The promise and perils of agentic AI in payments

Insights

• Payment processing remains largely manual, leading to inefficiencies despite advancements in financial technology.
• Current automation lacks true end-to-end solutions, creating an opportunity for agentic AI, which operates autonomously.
• Promising solutions:
  1. Single-use virtual cards: AI agent can secure transactions by generating unique card numbers for purchases.
  2. Programmable money: AI agents could automate transactions via digital wallets and smart contracts.
  3. Automated reconciliation: Agentic AI can automatically match payments with invoices, streamlining the reconciliation process.
  4. Dynamic payment settlements: Agentic AI use real-time data to negotiate and settle payments, adjusting terms based on factors like market conditions or currency fluctuations.

Despite advances in financial technology, processing payments continues to be a largely manual and labor-intensive process. That leaves businesses and consumers alike frustrated by inefficiencies, errors, and delays. The global payments industry handles 3.4 trillion transactions annually, accounting for $1.8 quadrillion in value.

Traditional automation and current AI solutions, while helpful for specific tasks such as fraud detection, fall short of delivering true end-to-end automation of the payment life cycle. This gap has created an opening for agentic AI. These agents act autonomously and are capable of making decisions and executing tasks without human intervention, potentially changing how money moves through the global economy.

Challenges to agentic AI adoption

Of course, the journey from pilot to production is rarely smooth. And the hurdles to adopting agentic AI are formidable.

• The regulatory quandary: Perhaps the most formidable barrier to agentic AI adoption lies in regulatory compliance. The complexity multiplies when considering cross-jurisdictional requirements, as regulatory frameworks vary significantly across markets.
  
  The cost of non-compliance provides a sobering context for institutional caution. The global non-compliance costs have reached $14 billion. Anti-money laundering (AML) fines alone hit a record high of more than $6 billion in 2023. Against this backdrop, banks' reluctance to deploy autonomous agents that could potentially amplify compliance risks becomes understandable.
  
  Regulatory uncertainty compounds these challenges. The Basel Committee on Banking Supervision continues to develop frameworks for AI in finance, but specific guidance for autonomous financial agents remains limited. This ongoing regulatory development creates timing challenges for institutions seeking to implement agentic AI solutions while maintaining compliance certainty.
• The trust deficit: Consumer trust in AI-powered financial services remains a critical constraint. Recent surveys indicate that only 29% of UK consumers would trust AI to make small, automated payments on their behalf. In the US, the percentage of consumers who trust and use AI to pay is much lower at 16%. The trust challenge extends beyond consumer sentiment to institutional confidence. This skepticism reflects broader concerns about AI decision-making in high-stakes financial contexts, where errors can have immediate and significant consequences.
• Technical limitations and investment hesitancy: The technology underlying agentic AI, while promising, faces scalability and robustness challenges for widespread enterprise deployment. Globally, the banking sector spends only 10% of technology budgets on AI, indicating measured investment approaches that reflect ongoing technical and operational concerns.
• The scale of payment inefficiency: The cost of maintaining antiquated payment processes is substantial. Research indicates that processing a single invoice manually costs between £4 to £25 ($5 to $34) in the UK, while costs to process invoices in other geographies fall in the same range, with the majority of expenses attributed to labor-intensive workflows. These figures underscore a broader challenge: Payment processes cost businesses an average of £1.5 million annually ($2 million), representing 12% of operational expenditure.
  
  The inefficiency extends beyond mere cost. Manual payment processing takes an average of eight days per invoice, creating cash flow constraints and operational bottlenecks. This sluggish pace stands in stark contrast to consumer expectations shaped by instant digital experiences in other sectors.

Promising agentic AI use cases on the horizon

Despite current limitations, several applications demonstrate the potential for agentic AI tools that can negotiate, reconcile, and settle payments end to end:

1. The rise of the single-use virtual card

The single-use virtual card is a modest but effective innovation in payments. Unlike its plastic ancestor, the virtual card generates a fresh number for every transaction. Here, the AI agent — acting as the payer — can access only the sum pre-approved by the customer, and only for the specified purchase. The customer, in turn, sets the rules of engagement: What to buy, where, and how much to spend, whether that’s a fixed sum or a flexible range. Visa’s agentic AI and Mastercard’s Agent Pay are already experimenting with such digital guardians, ensuring the agent spends only within its brief and not a penny more. This approach is not only more secure — virtual cards account for just 9% of fraudulent transactions — but also offers granular control to consumers and businesses alike.

2. Programmable money and smart contracts

The next step up the ladder is the AI agent with its own programmable wallet. Imagine a Web3 wallet, complete with a unique blockchain address, holding either stablecoins or tokenized deposits. The agent funds its wallet via open banking — transferring money off-chain from a bank account to on-chain by executing a smart contract. These digital agreements execute only when pre-set conditions are met, reducing the risk of errors and fraud. The agent ensures that funds are released only when goods are delivered, or when a service meets specified criteria. The contract then dispatches the stablecoin to the merchant’s account, or, if necessary, converts it into fiat currency before settling up. In this arrangement, the AI agent handles the nuts and bolts of payment, while the customer retains oversight and control.

Stablecoins are gaining traction — transaction volumes have surged, exceeding $450 billion per month in 2024 and reaching $710 billion by March 2025. The number of unique stablecoin addresses grew by 50% year-on-year to $35 million, suggesting that programmable money is more than a passing fad.

3. Corporate payment transformation

The complexity of business-to-business payments makes them ripe for agentic AI intervention. For example, a supplier’s invoice-to-cash agent can generate invoices and keep up a running dialogue with the buyer’s procure-to-pay agent. These digital emissaries don’t just swap paperwork; they could sift through public news, credit ratings, and economic indicators, and even arrange working capital financing or invoice discounting when the market demands. Once payment is made, the supplier’s agent ensures every penny is matched and reconciled, moving funds across accounts to maximize returns efficiently.

4. Collections and debt recovery: Empathy by algorithm

Collections and debt recovery are rarely associated with empathy, but agentic AI offers a chance to improve both efficiency and customer experience. Agentic AI can predict which accounts are likely to go delinquent, rank borrowers by risk, and recommend corrective measures tailored to each individual’s behavior and preferences. The agent contacts borrowers at the right time and through the right channel, aiming for resolution rather than resentment. An effectively implemented agent could transform debt collection from an exercise in attrition to one of genuine customer engagement.

Making agentic AI adoption work

Realizing agentic AI's potential in payments requires addressing current constraints through measured, strategic approaches:

1. Adopt modular, interoperable tech stacks: Adopting modular, interoperable technology stacks makes it easier to build agentic AI, at scale. The combination of blockchain for transparency, programmable money for automation, and agentic AI for orchestration forms the backbone of this approach. Standardized communication protocols and data formats are essential, enabling agents to interact without friction. This also helps avoid the creation of AI silos, where systems operate in isolation. Moreover, a modular, interoperable stack gives institutions the flexibility to upgrade or replace individual components — be it large language models, orchestration frameworks, or memory modules — as technology evolves, without the need for wholesale system overhauls.
2. Integrate advanced KYC and AML solutions: AI-powered identity verification, blockchain audit trails, and real-time transaction monitoring are essential to meet regulatory requirements. The global RegTech market is projected to exceed $22 billion by mid-2025, reflecting recognition that technology solutions are essential for managing complex regulatory environments.
3. Regulatory collaboration: Proactive engagement with regulators is essential for establishing clear guidelines for AI autonomy and wallet management. The UK’s Financial Conduct Authority emphasizes principles-based regulation, providing a framework for engagement. However, specific guidance for autonomous financial agents remains limited, creating uncertainty for institutions considering deployment.
4. Enhance trust mechanisms: Leverage decentralized identity, multi-signature wallets, and transparent smart contracts to build consumer and institutional trust. Decentralized identity — where individuals or institutions control their own digital credentials, storing them on a blockchain — allows users to make identity verification more robust and less intrusive. Multi-signature wallets add another layer of security by requiring approval from multiple parties before a transaction can be executed. These combined with transparent smart contracts build trust based on strong technological transparency, security, and user empowerment.
5. Invest in talent and training: Teams must combine expertise in AI, blockchain, and compliance — no small feat, given the current talent shortage.
6. Run controlled pilots: Pilot programs offer the most prudent path forward. Virtual cards for e-commerce transactions and automated invoice reconciliation represent low-risk, high-impact scenarios that can demonstrate value while building institutional confidence. These pilots should incorporate comprehensive monitoring and risk management frameworks to ensure regulatory compliance and operational stability.

The path forward

Agentic AI in payments resembles not a big bang but a cautious evolution. The technology exists — Visa’s and Mastercard’s agents, and Circle’s blockchain integrations prove that. Yet until regulators, banks, and consumers align on security and standards, autonomous payments will remain fragmented. For an industry processing $26 trillion in payments annually, even incremental efficiency gains promise substantial rewards.

The technology represents an opportunity to address long-standing inefficiencies in payment processing while creating new capabilities for customer service and operational excellence. However, successful implementation requires careful attention to regulatory compliance, security considerations, and stakeholder trust. The question is not whether agentic AI will transform payments, but rather how quickly institutions can navigate the complex path from promise to practice.