AI is moving closer to the moment a payment actually happens, evolving from a tool that supports decisions to one that can help orchestrate payment processes in real time. In buy now, pay later (BNPL), its role is expanding beyond analysing risk, detecting fraud, and personalising customer experiences, to actively participating in how transactions are initiated and managed.
Agentic AI could change that relationship. Autonomous AI agents can increasingly act within defined parameters to initiate purchases, make decisions, and coordinate payment processes. Klarna and Stripe’s move to bring BNPL into AI-powered checkout experiences offers an early indication of where this could lead. This raises a bigger question for BNPL providers and their payment partners: what happens when AI becomes an active participant in the transaction itself?
When AI starts acting on payment decisions
Much of the discussion around AI in financial services has focused on prediction. Models analyse large volumes of data to assess risk, identify suspicious activity, or determine the likelihood of a particular outcome. Agentic AI introduces a different capability: action.
An AI agent can interpret information, select an action based on predefined objectives and controls, and execute the next stage of a workflow. This matters because BNPL transactions involve multiple interconnected processes, including authorisation, fraud checks, payment routing, settlement, reconciliation, and exception management.
The potential is for payment operations that respond continuously to changing conditions. Rather than simply identifying an issue, an agent could determine the appropriate response and execute it with clear pre-approved rules and clearly defined boundaries.
BNPL scale creates a different operational challenge
The simplicity of BNPL for consumers masks considerable complexity behind each transaction. In March 2026, more than 9 million transactions were recorded across nearly 4 million consumers, worth £621 million. Average transaction values remained relatively modest at around £60, suggesting consumers are typically using BNPL for lower-value, everyday purchases rather than larger big-ticket items.
For providers, those volumes put greater pressure on operational efficiency. Regardless of value, each transaction must still pass through the necessary payment, risk, fraud, and settlement processes.
Agentic AI could help manage this complexity by embedding intelligence across the payment lifecycle, coordinating validation, fraud controls, routing, settlement, and exception management in real time. When a payment encounters an exception, an autonomous agent could assess the issue and determine the appropriate resolution, while intelligent routing could adapt to transaction characteristics, operational conditions, cost considerations and settlement requirements.
Fraud becomes a moving target
The same technologies making commerce more intelligent will also alter the fraud landscape. AI can help providers detect unusual behaviour more quickly, but malicious actors can use similar capabilities to automate attacks and adapt their techniques.
Agentic systems could strengthen the defensive side by monitoring transaction behaviour continuously and responding when patterns change. An agent could respond to an anomaly by initiating additional checks or changing how the transaction is handled within predefined controls, with cases escalated when human judgement is required.
Greater autonomy will require equally robust accountability. In regulated financial services, effective implementations will need to combine automation with clearly defined confidence-based controls and auditability, alongside clear routes for human-in-the-loop intervention.
Setting boundaries for autonomous decisions
The more responsibility organisations delegate to AI, the more important it becomes to understand how decisions are made and where authority sits. A system recommending an action presents one level of risk; a system capable of initiating a transaction or resolving an operational exception presents another.
Governance therefore needs to be embedded into how agentic systems operate. Organisations must determine which actions agents can take independently, which require approval, what data they can access, and what happens when behaviour falls outside expected parameters. Frameworks such as the NIST AI Risk Management Framework provide a useful general starting point for thinking about how trustworthiness can be incorporated into the design and deployment of AI systems.
For BNPL providers, this also requires visibility across the payment lifecycle. Human oversight becomes most effective when teams can understand what an agent has done, why it acted, and when intervention is necessary.
Competition moves beneath the interface
Agentic commerce could also change where competition takes place. Providers have traditionally competed heavily on the convenience of the customer interface, but AI agents may increasingly mediate that experience by evaluating purchasing options and initiating transactions on a consumer’s behalf.
This places greater emphasis on the infrastructure beneath the interface. That infrastructure increasingly depends on intelligent orchestration capable of coordinating transactions across multiple payment rails, systems, and operational workflows. BNPL providers and their banking and payment partners will need systems capable of processing agent-initiated transactions securely and reliably while supporting the payment choices consumers expect.
The competitive question extends beyond which provider offers the smoothest checkout. Increasingly, it will concern which providers can support intelligent, automated commerce at scale without sacrificing reliability, interoperability, oversight or payment choice.
Preparing BNPL for agentic commerce
Agentic commerce is still developing, and it would be premature to assume every purchasing decision will soon be handed to an autonomous AI agent. What is becoming clearer is that AI is moving closer to the point at which transactions are initiated and executed.
For BNPL providers, preparing for that shift will require visibility into how AI-driven decisions affect payment operations. Fraud detection, credit assessment, routing, and exception handling can each benefit from intelligence, but the larger opportunity comes when those capabilities work together across the payment lifecycle and connect securely with the underlying payment infrastructure. As payment ecosystems become increasingly multi-rail, intelligent systems will also need to operate across traditional payment networks, real-time payment infrastructures, and emerging digital payment models while maintaining consistent controls, visibility, and interoperability.
The next stage of AI in payments will be defined by how safely intelligent systems can act on the information available to them. For BNPL, that could mark the transition from AI as a tool supporting the transaction to AI becoming part of the operating layer through which the transaction itself is managed.
