How bank CFOs can govern AI in ERP systems

If you’re a finance leader at a bank or capital markets (BCM) institution, you may have just spent a lot on a modern Enterprise Resource Planning (ERP) or Enterprise Performance Management (EPM) systems. You know AI can help deliver the kind of value that you and your stakeholders demand:

• Faster outputs
• Lower costs
• Impeccable compliance
• New insights for the business, based on more holistic and timely data
• Automation of routine tasks, to free up your people for higher value work

There’s a path to achieve these benefits without a big upfront investment: the AI capabilities embedded in your enterprise platforms. And if you activate AI within these tightly governed platforms, you can address other concerns too, such as compliance, controls, and data and output quality. You can also tackle what may be a key reason why many companies haven’t yet seen much upside from AI: Only 13% of workers actually use it daily. But if you provide people with reliable, risk-managed AI within the tools and workflows they already use, AI adoption can surge.  

For banking finance, AI has to be transparent, explainable, and auditable, just like your other tools and processes. But that’s not a reason to wait to deploy AI—on the contrary. If your core ERP platform is up to date, it not only comes with “AI inside”: specialized tools designed for accounting, procure-to-pay, close and reconciliations, planning and forecasting, and more, but it can also provide the capabilities you need for governance and compliance:

• End-to-end data lineage across general ledgers (GLs) and subledgers
• Embedded controls, approvals, and segregation of duties
• Human-in-the-loop and override mechanisms

If you activate agentic AI within these tightly governed platforms that you are already using every day, you can apply risk-tiered oversight, keep key decisions in human hands, and maintain audit trails. With governance largely solved, you can advance quickly and confidently with AI.

Here’s how it can work in four key workflows that typically can be up and running in under four weeks, starting with automation that delivers productivity and speed, and then advancing to agentic workflows that can create new, strategic value for you and the business.  

If your finance function is like many in BCM, you have teams of analysts manually determining and coding account combinations for non-purchase order (non-PO) invoice lines. It’s a lot of human effort spent on repetitive work. And you’ll likely have some variance and posting errors that may cause problems for reconciliation.

An AI tool that comes embedded in your ERP can change that. An intelligent approach to account combination can:

• Analyze your historical invoice data to identify patterns and coding rules
• Generate suggested entries for your analysts to approve or improve
• Learn from this human feedback to improve its suggestions
• Observe high prediction accuracy (frequently above 80–90%) when sufficient, high-quality historical data is available, though results may vary by data quality and volume.

This approach not only elevates speed and reliability. With routine tasks automated, people can spend more time on complex risk postings and on more strategic work. And since human analysts should approve the AI’s decisions, you can manage risks and grow stakeholder trust while retaining the governance, audit trails, and workflow approvals you already use.  

In BCM, Financial Planning and Analysis (FP&A) teams have extra work: projecting balance sheet movements, liquidity, net interest income sensitivity, capital adequacy, and stress scenarios—all under regulators’ watchful eyes. Accomplishing these tasks usually requires manual modeling across siloed systems. That limits speed and can lead to errors, potentially causing issues for interest rate risk positioning, asset liability and capital buffer management, and more.

With the help of AI capabilities embedded in your EPM—a mix of machine learning (ML) models, algorithmic forecasting, scenario modeling engines, and explainable analytics— you can transform FP&A. These AI tools can:

• Collect and clean data from multiple, previously siloed sources
• Spot patterns in historical and real-time data
• Identify key drivers of forecast variances
• Enable rapid scenario modeling to assess a range of possible outcomes
• Produce confidence ranges and risk signals for key forecasts

The result? Less manual effort, faster cycle times, and richer, earlier insights—giving your team new "free time” to turn these insights into action. Since the tools work within your ERP’s governance and oversight framework, they can offer the traceability, risk management, and compliance that you and your stakeholders demand.  

Accounts payable (AP) is no simple matter in BCM. Regulations are strict. Vendors often cross legal entities, business lines, and jurisdictions. And for your AP teams, work is intensely manual—and if they make a mistake or fall behind, it can impact accrual accuracy, expense recognition timing, working capital reporting, and more. Problems can even cascade across legal entities and cost centers, making remediation painful.

“Vendor assist” AI agents, which you can embed in your ERP, can lighten your people’s burden and reduce the risks to you and the business. If you activate vendor assist agents and carefully customize them for your requirements, they can:

• Continuously scan the shared AP inbox
• Interpret and classify vendor inquiries
• Retrieve real-time invoice, payment, and vendor data
• Analyze data to support both their own decisions and those of human decision-makers
• Answer routine inquiries and escalate complex ones to the appropriate person

In PwC’s experience, agents can assist on roughly 75% of inbound AP inquiries. Their automation boosts consistency. Their speed and access to data can help spot issues early. And, since they can be deployed natively in your ERP, they can be time-stamped, traceable and follow your rules for role-based access and segregation of duties. To further manage risks, your people can oversee them and make any high-risk decisions.  

Closing ledgers can be one of your most difficult tasks because potential issues rarely stem from a single, obvious error. Instead, there are subtle shifts across balance sheets, portfolios, and legal entities which skilled people—using periodic reviews, static thresholds, and their own experience—work hard to spot. It’s a time-intensive process, and it tends to be reactive. And in complex, multi-entity BCM environments, some critical signals may get missed.

For such a subtle responsibility, you can’t just switch on an AI tool. But you can draw on AI capabilities that come with your enterprise systems platform—and operate within its tightly governed data and secure environment—to create customized detection and reconciliation agents. These agents can:

• Continuously monitor general ledgers, subledgers, and planning systems
• Compare this data against historical patterns, forecasts, and peer portfolios
• Detect anomalies—such as liquidity and funding shifts, or balance sheet moves that don’t align with rate expectations—even before they breach formal thresholds
• Offer your personnel data-backed signals, such as reconciliation trends that previously led to late-stage adjustments, or indications that forecast assumptions are out of date.

This agentic capability can help you “look around corners” to spot problems before they start. That could mean fewer late-stage journal adjustments and re-forecast cycles, and faster, smoother ledger closes and related reporting. Thanks to these agents’ explainable, traceable tracking of data and signals, you can also respond faster to audit and supervisory inquiries.

Humans remain in charge, making the decisions. But they have less manual work and more timely data analysis—done by agents. Here too, these agents can follow your role-based access controls and governance, preserve segregation of duties, and operate within your existing monitoring and review processes.  

For many banks, one of the top barriers to AI success isn’t the tech, but rather knowing how to help people use AI capabilities effectively, compliantly, and securely. People will need new skills related to judgment and critical thinking, rather than task proficiency, since AI can do so many finance tasks. For example, your controller may no longer need to manually scan balance sheet accounts. Instead, they can evaluate why an anomaly agent flagged three accounts and determine whether the issue is timing, business-driven, or a true control concern.  

But banking finance professionals will need to understand why AI is making a choice or recommendation, so they can know when to challenge and override AI—since people should be accountable for decisions and outputs. That may require your finance professionals to cultivate broader process awareness. For example, to keep with up with agents operating across procure-to-pay, record-to-report, and planning, FP&A analysts should also understand these processes too (including upstream AP accrual behavior and vendor payment timing), so they can interpret forecast confidence bands.

Some people may be unnerved by the change in their roles—or fear that AI may replace them. But if you give them the proper support and make clear how AI can make them more valuable, they can become AI enthusiasts.

Achieving an AI-powered finance function within banking may be a journey—but thanks to your existing platform’s embedded capabilities and governance, you can get started today.

1. Start where the “lift” is easier. Common initial use cases typically share three characteristics: They’re high-volume, repeatable processes; they're already running on your current platform; and they have pain points that your platform’s existing AI capabilities can address—within your existing governance and control framework.
2. Measure business outcomes—not AI metrics. To measure and drive concrete AI value, align your stakeholders around business relevant key performance indicators (KPIs) for AI initiatives, such as close speed, forecast precision, or cost-to-serve.
3. Make accountability explicit—and built in. Even in early, ERP-centered use cases, be explicit about how to oversee and observe AI outputs, when and how humans should intervene, and who owns which decisions. Responsible AI can enable compliance and speed up value recognition by preventing costly, time-consuming mistakes.
4. Revisit your workforce strategy. A new, AI-enabled finance function begins with redefining roles and clarifying how humans and agents should work together. As agents execute tasks, people’s responsibilities should shift to orchestrating teams of agents, managing exceptions, and owning decisions and outcomes. Over time, your workforce model—including its shape, size, and composition—will likely change. Continue to invest in entry-level talent to build AI, business, and technical skills, and redefine expectations of managers to help them lead hybrid teams with both humans and agents.
5. Take the lead on a new workforce. To help your people become enthusiastic about cultivating a new skill and mindset for AI, you and your fellow finance leaders can model human-agent ways of working: demonstrating the “why” and “how” of AI use, including ownership of decisions and outcomes. Engage with your grassroots AI accelerators who can energize teams from the ground up, and offer psychological safety, so people feel confident that AI can grow their value.

Competitive advantage awaits

A key misconception for finance functions in banks is that you have to wait for internal IT solutions to arrive. The opposite is true: Mature technologies and tested controls already exist. Many are or can be embedded in your ERP and can operate within your current governance framework. This can quickly boost your speed, precision, and ability to deliver insights to the business.

It’s a competitive advantage for you and your stakeholders, waiting for you to seize it.