Thredd: Inside the Payments Overhaul Powering Faster, Smarter Transactions

Banks and FinTechs, aided by issuer-processor Thredd, are embracing agentic artificial intelligence (AI) to pioneer intelligent transaction orchestration, moving beyond traditional automation to unlock improved efficiency and customer experience.

This shift aims to redefine how financial institutions issue and manage cards, optimize real-time decisions and combat sophisticated fraud, charting a course toward a more responsive financial ecosystem.

The journey toward this intelligent future, however, is not without its challenges, primarily centering on establishing the infrastructure to support widespread AI deployment.

Infrastructure Shifts

Edwin Poot, chief technology officer at Thredd, pointed to the underestimation of these infrastructural demands. His comment came as part of the “What’s Next in Payments” series focused on agentic AI.

“I think what people usually tend to forget is … once this takes off and you’ll deploy agents per transaction, this will require changes to the infrastructure and the ways in which you manage those agents. People can underestimate that,” he said.

He elaborated that while many focus on specific use cases, the crucial question remains whether the underlying infrastructure is prepared to support potentially thousands (or tens of thousands) of agents running concurrently and accessing application programming interfaces (APIs) at speeds exceeding human capabilities.

This intense activity places immense strain on existing APIs and infrastructure, further complicating the need to authenticate and ensure agents are not malicious. Scaling these solutions to a large, business-ready level remains the central challenge for the coming years.

In response to these escalating demands, Thredd is rebuilding its platform to underpin the expansive capabilities of agentic AI. This overhaul is centered around event-driven, real-time APIs that feature fine-grained, policy-driven access to data, a capability Poot deemed essential for enabling agent decision-making in milliseconds.

Given that multiple agents may run in parallel for individual clients, which can strain infrastructure, Thredd is investing in serverless compute to scale decisions.

Thredd is also integrating tokenization-as-a-service to secure agent transactions and implementing a federated data access layer.

This layer ensures that agents only access data they are explicitly allowed to see, including device context and risk signals across various domains, all while upholding privacy constraints. Fundamental building blocks include robust APIs with intricate policy mechanisms, along with infrastructure capable of supporting  agents, train models and serve memory across clients — which helps satisfy the demands of a shifting regulatory landscape.

Testing the Models

Crucially, Thredd is also opening its simulation environments to clients, allowing AI models to be thoroughly tested against lifelike transaction flows before deployment. This testing is considered critical for building trust and ensuring consistent, predictable agent behavior.

When Thredd processes transactions, it receives them via schemes from the acquirer into its platform, where the speed of decision-making is paramount. The process typically involves applying a series of heuristic rules, sometimes sequentially and at other times in parallel. In certain scenarios, Thredd may forward the authorization request to its client for a decision, subsequently updating this information in its ledger before transmitting it back to the schemes, Poot added.

More often, Thredd makes the decision for its clients by leveraging its own rules, which incorporate a fraud engine, 3D Secure protocols and other proprietary checks, all aimed at completing the transaction swiftly.

The introduction of agentic AI enhances this process by enabling a much more nuanced understanding of transaction intent, reducing the need to block transactions unless necessary.

“Agent technology will be everywhere,” Poot predicted.

The adoption of AI-driven processes is delivering a measurably better client and end-user experience. With a memory component, agentic AI can analyze historical transaction patterns and intent at the card or instrument level, enabling faster and more accurate decisions without impeding the end-to-end processing flow. This leads to a reduction in false positives, such as incorrect fraud detections, benefiting both Thredd and its clients.

Poot emphasized that each agent must be assigned a specific role to maintain efficiency and speed. For instance, one agent might focus solely on the core processing flow of a transaction. However, other specialized agents can be deployed to handle post-transaction scenarios, such as when a client or end-user disputes a transaction.

AI agents can act on behalf of consumers or FinTech applications to optimize real-time decisions, such as advising which card to use, when to split payments, or how to defer or convert installments.

The company envisions what Poot calls “a completely self-serve model” where clients can select from a library of agents, deploy them and activate them for specific use cases or scenarios.

Battling the Fraudsters

Thredd is enhancing its fraud approach by modeling agent behavioral profiles to differentiate between legitimate automated processes and suspicious automation.

The company is integrating explainable AI (XAI) to provide justifications when a transaction is flagged, a crucial feature in environments where agents act autonomously without human clicks.

Looking ahead, agentic AI is set to reshape the landscape of B2B issuing and embedded finance. Autonomous agents will be capable of dynamically issuing funds, managing virtual cards, optimizing working capital and orchestrating vendor payments across different geographies.

This presents a “goldmine of opportunity” for transforming large, global and cross-border transactions. Early pilot programs suggest that agentic AI can shrink cross-border settlement times and improve foreign exchange rates, impacting supply chains and various commercial settings. Key metrics for gauging success in this arena include speed to decision — as agents can execute tasks far quicker than humans — and an uplift in conversion rates for clients.

Poot noted that even 50-millisecond reduction in transaction latency, enabled by an AI agent, can yield a “huge difference.”

Other indicators of success include higher authorization approvals through smarter routing and increased repayment conversion rates by offering real-time installment options.

As the ecosystem matures, a key future metric will be agent-to-agent interoperability, as multiple agents with different roles will need to seamlessly hand over communication flows. Thredd’s ambition is to become the network where AI agents not only transact but also collaborate, defining the scalability of this emerging ecosystem.

Thredd is also piloting agent identity registries, assigning each AI agent a verifiable ID and an access policy attached to its owner, crucial for enforcing auditability.

“We’re just scratching the surface,” of AI’s potential, Poot said. “The future looks really bright.”