Transaction banking has spent decades automating the movement of money. Nearly all domestic payments now achieve straight-through processing (STP), as do more than 90% of cross-border payments in many corridors. That success reduced the segment’s sense of urgency about investing in agentic and generative AI. Because most transactions already flow without human intervention, the remaining opportunities for improvement seemed limited.
That calculation has changed. Instant payments, agentic AI, and programmable money are converging to make treasury more continuous and automated. Money can be transferred around the clock, agents can initiate and manage treasury activities, and stablecoins and tokenized deposits can execute transactions that meet predefined conditions based on smart contracts. As more of this activity flows into corporate systems, the role of transaction banks will extend beyond processing payments to include participating in decisions that determine when, where, and how money moves.
Banks still have substantial work to do inside their own operations. Manual effort remains concentrated in exception handling, payment investigations, trade documentation, and client service at a time when enterprise resource planning (ERP) and treasury management system (TMS) providers, along with specialized fintechs, are embedding intelligence directly into corporate workflows. To stay competitive, transaction banks must use AI to eliminate residual operational friction while ensuring that their platforms remain relevant to an increasingly automated corporate treasury.
Corporate Treasury Is Evolving into a Machine
Historically, corporate banking has relied on human relationships and manual interactions through bank portals. Today, multinationals and other large companies are systematically standardizing and automating core finance functions, including procure-to-pay, order-to-cash, liquidity management, and foreign exchange hedging.
This shift mirrors a broader evolution toward agentic workflows and autonomous execution. (See the exhibit.) Initial AI adoption focused on aggregating transaction data across ERPs and banks to optimize real-time cash forecasting, simulate risk scenarios, and flag fraudulent transactions. For frontrunners, however, AI is moving from insight to action. Instead of treasury teams logging into bank portals to execute transactions, AI agents are starting to operate directly on behalf of the CFO or treasurer. They route payments in real time on the basis of cost, speed, and STP criteria while managing currency exposure and cash pooling within defined corporate policies.
Software providers and fintechs are already capitalizing on this transition. TMS providers such as Kyriba, Serrala, and Bottomline Technologies offer AI-native cash forecasting and risk tools built directly into corporate workflows, and platforms such as Ripple Treasury manage fiat and digital assets simultaneously.
Others, such as Ramp, are going further by granting AI agents direct access to accounts from which they can transact. These agents can complete real financial tasks autonomously, including paying suppliers, issuing cards, and transferring money, leaving human treasury professionals free to focus on oversight, approvals, and anomalies.
In a machine-to-machine model, control of the orchestration layer will help determine the provider that ultimately owns the client relationship and captures the higher-value work.
Agents Can Give Incumbents the Edge They Need
Modernization during the past decade has given incumbent banks more muscle than many observers assume. The migration to ISO 20022 messaging and the shift from batch to real-time processing solved the primary bottleneck for AI adoption by forcing rich, standardized data into every transaction layer. Banks can now use that foundation to automate their operations and help clients redesign finance functions around instant payments. Five areas of action can help banks defend margins and gain the intelligence layer they need:
- Eliminate the document bottleneck in trade finance. Trade finance remains stuck at a 30% to 70% STP rate because many documents still require subjective review. AI can materially increase automation in this area. By deploying AI for document intelligence—an effort that has yielded accuracy levels of better than 85% for early adopters—banks can automate the entire trade life cycle, from basic bills of lading to advanced anti-money-laundering controls.
- Move payment operations closer to zero-touch. The remaining non-STP payments generate a disproportionate share of operating costs. AI can automate wire repairs (a capability that institutions such as BNY have already proven at scale in production); distinguish true positives from false positives in sanctions screening; document the reasoning behind decisions to audit, and detect anomalous transaction patterns to avoid fraud before funds are routed. Agentic models can also handle complex customer service requests, including disputes, chargebacks, reporting requests, and technical issues in account-to-account payments.
- Accelerate legacy architecture modernization. Transaction banking remains severely constrained by outdated technology platforms wrapped in legacy code. Rather than deploying AI purely as a front-end feature, leading banks are using it as a high-leverage structural software engineering tool. A recent BCG proof of concept with a G-SIB transaction bank demonstrated 20% to 40% efficiency gains in software engineering projects by using BCG’s agent-enabled Dark Software Factory. By applying AI to automate requirements analysis, test case generation, test automation, and code documentation, banks can rapidly accelerate legacy system translation and mitigate the execution risk of core platform replacements.
- Protect the corporate treasury relationship. As AI agents take on more responsibility for payment routing, liquidity, and risk decisions, the center of the banking relationship will shift from transaction execution to the rules and intelligence that govern those decisions. Banks that embed these capabilities into client workflows can remain their clients’ primary adviser and orchestration partner.
- Use AI to improve the efficiency of sales and key account management teams. Transaction banking product specialists often work hand in hand with relationship managers to deepen connections with corporate customers through current accounts, domestic and cross-border payment execution, and cash-pooling solutions to generate valuable and stable fee income. AI tools can help relationship managers qualify leads, strengthen retention and cross-selling capabilities in business banking, and improve the speed and quality of content production in the request-for-proposal process.
Smart Deployments Start with Strategy
Rather than getting stuck in experimentation, transaction banks can build scale and value in four ways:
- Quantify the true customer experience and financial cost of non-STP activities. Banks should audit their transaction value chains to identify where and how often manual intervention occurs, and at what cost to customer experience, agility, and the bottom line. They can then rank use cases by value, implementation difficulty, and time to deliver, directing AI investment first to the largest addressable sources of cost.
- Architect for machine counterparties. Leaders need to assess whether their technology, process standardization, and data layers can support AI beyond isolated use cases. They should also prepare their systems to expose secure, real-time data that AI agents can query and use, with authorization, monitoring, and audit controls designed specifically for nonhuman actors.
- Close the capability and control gaps. Banks should identify the product, engineering, risk, and change-management skills they need to build, govern, and adopt AI at scale. In parallel, they should define the authorization limits, accountability, and compliance controls required before agents can act on behalf of the bank or its clients.
- Decide what the bank must own. Leaders must determine where proprietary control of the agentic treasury layer is essential to protecting the client relationship and where partners can provide greater speed or scale. Those decisions should guide their determination of whether to build orchestration capabilities or to integrate them through a tightly governed ecosystem of hyperscalers and AI providers.
The infrastructure is ready, the technology is proven, and the competitive landscape is shifting. Transaction banking leaders must act now to build the intelligence layer into their own platforms, thereby ensuring that they will be equipped to serve the automated and AI-supported corporate treasuries of the near future.
