A fundamental shift is underway in how enterprise software vendors are managing AI agent access to their platforms. SAP has moved most explicitly, requiring approval for third-party AI agents—software that autonomously acts on your data—to operate on its systems. ServiceNow has taken the opposite approach, opening its platform to any AI agent through a governed, consumption-based layer. Others across the industry are charting different directions from SAP and ServiceNow. The upshot: The terms on which AI tools access enterprise data are actively changing at different speeds and with outcomes that are genuinely hard to predict.

Layered on top of this is a broader commercial shift, as vendors move from seat-based licensing toward consumption pricing, charging for each operation an AI agent completes. For CFOs, that introduces a new category of cost that is unpredictable, volume driven, and more difficult to model than traditional software contracts. The access landscape and the cost landscape are both in motion simultaneously.

For CFOs, the uncertainty itself is a defining moment. Organizations that have built their own governed data foundation can adapt as vendor approaches evolve. Those that have not are increasingly exposed to whichever commercial decisions their vendors make next. The path to genuine AI value in finance runs through a set of deliberate architecture decisions, and most organizations have not made them explicitly. Here is what they are, and why they now determine who wins.

The destination is clear, but most organizations are stalled on the way there

Leading finance functions are on a journey toward what the industry is calling “autonomous finance,” with AI handling routine processes continuously, humans focusing on judgment and strategy, and finance acting as the enterprise's real-time nerve center. Leaders across industries have demonstrated what this looks like: double-digit cost savings, faster close cycles, and AI-driven forecasting that keeps pace with the business.

CFOs are now turning that ambition toward their own function. Bain's 2026 CFO Survey finds that 83% of CFOs plan AI budget increases above 15% over the next two years, with a significant share directed at the finance function itself. Yet, only 31% currently rate AI outcomes in finance as strongly positive, a gap that widens the longer the underlying data and architecture issues go unresolved.

The gap between these leaders and the majority of companies is not the technology they bought; it turns instead on whether they built a solid data foundation before deploying AI on top of it.

Think of that foundation as a stack. At the base are the systems of record, such as enterprise resource planning (ERP), procurement, supply chain, customer relationship management (CRM), and HR platforms. Above that, a data layer brings information from all of those systems together, standardizes it, and makes it consistent. Above that, finance builds trusted data products, including agreed-upon definitions of revenue, gross margin, and EBITDA from which everyone works. At the top, AI and analytics tools turn that governed data into real-time decisions and insight (see Figure 1). 

Figure 1

The challenge is more complex than it first appears because modern finance can't work from its own systems alone. Business leaders expect finance to connect financial outcomes to the operational drivers that explain them—for example, revenue performance tied to the sales pipeline in CRM, or margin pressure linked to input costs and fulfillment data in supply chain systems, or headcount costs connected to productivity metrics in HR platforms. Each of those connections requires the same governed foundation—namely, a layer where data from systems that were never designed to talk to each other is reconciled and standardized before AI can reliably use it.

The most consequential and most commonly skipped is the middle layer, where data from all those systems is reconciled into a single, trusted version. Without it, AI is reasoning from numbers that different parts of the organization define differently. When vendors restrict which tools can access their data directly, organizations without this layer lose their ability to chart their own AI destiny. The evidence from finance leaders is consistent: Across every major finance process, from financial planning and analysis to tax to accounts receivable, systems integration and data quality together account for 28% to 41% of top-cited AI blockers (see Figure 2).

Figure 2

Six decisions that most CFOs haven't explicitly made

These choices don't get made by selecting a platform. They require deliberate decisions that each have compounding consequences for AI readiness.

How standardized does your financial systems backbone need to be? There is no universal right answer. What matters is having an explicit strategy: When data comes from different systems, where is it reconciled, who owns that process, and how do you prevent conflicting numbers from propagating? The absence of a deliberate answer is itself a structural liability.

What is your default when buying new financial tools? Most organizations run a mix of core ERP and specialist software. The key question is what happens by default when a new tool is needed. Without a clear policy, organizations accumulate one-off integrations that eventually prevent AI from working reliably.

Where does financial truth live? Running reporting directly out of an ERP works at small scale, but it breaks down when AI needs to reason across multiple systems. A dedicated data layer where financial data is brought together, cleaned, and governed is increasingly the make-or-break prerequisite for AI in finance.

How do your systems talk to each other? Direct connections between individual systems are easy to build and difficult to maintain at scale. A managed integration hub—that is, a central layer governing how systems exchange data—creates reliability and independence from vendor-imposed restrictions.

Where should AI sit in your architecture? AI built into platforms works well for high-volume transactional tasks. AI built on top of a governed data layer works better for planning, forecasting, and strategic insight. Deploying AI before the data foundation is ready consistently produces outputs that no one trusts.

Who owns the data, and who owns the platform? Finance should own what the numbers mean (definitions, allocations, authoritative sources); IT should own the underlying platform and integration layer. When these accountabilities blur or when investment is structured as one-off projects rather than ongoing operating costs, the architecture rarely stabilizes, and AI can’t scale.

Implications for CFOs

The window to act ahead of the competition on these decisions is still open but narrowing fast. Three areas where CFOs can make six moves now:

Establish the foundation before adding capability.

• Decide where financial data gets reconciled, and assign clear ownership before approving the next platform or AI investment.
• Agree on financial definitions before building anything on top of them. Who defines revenue? Who resolves disagreements? These questions determine whether any AI system touching financial data produces outputs that the organization can trust.

Control your architecture; don't let vendors control it for you.

• Set a default integration standard before adding more systems. A managed hub as the default creates predictability and independence from vendor-imposed restrictions.
• Let your own architecture roadmap drive sequencing. Platform vendors have their own commercial interests. CFOs who allow those to set the pace will find their AI options narrowing as vendors consolidate control over data access.

Deploy and measure differently.

• Build the data foundation before deploying AI at scale. Governed financial data is the prerequisite for AI investment, not a follow-on. Deploy on top of contested data, and adoption will stall.

• Measure outcomes at the enterprise level. Finance architecture is not a finance project. Time to insight, forecast accuracy, and capital reallocation speed reflect whether the enterprise makes better decisions faster and should be on the CFO's dashboard.