, and faster insight. It is not simply about how soon I can close, or how many fewer people I need for the task. The real value is fewer days to close, fewer surprises, fewer corrections, and more time for insight and decisions.
CIO&Leader: Looking three to five years out, what does a fully autonomous finance function look like, and what should CFOs be doing today to prepare their organisations for that shift?
Madhukar Uniyal: Fully autonomous finance would look like nobody’s there. It would keep walking on its own, like a self-driving car. But that is not what a fully autonomous finance function will mean.
It will mean agents managing routine, repetitive tasks, doing reconciliation, monitoring, and orchestration in a continuous cycle. People will set policy, approve material judgments, investigate exceptions, and advise the business. CFOs should prepare now by standardising processes and moving towards a unified cloud platform, which is fundamental and ground-up thinking in Fusion. They should also strengthen data governance, define AI accountability, and redesign roles around judgment and outcomes.
The future of the finance team will manage outcomes and exceptions, not queues and transactions. One of Oracle’s recent messages on agentic apps is that applications are not just a system of record, but a system of outcome.