CIOs have decided which enterprise software gets bought for most of the last two decades. That’s breaking down, and finance is where the break started, which surprised almost everyone except the CFOs themselves.
The CFO as gatekeeper
A Gartner survey of 303 CFOs and senior finance leaders, conducted last fall, found nearly 60% planning to raise AI investment in the finance function by 10% or more this year.Â
In the conversations I have with finance leaders, the more telling shift shows up in how many are now setting the vendor criteria and ROI thresholds that marketing, HR, and supply chain have to clear before an AI purchase gets approved. A few years ago, that review sat with IT procurement. Now it often starts, and sometimes ends, on the CFO’s desk.
What that review actually looks like varies by company, but the shape is becoming familiar: a standard intake form for any new AI tool, a required ROI estimate before a pilot gets budget, and a finance sign-off before a department’s card gets used for a new subscription.Â
None of that existed for AI purchases two years ago. It existed for capital equipment and licensed enterprise software, and finance is now applying that same discipline to a category of spending that used to move through expense reports and one-click trials. Departments that never used to loop in finance before adopting a new tool are doing it by default now, not because a policy told them to, but because the vendor’s sales team already knows finance has to sign off.
Governance is following the same trajectory. Deloitte’s Q2 2026 CFO Signals survey found 96% of finance chiefs confident in their organization’s AI governance framework, even as they flag real external risk: 43% point to potential litigation tied to the use of protected or private content, 41% cite cybersecurity, and 36% cite regulatory complexity. Someone has to own that risk conversation in terms a board can actually follow, and audit trails and accountability were already built into how finance operates long before AI showed up.
That transfer makes sense once you think about it. Finance already runs formal controls testing, segregation of duties, and audit trails for every material process under Sarbanes-Oxley. When an AI agent starts touching those same processes, the natural instinct is to treat it the same way: log what it did, who approved it, and why, so a human can trace the decision back if a regulator or an auditor asks. Most other departments don’t have that muscle built in. Finance has been doing a version of it for two decades, which is why CFOs ended up holding the pen on AI governance more than any strategy memo or org chart redesign could have given them.
Why finance moved first
None of this happened by accident. Finance is one of the only functions in a company where the value of a new system shows up as a number someone can check without an argument. An invoice matches automatically, or it doesn’t. The books close in nine days one month, four the next, and everyone downstream feels the difference. Compare that to a chatbot that makes a brand’s customer service feel warmer, or a copilot that helps someone draft a slide deck faster: real value, maybe, but nobody can point to the specific dollar it saved. Finance has the receipts to prove it.
And the math is not subtle. APQC benchmarking data, drawn from roughly 1,500 organizations, shows top-quartile accounts payable teams processing invoices for about $2.07 each, compared with $10 or more at the bottom quartile. The same split shows up in staffing: top-tier teams handle every $1 billion in revenue with about two people, while bottom-tier teams need five times the headcount for the same volume. That gap is a budget line with a payback period a CFO can defend to a board without hedging.
CFOs are becoming the de facto architects of enterprise AI adoption. Finance was the first place the numbers could be checked, and once you’re the one checking the numbers, you end up writing the rules everyone else has to follow.
The finance leaders getting the most out of this moment are treating it as a mandate rather than an accident. They’re building governance discipline before the next wave of AI spend lands. They’re asking vendors for outcome data instead of adoption metrics, and they’re willing to let some low-stakes bets fail fast instead of subjecting every pilot to a business case. That combination, rigor where it matters and speed everywhere else, is quietly turning the CFO’s office into the place where enterprise AI either earns its keep or gets sent back to pilot.
Author’s Biography: Rob Bearden is co-founder and CEO of Sema4.ai. He was co-founder and CEO of Hortonworks, a publicly traded open-source company that merged with Cloudera in 2019. He was then CEO of Docker in 2019 and remains on the board. Rob returned to Cloudera in late 2019 to serve as CEO where he led the restructuring and sale to private equity firms KKR and CDR for $5.3B. Previously, he served as President and COO of SpringSource, a leading provider of open-source developer tools, until its acquisition by VMWare in 2009. Prior to joining SpringSource, Rob served as Entrepreneur in Residence at Benchmark Capital. He also served as President and COO of JBoss, a leading open-source middleware company, until its acquisition by Red Hat in 2006
