The problem with fragmented payments infrastructure is usually described as technical debt. For today’s chief financial officers, it is becoming one of financial duplication. Separate payment systems tend to accumulate separate controls, approval processes, data models and reconciliation workflows. That makes relatively basic financial questions surprisingly difficult to answer across an enterprise.
“Historically, a lot of organizations will have accumulated payment infrastructure one decision at a time,” Tungsten Automation Head of Payments and Embedded Finance Andrew Ng told PYMNTS for the September edition of the “What’s Next in Payments Series: The Fall Draft.”
And fragmentation means finance organizations can spend heavily on payment execution while still requiring employees to reconstruct the financial picture above it.
“Payment execution alone is increasingly commoditized. The durable value as I see it today is in making the right payment decisions,” Ng said.
He added that he continues to encounter large companies that see employees email or message colleagues on Teams for payment approvals.
“It is not through lack of desire,” Ng said. “It is a consequence of this buildup over time.”
The CFO opportunity, as a result, may not be adding an artificial intelligence budget on top of the existing payments stack. It may be using AI as the catalyst to decide which parts of that stack no longer warrant separate infrastructure, controls and operating expense.
AI’s Biggest Payments Opportunity Is Everything Before the Payment
Rather than treating AI as another automation layer bolted onto fragmented payment systems, companies are confronting a more fundamental problem. Their payment infrastructure often lacks the structured, contextual data AI needs to make useful decisions.
Ng said complex legacy systems can struggle to answer such basic questions as, “What is the exposure to one counterparty for payments being processed over four different systems? What is the total cash in versus cash out?”
But one unlikely catalyst for change is already coming from tax authorities. Governments around the world are expanding electronic invoicing requirements, requiring businesses to replace PDFs and other loosely structured documents with standardized transaction data. The mandates are typically treated as a compliance headache, but Ng said they represent something more consequential.
“What e-invoicing gives us that is really exciting for payments is a structured schema, validated counterparty identified set of transaction data at genuine scale arriving over a network instead of in a PDF,” he said, adding that this amounts to possibly “the largest involuntary data cleansing exercise for B2B payments, probably in commercial history.”
Companies would ordinarily have to justify the cost of creating that foundation against an uncertain future AI payoff. A mandate changes the calculation as the investment has to happen anyway.
Autonomy Is Not the Point of Payments Modernization
Governments are, in effect, inadvertently building some of the infrastructure agentic finance will need to scale. Structured invoice information can give an AI system context about what is being purchased, who is being paid and why a transaction exists. Combine that with payment history, counterparty information, policies and payment-rail data, and artificial intelligence can potentially help determine how a transaction should be executed while improving reconciliation and compliance.
“This builds a solid substrate for agentic payments that no one is talking about in that language,” Ng said.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
That does not mean enterprises should hand AI agents the corporate checkbook. For all the excitement around autonomous agents, Ng argued that payments expose the limits of the concept particularly clearly. Money is difficult to recover once sent, and a transaction executed incorrectly can create fraud, sanctions and compliance consequences.
At the same time, an agent does not need final authority over a payment to create substantial economic value. It can collect information, validate instructions, screen counterparties, recommend a payment rail, flag anomalies and prepare transactions for approval.
“Almost all compliance monitoring is rules-based today,” Ng said, adding that AI can perform much of that evaluation while pushing exceptions and low-confidence decisions to people.
The biggest near-term opportunity may not be finding another workflow to give an agent. It may be recognizing that billions of dollars already being spent on mandatory financial infrastructure are creating assets that can be reused for AI.
Office of the CFO Moves From Software Budget to Control Plane
Ng’s proposed architecture points toward a different allocation of finance technology spending. Instead of maintaining infrastructure around individual payment rails, companies can increasingly concentrate investment in what he describes as an overarching control plane: shared data, policy and approval controls, AI-driven recommendations and access to multiple payment rails.
“Multi-rail in one place,” Ng said, contrasts with a legacy environment where “so much legacy payments is infrastructure per rail.”
This does not necessarily mean ripping out every existing payment system. Ng’s argument is closer to an inversion of the traditional modernization strategy. Companies have often attempted large payments transformations by replacing legacy infrastructure and hoping operational improvements follow. Ng instead describes a layer capable of coordinating systems that may continue changing underneath it.
“The models will change and will evolve when we think about AI,” he said. “The payments ecosystem around us will change and evolve with new payment standards, new payment schemes, more instant payments.”
A shared layer does more than determine whether a transaction should proceed. Ng points to “true intelligence” around risk and compliance, working capital, counterparty exposure and reconciliation.
Those are traditionally treated as adjacent finance functions. A richer payment architecture begins connecting them.
Watch the full PYMNTS TV interview with Andrew Ng to hear more about:
- Why AI could force CFOs to rethink what they’re already spending on payments.Ng says companies have accumulated payment infrastructure one decision at a time, leaving finance teams with duplicated systems, controls and workflows that AI could help consolidate rather than simply automate.
- Why the real payments opportunity is shifting from execution to the control plane.Ng says payment execution is increasingly commoditized, putting more value on the shared intelligence layer that can connect payment decisions with risk, compliance, working capital, counterparty exposure and reconciliation.
- Why full payment autonomy may be the wrong AI target.Ng argues that “governed autonomy” is the more practical model, with AI preparing, validating and recommending transactions while humans retain oversight of consequential decisions such as releasing funds.
Add as Preferred Source
