The CFO in the Age of AI: What India's Finance Leaders Actually Think

We brought together finance leaders from some of India's fastest-growing companies for a candid, no-deck conversation on AI, real-time decision-making, and cross-border payments. Here's what came out of the room:

• Transparency in cross-border payments is still an unsolved problem at the banking layer. Companies moving millions of dollars a month are often unaware of the FX spread they're being charged.
• AI is a signal generator, not a decision-maker. Every finance leader present drew a hard line between AI-generated insight and the act of signing off. Accountability still belongs to the human.
• Real-time data hasn't produced real-time decisions yet. CFOs are getting better information faster, but the organisational muscle to act on it at the same speed is still being built.
• There are workflows finance teams are ready to hand over and ones they're not. The line is clearer than most AI vendors would have you believe.
• The CFO's edge in an AI world is mastering the outliers. Geopolitical shocks, regulatory reversals, a Sunday-night tweet: these are where human judgment still has no substitute.

The event was a roundtable co-hosted by Xflow and ET CFO, held at The Oberoi, Bengaluru on 24th June, 2026. Around the table were CFOs and senior finance leaders from companies including RSA, Infoblox, Verint, Mathco, Encora, Zapcom Solutions, Prophecy, HiLabs, Infinite Locus, and Chainbrain: high-growth businesses, most operating across borders, all dealing with the same pressures: volatile FX, tightening compliance requirements, and the question of how much to actually trust the AI tools landing on their desks.

The conversation was structured around one question: is the modern CFO a real-time decision-maker? The answer, it turned out, was more complicated than the question.

The conversation's final stretch moved to cross-border treasury specifically and surfaced some of the most concrete pain points of the morning.

For companies operating across borders, cross-border payments sit at a frustrating intersection: high frequency, high value, and surprisingly low visibility. Most finance teams don’t exactly know their banking fees. Banks don’t give you a clear picture of what they're actually paying in FX conversion costs, because banks aren't required to show it and most don't.

The scale of this blind spot is striking. Companies moving hundreds of thousands of dollars a month internationally have, in documented cases, been operating under the assumption that their FX costs were negligible, only to discover when they actually looked that the spread being applied was costing them several thousand dollars on every transaction cycle.

Beyond cost visibility, there's the operational complexity of the correspondent banking network itself. When money moves across multiple banking relationships: an originating bank in India, an intermediary in Singapore, a receiving bank in the US, it can effectively disappear for three to five days with no reliable way to track it. For finance teams managing cash flow forecasts and working capital positions in real time, that opacity is a genuine operational problem, not just an inconvenience.

The compliance layer adds further weight. For Indian companies in particular, cross-border transactions carry significant compliance obligations including GST treatment, FIRC documentation, SoftEx filings, and EDPMS. As transaction volumes scale, managing these obligations becomes a full-time job that most finance teams aren't staffed to handle alongside their core work.

These aren't new problems, but they remain largely unsolved at the banking layer. The finance leaders in the room recognised them immediately, because most were living with them.

There's a version of the AI conversation in finance that goes: better data, faster insights, smarter decisions. What the leaders in the room described was something more granular and more honest.

The broad consensus was that AI has meaningfully improved the quality of information available to finance teams. Dashboards are richer, forecasting models are more sophisticated, and routine data work that used to consume analyst bandwidth is increasingly automated. That part of the story is real.

But the leap from better information to faster decisions is not automatic, and for high-stakes calls, most in the room aren't making it. The issue isn't distrust of the technology per se. It's that AI, at its current stage, operates probabilistically. It surfaces patterns from historical data and generates outputs based on those patterns. What it cannot do is account for the thing that wasn't in the training data: the geopolitical event, the regulatory reversal, the market move driven by a single individual's social media post.

Finance leaders are also acutely aware that accountability doesn't transfer when they hand a decision to a model. When a CFO signs off on a valuation, a treasury position, or a compliance filing, they own that call regardless of what the AI said. That reality shapes how the technology gets used in practice: as a briefing tool, not a decision-maker.

As Dhiraj Choraria, CFO at RSA, put it: "When the output comes in, who is reviewing it, who is authorising it, who is signing off? I don't think you can just go ahead and believe everything AI is giving today."

If there was one thread that ran through every part of the morning's conversation, it was this: the speed at which finance teams receive information has improved dramatically. The speed at which they act on it has not kept pace.

Abhilash Kumble, Manager - Finance at Chainbrain, set the tone early: "We are still evolving towards making real-time decisions. We have not reached there. Decision-making accuracy is much easier with this kind of data, but the speed of decision is still something we need to become better at."

Part of this is structural. Enterprise systems are complex, and even well-resourced finance teams operate within processes: approval chains, risk committees, policy constraints that were not designed for real-time response. Plugging in a better dashboard doesn't change the underlying decision architecture.

Part of it is also about context. AI models are trained on historical data, and history has limits as a guide to the present. A forecast model might correctly identify that holding USD exposure is the right call based on everything it knows, right up until a development occurs that it had no way to anticipate. Harsha Kaalla, Finance Controller at Prophecy, described exactly this scenario: a well-reasoned treasury position, built on solid data, that had to be unwound in a matter of hours when news of a US-Iran peace development broke overnight and shifted the rupee direction entirely. "There is definitely a gap. The human element has to make the decisions. You have to be very smart to do that."

This is the gap that finance leaders are actually navigating: not a gap in data quality, but a gap between what models can see and what the world can produce. Mr. Venkatraman, CFO at Mathco, framed it plainly: "We need to be realistic with our expectations. Sooner is not the answer for all your problems."

The practical upshot from the room was consistent: the value of AI in finance right now is not in replacing the decision, it's in compressing the time it takes to get to a decision. Better-prepared, better-informed, faster to the point where human judgment takes over: that's the realistic near-term use case.

Despite the healthy skepticism on high-stakes decisions, there was genuine clarity in the room about the workflows AI can and should own today.

Transactional and operational work came up repeatedly as the clear near-term opportunity. Expense claim processing, SOW review, accounting consolidation, cash application, receivables management: these are rule-based, data-heavy processes where AI can operate with minimal human intervention once the parameters are set correctly. Several leaders noted that their teams are already running versions of this, and the productivity impact is significant.

One leader described a sales performance bot built at a previous company that gave sales leaders real-time pipeline visibility without any involvement from the finance team: a small example, but illustrative of where the early wins are.

The clearer picture, though, was on the other side of the line: the workflows where human judgment remains non-negotiable. Compliance and legal decisions ranked highest. The challenge here isn't just accuracy; it's that the consequences of getting it wrong are asymmetric, and the regulatory environment changes in ways that models lag behind. Strategic decisions including pricing, M&A valuation, and capital allocation were equally off the table. These require an understanding of competitive context, stakeholder dynamics, and organisational intent that AI simply doesn't have access to.

Pankaj Sharma, Senior Director - Finance at Zapcom Solutions, captured the practical approach most teams are taking: start narrow, prove it, then scale. "We are picking up a use case, trying to do a POC. If that is successful, that is where we go and monetize." Enterprise-wide AI transformation is a vendor pitch. Focused deployment that solves a specific problem is what's actually working.

The most useful framing of the morning came near the end, and it cut across both threads of the conversation: AI and cross-border payments are, at their core, the same problem. Both involve navigating systems that give you better information than you had before, without ever giving you complete information. Both require finance leaders to know exactly where the model ends and their judgment begins.

On the cross-border side, that line is already visible. The correspondent banking network will tell you money is moving. It won't tell you where it is or what it actually cost you. Compliance frameworks will tell you what's required. They won't tell you how a bank in a different jurisdiction is going to interpret it on a given day. The finance leaders who manage cross-border treasury well aren't the ones who've automated everything. They're the ones who know which parts to trust, which parts to verify, and which parts require a phone call to someone who knows the answer.

The same logic applies to AI. As Ashutosh Gupta, Global Finance Controller at Encora, framed it: no model anticipates the outlier. Black swans, paradigm shifts, the development that by definition wasn't in the training data. History is full of them and cross-border finance, with its exposure to geopolitical risk, currency volatility, and regulatory change across multiple jurisdictions, sits closer to that edge than most domains in finance.

The CFO's value isn't in the work that systems do well. It's in the judgment that kicks in precisely when they don't. That's true of AI tools, and it's equally true of the banking infrastructure that moves money across borders. The leaders who navigate this transition well will be the ones who know, in real time, which one the moment calls for.