Enterprise finance accounts for 75 companies and $9.1 billion in combined funding in the AIFI directory, making it the single largest segment we track. That concentration of capital tells you something about where buyers are spending; CFOs aren't chasing chatbots, they're writing checks for software that closes their books faster and catches errors before auditors do. I've spent the last six months mapping what I call the "AI CFO stack," and the picture that emerges is less about any single breakthrough and more about a steady, methodical layering of machine learning into workflows that have run on spreadsheets and manual approvals for decades.
The stack has clear tiers. At the bottom sits accounts payable automation. Above that, expense management and corporate cards. Then treasury and cash management. Then FP&A, the planning and forecasting layer. And finally, payroll and HR finance, which touches every employee in the organization. Each tier has its own dynamics, its own incumbents, and its own version of the question every CFO is asking: where does ML actually reduce errors, and where is it just a marketing label?
Accounts payable: the $3 billion entry point
AP automation is where most CFOs first encounter AI in their own department. The mechanics are straightforward: invoices arrive in varied formats (PDF, email, EDI, sometimes still paper), need to be captured, matched against purchase orders, coded to the correct GL account, routed for approval, and paid. Every step is a candidate for automation, and the market for AP automation software sits around $3 billion today with double-digit annual growth.
Stampli has built its product around the idea that invoice approval is a collaborative process, not a linear one. Their AI learns each organization's approval patterns over time, routing invoices to the right people based on vendor, amount, department, and historical behavior. The system flags exceptions rather than trying to handle everything autonomously, which matters in a finance context where a misrouted $500,000 invoice isn't just inconvenient; it's a control failure. Stampli's approach reflects a broader pattern I see across the stack: the most successful products augment human judgment rather than replacing it.
Tipalti takes a different angle, focusing on global payables. They automate payment execution across 196 countries, handling currency conversion, tax compliance, and fraud detection for over 4,000 companies. Their ML models score payment risk in real time and flag transactions that match known fraud patterns. For companies with large supplier networks spanning multiple countries, the value proposition isn't intelligence per se, it's the elimination of manual work that scales linearly with supplier count.
Expense management gets competitive
The corporate card and expense management space has become one of the most competitive corners of fintech. Ramp has crossed $300 million in ARR and is the fastest-growing corporate card in the US by most measures. Brex, which started as a card for startups, has shifted toward mid-market expense management. Both companies use ML in similar ways: real-time spend categorization, automated policy enforcement, and anomaly detection that flags unusual charges before they become audit findings.
Ramp's AI assistant is a good example of where LLMs are starting to show up in enterprise finance. It can auto-draft expense reports from receipt images, flag duplicate charges across employees, and surface spending patterns that a human reviewer would miss in a stack of line items. The underlying models aren't exotic. They're classification and matching algorithms trained on millions of transactions. But the product experience is genuinely different from the old workflow of filling out expense forms manually, and adoption rates reflect that.
What's interesting about this tier is the speed of commoditization. Real-time categorization and duplicate detection were differentiators two years ago. Now they're table stakes. The competition has shifted to integration depth, how well these platforms connect to ERP systems, banking partners, and the rest of the CFO stack. That's a pattern worth watching across every tier.
Treasury and cash management
Treasury sits at the center of the CFO stack, and HighRadius has become the dominant AI-native player. Valued at over $1 billion with more than 800 enterprise customers, HighRadius automates the order-to-cash cycle and treasury management. Their ML-based cash forecasting models ingest historical cash flows, receivables aging, payment patterns, and macroeconomic indicators to produce forecasts that, according to their published case studies, reduce forecast variance by 30-40% compared to spreadsheet methods.
The cash forecasting use case is a clean fit for ML because it involves structured tabular data, clear feedback loops (you can compare the forecast to actual cash positions), and high business value. A 10% improvement in cash forecast accuracy directly reduces the amount of precautionary cash a company needs to hold, freeing working capital. For a large enterprise, that can mean tens of millions of dollars.
Kyriba operates in the same space with a SaaS treasury management platform used by large corporates and banks. Their product covers cash management, payments, risk management, and supply chain finance. The ML applications are more targeted: anomaly detection in bank statements, cash flow pattern recognition, and automated bank reconciliation. Treasury management is conservative by nature; treasurers are paid to avoid surprises, not to chase upside. That conservatism shapes how AI gets adopted here. Incremental accuracy gains matter. Flashy demos don't.
FP&A: the shift from Excel
Financial planning and analysis might be where the gap between current practice and ML potential is widest. Most FP&A teams still build their forecasting models in Excel, with manual data pulls, hardcoded assumptions, and version control that consists of filenames like "Budget_v14_FINAL_revised_ACTUAL.xlsx." The shift to ML-enhanced forecasting platforms is real but slow, mostly because FP&A analysts have deep expertise in their company's business drivers and are understandably skeptical of black-box models.
Pigment and Vareto represent the new generation. Pigment offers a business planning platform where revenue forecasts, headcount plans, and scenario analyses are built on connected data models rather than disconnected spreadsheets. Their ML features focus on scenario generation (automatically surfacing the variables that most affect outcomes) and anomaly detection (flagging when actuals deviate from plan in unexpected ways). Vareto takes a similar approach with a focus on real-time variance analysis and collaborative planning.
The real challenge in FP&A isn't building a better model. It's getting the data into a state where any model can work. Most companies have financial data scattered across ERP systems, CRM platforms, HRIS tools, and dozens of spreadsheets maintained by individual analysts. The companies winning in this tier are the ones solving the data integration problem first and layering ML on top second.
Payroll and the platform play
Rippling sits at an interesting intersection. It combines HR, IT, and finance in a single platform, running over $12 billion in annual payroll. Their approach is to treat employee data as a unified graph: when someone is hired, their laptop is provisioned, their benefits are enrolled, their payroll is set up, and their expense card is issued, all from one system. The AI applications span the full surface area, from automated tax compliance (calculating withholdings across jurisdictions) to anomaly detection in payroll runs to intelligent routing of approval workflows.
Rippling's bet is that the CFO stack shouldn't be a stack at all. It should be a platform where finance, HR, and IT share a common data layer. If that thesis plays out, the standalone point solutions in each tier face pressure from platform players that can offer "good enough" functionality with better integration. It's too early to say whether that will happen, but the trajectory of Rippling's growth suggests the market is receptive.
What CFOs actually want
I've spoken with several CFOs over the past quarter about their AI priorities, and the consensus is remarkably consistent. They want accuracy, not intelligence. Specifically, they want AI that reduces errors in reconciliation, speeds up the monthly and quarterly close process, and produces audit trails that satisfy SOX compliance requirements.
SOX is the quiet constraint that shapes everything in this market. Section 404 requires that every material financial process has documented internal controls, and those controls need to be testable by external auditors. When an AI system makes a decision, whether it's approving an invoice, categorizing a transaction, or flagging an anomaly, that decision needs to be explainable and auditable. This requirement eliminates a lot of the more aggressive ML architectures and favors simpler, more interpretable models. Gradient boosted trees and logistic regression outperform neural networks in this context not because they're more accurate, but because auditors can understand them.
The other consistent theme is integration. CFOs don't want to buy six AI point solutions that each require their own implementation, training, and maintenance. They want their existing ERP system (SAP, Oracle, NetSuite) to get smarter, or they want a platform that replaces multiple tools at once. The vendors winning deals are the ones that can demonstrate value within existing workflows rather than requiring process redesign.
Where the stack goes from here
I expect the next two years to bring consolidation. The AP automation tier is already seeing M&A activity, and the expense management space has more funded competitors than the market can sustain. The winners will be companies that have built proprietary data advantages, either through network effects (more transactions make the models better) or through deep integration with enterprise systems that creates switching costs.
The FP&A tier has the most room for growth, but it also faces the highest adoption barriers. Convincing a CFO to move forecasting out of Excel is a cultural change as much as a technology sale. The companies that crack this will likely do it by augmenting Excel rather than replacing it, meeting finance teams where they already work. As for the broader question of whether ML is reshaping enterprise finance, the answer is yes, but it's happening through thousands of small accuracy improvements and workflow automations rather than through any single dramatic shift. That's not the story most vendors want to tell. But it's the one the data supports.