AI can help a corporate treasurer produce a cash forecast within minutes. The harder questions begin when that forecast points to action. Should surplus liquidity be moved, should funding be raised and how much authority should AI have to recommend, prepare, approve or execute those actions?

These decisions are becoming more complex as treasurers operate across more suppliers, markets, currencies and payment systems. Changing supply chains and geopolitical disruption have increased the importance of resilience and flexibility alongside cost and efficiency.

Siddarth Gupta, Head of Global Payments Solutions for APAC Financial Institutions Group and Co-head of Global Non-Bank Financial Institutions at Bank of America, and Chandana Thanthrige, Head of APAC Financial Institutions and Non-Bank Financial Institutions Transactional FX and Commercial Cards, see transaction banking shifting from processing transactions to helping companies manage commerce in real time.

As banks gain access to similar AI models, payment networks and digital platforms, basic payment execution will become less differentiated. Gupta and Thanthrige expect clients to choose banks for their ability to connect networks, optimise liquidity, safeguard transactions and turn data into better decisions. Competitive advantage will depend on combining these capabilities with global reach, local market expertise, resilient infrastructure and trusted advice.

Siddarth Gupta, Head of Global Payments Solutions for APAC Financial Institutions Group and Co-head of Global Non-Bank Financial Institutions, Bank of America

Chandana Thanthrige, Head of APAC Financial Institutions and Non-Bank Financial Institutions Transactional FX and Commercial Cards, Bank of America

From processing transactions to anticipating decisions

Gupta and Thanthrige expect real-time payment systems, application programming interfaces (APIs), ISO 20022 and embedded finance to integrate banks more deeply into clients’ workflows, making information almost as valuable as the movement of money.

By bringing together payment, balance, receivables and remittance data across banking relationships and corporate systems, these tools can give AI the consistent and timely information needed to identify patterns, forecast cash positions and highlight where action may be required.

A treasurer could understand expected cash flows, the reasons for variances, where excess liquidity is held, which collections may be delayed and whether funding or other risk responses should be considered. Banks could also identify whether financing should be provided earlier in the working-capital cycle.

This changes the role of the transaction bank. Its value increasingly comes from helping clients anticipate changes and assess possible responses, alongside executing the resulting transactions.

These capabilities also extend into trade. Bank of America’s Open Account Automation synchronises procurement, logistics and payment data within a common workflow. As companies diversify sourcing, supplier decisions increasingly affect liquidity, foreign exchange exposure, counterparty risk and capital allocation. Gupta and Thanthrige see supply-chain finance supporting supplier liquidity and resilience, including by extending financing to smaller businesses further along the supply chain.

Financial institutions face related pressures involving nostro rationalisation, broader clearing reach and liquidity efficiency. For the Bank of America executives, a banking partner must combine network depth and global consistency with local market knowledge. This is particularly important in Asia Pacific, where advanced payment infrastructure exists alongside wide differences in currencies, regulation and domestic market systems.

Applied AI reaches institutional scale

Gupta and Thanthrige believe applied AI is the technology closest to achieving meaningful institutional scale in transaction banking. It is already being used in fraud detection, cash forecasting, reconciliation and client servicing.

In payments, AI can support instruction validation, routing and exception handling. In receivables and trade finance, it can improve matching and identify risks or financing opportunities. These applications fit within established banking processes and control frameworks, allowing institutions to deploy AI without redesigning the underlying forms of money or lines of accountability.

Bank of America is applying AI across its transaction banking services. CashPro is its principal digital interface through which corporate clients manage payments, liquidity and other treasury activities.

The executives cited CashPro Forecasting, which uses machine learning and account data from Bank of America and other financial institutions to produce forecasts within minutes and provide related insights. Intelligent Receivables applies AI and advanced data capture to match payments with remittance information, while CashPro Chat uses Erica, the bank’s AI-powered virtual financial assistant, to provide account information, track transactions and respond to service enquiries.

Gupta and Thanthrige expect the next stage to move further into predictive and personalised decision support. In treasury, AI could recommend possible actions involving liquidity, funding and working capital, helping clients decide what to do with the information produced.

The executives do not describe the intended outcome as autonomous finance. Bank of America’s approach is based on “augmented intelligence”, with AI supporting analysis while people remain accountable for material financial outcomes.

Defining the limits of autonomous action

Moving AI closer to financial decisions and execution raises questions involving data, identity, authority, explainability, governance and accountability. These become especially important in payments and treasury, where speed and finality can amplify the consequences of errors or fraud.

Gupta and Thanthrige identify five connected requirements. AI models need accurate, permissioned and traceable data. Banks must verify whether an instruction comes from an authorised person, system or AI agent. They must also define what a system is permitted to recommend, prepare, approve or execute.

Governance must cover model selection, testing, monitoring, audit trails and escalation procedures. A person or legal entity must remain responsible for the outcome.

Their preferred path is “graduated autonomy”. An AI system may first forecast a cash position, then recommend a funding or liquidity action and eventually prepare a transaction for approval. Each step requires clearer permissions, stronger identity controls and more complete audit trails. Approval and execution carry greater consequences, particularly when payments are immediate and final.

Banks can therefore begin with lower-risk tasks, retain human review for material decisions and expand a system’s authority after its performance and controls have been demonstrated.

Governance then becomes part of the client proposition. A treasurer considering greater use of AI will need to know who authorised an action, what information informed it and who remains accountable. Cybersecurity, operational continuity and identity controls become integral to how the service is designed and delivered.

Taking responsibility for a multi-rail system

The need for certainty and control extends beyond AI to the payment infrastructure itself.

Gupta and Thanthrige expect correspondent banking, instant-payment systems, wallets, stablecoins and tokenised deposits to coexist because they serve different requirements. Correspondent banking will remain important for global reach, foreign exchange and liquidity, while instant-payment systems provide greater speed and availability. Stablecoins and tokenised deposits may support selected uses requiring continuous availability, programmability or on-chain settlement.

They point to the India-Singapore real-time payment linkage and multilateral initiatives including Project Nexus as examples of how markets in Asia Pacific are extending payment connectivity.

Bank of America has launched a global cross-border real-time payment service for corporate, commercial and financial-institution clients through Swift or CashPro. According to the bank, the service is designed for high-volume, low-value international payments and is intended to provide real-time tracking, full-principal preservation and lower costs. The bank is also participating in shared-ledger initiatives and exploring regulated digital money with other leading banks.

Fintechs and technology providers can address specific payment problems, while banks contribute trusted relationships, balance-sheet capacity, liquidity, security, regulatory discipline and scale. Gupta and Thanthrige see banks using these capabilities to connect clients with different platforms and providers.

Clients want the right amount delivered at the required time, with clear fees, strong controls and end-to-end visibility. A treasurer could specify the delivery time, cost or currency without selecting the underlying rail. The bank could then apply the client’s rules and permissions to choose the route while integrating pre-validation, foreign exchange, sanctions and fraud controls, payment status and reconciliation.

Connecting several networks and providers will also require common data and identity standards and clear responsibility for each transaction. For Gupta and Thanthrige, the urgent question is how to continue innovating without creating new fragmentation or weakening trust.

Technology becomes repeatable, execution does not

Gupta and Thanthrige believe advantage will depend on how effectively banks combine digital capabilities with client relationships, personalised advice, high-quality data, resilient infrastructure and disciplined execution. Successful institutions will start with a client problem, build reusable capabilities and assess them against outcomes such as forecast accuracy, straight-through processing, risk reduction and working-capital improvement.

Bank of America’s AI strategy follows a “build once, reuse at scale” approach. The executives cited Erica as an example of how a reusable AI capability can support both client-facing and employee applications. They also said the bank’s model-agnostic platform allows it to adopt new AI models without rebuilding the products and workflows that use them.

The approach avoids siloed pilots and allows data, governance and risk controls to remain consistent as new models are introduced. Institutional scale then becomes part of the advantage.

Global reach, liquidity and resilient payment infrastructure will remain fundamental, while service and advice become more valuable as basic processing becomes easier to reproduce. In Gupta and Thanthrige’s view, leading institutions will turn widely available technologies into measurable client outcomes by helping clients make better decisions, execute transactions with greater certainty and retain control.