The Missing Layer: Why the AI Economy Needs an ROI Operating System
Menlo Ventures reported at the end of last year that their data reported that enterprise companies spent $37 billion on generative AI in 2025, up from $11.5 billion in 2024, a 3.2x YoY increase. It is abundantly clear at this point that AI needs to be a P&L line item that is closely tracked and owned by CFOs.
A structural shift approaches in financial management of AI systems as AI agents take over all aspects of business operations. Autonomous agents, which are now becoming a fundamental part of organizations’ internal operations, operate continuously, trigger multi-step workflows, and can make thousands of LLM calls per session in order to complete tasks, bloating cross-org AI spend.
Organizations are noting these major changes and taking action: according to the FinOps Foundation, 98% of all organizations in 2026 actively manage AI spend, vs. 31% just two years ago, demonstrating that organizations are now centering AI spend as a key driver within their financials.
Yet, the conversation around AI financial observability largely stops at cost management. Gartner, for instance, has reported that an one-size-fits-all CFO AI approach proves to have limited value. Much less attention has been paid to how organizations can intelligently route AI spend to highest-ROI use cases while limiting spend in lowest-impact areas.
This lack of focus may be due to difficulty in measuring AI via existing tools available to CFOs today. CB Insights, in a survey of executive leaders, found that 40% of them cannot track or do not know their AI ROI, even though experts at organizations like Gartner have cited AI ROI predictability as a pre-condition for AI scale-up. We see an opportunity in AI revenue attribution that solves this fundamental problem. At Montauk Capital, we are actively exploring this space: if you are interested in joining us to build the ROI measurement layer for enterprise AI, reach out.
The cross-organizational win-win
Cost management alone has historically run into organizational friction. The tension between innovation and operational discipline is real: engineering teams want to experiment, while finance teams want predictability. Cost-cutting mandates often stall on the push and pull between advancement and governance, and cross-organizational buy-in is difficult to sustain.
ROI measurement changes this dynamic. The question shifts from “how much are we spending on AI?” to “what is each AI investment returning?”. Innovation teams gain the evidence to defend and scale their highest-performing use cases; finance teams gain the rigor to redirect spend from low-return workflows to high-return ones.
The historical parallel is instructive. In the early waves of enterprise SaaS adoption, organizations required ROI analysis of new software to both motivate adoption and prevent unnecessary spend. That discipline is what allowed software to graduate from a line item to a strategic investment. The same discipline is now required for AI. Outcome-level attribution is what allows AI to be deployed where it is needed most and retired where it is not, replacing gut feel with unit economics that CFOs, CTOs, and RevOps leaders can all act on.
What the market is already building
Today’s AI observability and cost management landscape is crowded with each layer producing strong builders solving meaningful problems. What we’re seeing, however, is limited focus on revenue management, with just a few players stepping into the AI revenue management space.
There is a clear opportunity to build the AI Economics OS that provides revenue attribution visibility as a core competency that can help close the loop between AI spend and business outcomes. The organizations that win the next phase of enterprise AI will be those that can answer, in real time, whether each workflow, agent, and model call is earning its keep. Building the infrastructure that makes that possible is one of the most compelling opportunities in enterprise AI today.
At Montauk Capital, we are actively exploring the intelligence and governance layer that makes this possible. Reach out if you are keen to compare notes.