PwC and OpenAI Launch AI-driven CFO Office Transformation
With 83% of finance chiefs planning higher AI spend, PwC and OpenAI are testing whether an AI-native finance model can finally deliver measurable returns.
PwC and OpenAI have expanded their collaboration to build what they are calling the first AI-native finance operation at enterprise scale, a move that targets the office of the CFO across the full range of finance activity, from planning and forecasting to procurement, treasury, tax, and the monthly accounting close. It arrives at an uncomfortable moment for the industry: CFOs are committing to AI at a pace that has no historical precedent, yet the returns are proving difficult to find. A Bain and Company survey of more than 100 finance chiefs found that 83% plan to raise AI spending by at least 15% over the next two years, with 42% targeting increases of 30% or more. PwC’s own 2026 Global CEO Survey found that 56% of chief executives report getting nothing measurable from those investments, and only 12% say the technology has delivered benefits to both costs and revenue. Where that money is going and what it is actually returning is the question PwC and OpenAI are now staking their reputations on answering.
The collaboration is notable less for its ambition than for its method. Rather than designing workflows in theory, PwC and OpenAI are building a procurement agent directly inside OpenAI’s own finance organisation, using that live production environment to test governance models, runtime controls, and the practical boundaries of human-agent collaboration. OpenAI describes its internal finance team as “customer zero” for the initiative, and the lessons from each internal deployment feed directly into the agents being built for enterprise clients.
The early results from OpenAI’s internal deployment are among the most concrete data points yet on what autonomous agents can deliver in a finance setting. Using Codex, OpenAI’s AI-powered coding assistant, the company’s finance team processed five times more contracts without adding headcount. A separate tool called IR-GPT managed more than 200 investor interactions during a recent fundraising round, a workflow that would ordinarily require significant analyst time, careful version control, and coordination across multiple stakeholders.
The scope of the agent buildout is deliberately broad. Rather than targeting a single workflow, the collaboration covers the full operating cycle of a finance function: procurement, contract review, accruals, the accounting close, forecasting, treasury, tax, and management reporting. Finance departments are unusual in that their workflows are deeply interdependent; a procurement agent that cannot communicate with the payments system or feed into the close process creates new bottlenecks as quickly as it eliminates old ones. By designing agents that operate across the entire cycle through shared enterprise connectors, PwC and OpenAI are attempting to avoid the partial-automation trap that has undermined earlier generations of finance technology.
Governance is built into the architecture from the start, not added afterwards. As agentic AI scales, CFOs will need clear visibility into AI usage, token consumption, and projected spend so they can manage adoption as an operating cost rather than an unmonitored overhead. PwC and OpenAI have embedded that requirement into the design of the initiative from the outset. The decision reflects a lesson the industry has learned from earlier automation cycles: finance tools that lack auditability and cost transparency tend to stall at the pilot stage as soon as boards and regulators begin asking questions that the technology was never designed to answer. Tyson Cornell, PwC’s US Advisory Leader, put the stakes plainly.
“Finance is at an inflection point, where organisations are moving from process efficiency to intelligent, decision-centric operations. Through our collaboration with OpenAI, we are helping clients embed agentic AI into the core fabric of the finance function, enabling more proactive insights, stronger controls, and a more adaptive operating model.”
Tyson Cornell, US Advisory Leader, PwC
PwC brings the finance transformation, controls, and implementation expertise necessary to move these workflows from prototype to production. OpenAI provides the models and products used to build and manage workflows across existing enterprise environments. The division of labour is deliberate. Enterprise finance is one of the most governance-sensitive environments in any large organisation; agents operating in procurement, treasury, and tax must follow approved internal processes, respect policy boundaries, and produce outputs that can withstand scrutiny from auditors and regulators. Neither company can do both halves of that convincingly without the other.
The collaboration positions PwC and OpenAI in direct competition with partnerships assembling quickly across the Big Four. Deloitte launched a dedicated agentic transformation practice with Google Cloud in April, built around Gemini Enterprise and spanning financial services, healthcare, retail, and government. KPMG has integrated Salesforce’s Agentforce platform firm-wide and launched its own multi-agent platform, KPMG Workbench, built on Microsoft Azure. Both arrangements pair consulting depth with cloud infrastructure. What differentiates the PwC model is that it is built on a direct relationship with the company that controls the underlying frontier AI, not a cloud layer sitting on top of it. Enterprise finance clients are not just buying implementation support; they are buying confidence that the agents they deploy will improve as the models improve, and that the firm advising them has a seat at the table where those models are built. For Sarah Friar, OpenAI’s chief financial officer and the executive whose team is running the initiative from within, the ceiling on what that means is considerably higher than cost reduction.
“Finance has always been about judgment, trust, and making decisions in environments filled with complexity and constant change. AI gives finance leaders a much deeper ability to see around corners and act faster. I believe we’re now entering a moment where the finance function itself gets reimagined to shape decisions in real time. The opportunity here is far bigger than efficiency. It is about giving finance leaders the tools to operate with greater foresight, agility, and strategic impact across the business.”
Sarah Friar, Chief Financial Officer, OpenAI
Under the operating model being developed, the role of finance professionals does not diminish. It changes. Rather than executing repeatable processes, finance teams will supervise, govern, and refine the AI agents performing that work. Human experts retain responsibility for judgment, compliance, and the handling of exceptions that fall outside the parameters agents can manage. That shift has significant implications for how CFOs hire and how they structure their organisations. The most valued skill in an AI-native finance function is not the ability to run a process, but the ability to govern the agent that runs it, to understand where it can be trusted, where it requires oversight, and where the stakes demand human decision-making from the outset.
PwC and OpenAI have structured this as a scaling exercise from day one, using OpenAI’s internal deployment not as a proof of concept but as a live production environment from which a repeatable enterprise framework can be extracted. The commercial logic is clear. If the model works, PwC gains a differentiated offering built on the world’s most recognised AI developer, and OpenAI gains a distribution channel into the finance functions of the Fortune 500 that no amount of direct sales could replicate at the same speed. What is less certain is whether the productivity gains visible inside a technology company’s own finance team will survive contact with the legacy ERP systems, fragmented data environments, and regulatory constraints that define finance operations at most large enterprises. That is the test this collaboration has yet to face. How it performs there will matter far more than how it performs at OpenAI.
