Ask an accounts payable (AP) analyst how they spend the Monday of close week, and much of the answer may involve gathering the information needed to approve, code, or resolve invoices. They open the enterprise resource planning (ERP) system and review the exception queue. One invoice may need additional transaction details. Another could require a matching expense report. A third requires clarification from a colleague about an unfamiliar charge. Before deciding, they must assemble information scattered across multiple systems and stakeholders.
The decision itself may take only minutes. The challenge can lie in gathering the information needed to make the decision confidently.
This pattern often exists across finance, procurement, customer operations, and service teams. Organizations rely on skilled employees to assess risk, resolve exceptions, and make decisions, yet much of their day can be spent navigating applications, documents, emails, and internal systems before they can apply that expertise.
This is a consequence of a successful, decades-long effort to digitize the enterprise. Companies can adopt specialized systems for finance, procurement, travel and expense management, payments, human resources, and customer operations. Each system can become the authoritative source for a specific business function.
A challenge is that important business decisions rarely depend on just one system. Employees have effectively become the integration layer, moving between applications to gather evidence, reconcile information, and establish context before they can act. That work between systems is where time can disappear, queues can build, and expertise can be consumed by information gathering instead of business judgment.
Roadblocks to Automate
Traditional automation often performs well when work is structured, repetitive, and predictable. Rules engines, workflow automation, and ERP matching logic can process large volumes of transactions when data follows predefined formats and paths.
The challenge can begin when information falls outside those boundaries. A supplier includes a note that says, “revised per our call.” An employee explains an unusual expense in a paragraph of text. A policy establishes guardrails but still leaves room for judgment. People can interpret these situations, but traditional automation may struggle to use the surrounding context.
Much of enterprise work combines structured and unstructured information. Historically, that combination often placed many processes in a gap between automation and human review, where both scale and judgment were required.
From Strategic Idea to Operational Value
With the emergence of large language models (LLMs) and AI agents, organizations have new opportunities to address this challenge.
To explore what that could look like in practice, Visa and Workato have developed two illustrative reference demos outlining implementations focused on accounts payable, where corporate payment, expense, and accounting policies can help meet the realities of exceptions and nuance. These examples are prototypes designed to help explore potential workflows; they are not descriptions of autonomous payment decision-making or guaranteed business outcomes.
The premise is straightforward: AI agents may help reduce the manual preparation that slows exception handling by gathering information from disconnected systems, interpreting unstructured inputs, applying configured business policies, and preparing recommendations for human review. Rather than replacing employees, this approach is designed to help shift work from investigation toward evaluation. Human review, approval, and accountability can remain central before any action is taken.
How AI Agents May Support Accounts Payable
Payment orchestration
In one illustrative implementation developed by Visa and Workato, a manufacturer is preparing a weekly payment cycle with approved invoices scheduled for payment:
- In this scenario, the company has already determined what it owes suppliers. The remaining questions are operational: How should each supplier be paid? Which available early-payment discounts align with configured business rules? Which invoices require additional review before funds are released?
- The implementation then shows how the AI agent can review the company’s payment policy, retrieve approved invoices from the ERP, review supplier payment history, identify discount opportunities based on configured rules, and apply routing criteria. The AI agent can then prepare a proposed payment plan for an authorized analyst, including potential discounts, exceptions held for review, and invoices that may be eligible for virtual card. The agent does not approve payments or make final decisions. Authorized personnel retain control of payment decisions, policy exceptions, and approvals.
Expense exception management
The second illustrative implementation focuses on travel and expense card transactions in an exception queue during close week. These transactions may include line-item details, employee explanations, policy nuances, and contextual information that existing automation may not use as effectively.
- In this scenario, the AI agent can retrieve available transaction details, gather line-item information, match charges to supporting expense reports, evaluate each item against configured policy criteria, and prepare a prioritized review queue. Routine items may be presented with recommended coding. Exceptions can be accompanied by available supporting evidence and, when needed, a request for clarification from the traveler.
- This approach is designed to help streamline manual investigation and give reviewers a more complete set of information. It does not replace the reviewer’s judgment or the organization’s established approval processes.
What Can Make This Practical
The architecture behind these examples can build on systems organizations already use. ERP can remain the authoritative system for approvals, invoices, accounting entries, and financial records. Existing business applications can continue performing the functions for which they were designed. The agent can then operate across those systems, and can help with tasks such as retrieving information, interpreting supporting documents, applying configured policies, and preparing recommendations with traceable evidence for human review.
In these illustrative implementations, Workato can provide the orchestration layer used to connect core applications, departmental tools, data platforms, repositories, and communication channels. In those scenarios, Visa can contribute payment controls, transaction data, authorization services, and traceability. Together, these components are designed to help ground recommendations in available evidence and established controls while helping keep human approval in the workflow.
This model may be particularly relevant for exception-heavy workflows with recurring volume, fragmented evidence, written policies, and skilled employees who spend substantial time gathering information before evaluating it. Accounts payable can be a useful starting point because exception handling often brings together transaction data, written explanations, policy nuance, and context from multiple systems.
Streamlining the Work Between Systems
For more than two decades, organizations have frequently invested heavily in digitizing individual business functions. Finance, procurement, customer operations, and other teams often now operate on capable platforms that produce valuable information.
The next opportunity may lie in connecting that information across systems. Policies reside in documents, transaction data lives in operational platforms, and supporting evidence appears in emails, reports, attachments, and third-party applications. Bringing those elements together remains a persistent source of friction in enterprise operations.
AI agents can offer one potential approach. By gathering information, evaluating context against configured policies, identifying exceptions, and preparing recommendations, they may help transform investigation-heavy processes into more streamlined review workflows. The goal is not to replace human judgment, but to help experts spend more time applying it.
Through their collaboration, Visa and Workato are exploring how AI agents may help organizations bridge information gaps, reduce operational friction, and support employees in making informed decisions. The first wave of enterprise software digitized individual systems. The next wave may help businesses operate more effectively across them.
Ship governed AI agents that work across your customers’ systems with Workato ONE for SaaS Platforms.
Disclaimers: This article contains depictions of ideas, concepts and details currently in the process of deployment, and should be understood as a representation of the potential features of the fully-deployed feature set.
These material and recommendations are provided “AS IS” and intended for informational purposes only. Visa neither makes any warranty or representation as to the completeness or accuracy of the information within this document, nor assumes any liability or responsibility that may result from reliance on such information. Materials and recommendations should be independently evaluated in light of your specific business needs and any applicable laws and regulations. All brand names and logos are the property of their respective owners, are used for identification purposes only, and DO NOT imply product endorsement or affiliation with Visa.
