Rick Caccia, CEO of Witness.ai.gettyA recent survey of more than 1,000 C-suite executives found that when CFOs are directly accountable for AI value, 76% of organizations achieve a high degree of success. Only 2% of CFOs actually hold that role. The obvious response is to give more CFOs that accountability. The real question is whether they’d have the information they need to act on it.​I've spent the past few years talking with CIOs and CFOs across industries and have learned that CFOs get access to the AI bills after the fact. They can see dollars spent but not value gained. They don’t know which teams drove the spend, what it was used for and whether any of it produced a result they could point to.​The Three Blind C-Level Execs​The standard accountability structure assigns AI responsibility to three roles: the CISO owns risk, the CFO owns cost and whoever holds the AI outcomes job—chief AI officer, CIO or COO—owns productivity and effectiveness. None of them has the tools to see their own piece of the picture clearly.​They are three blind C-level executives trying to find something they don't have the tools to see. They should be collaborating on a unified picture, but the information to do that doesn't exist yet.​When I ask CISOs to quantify risk in terms the board understands, the answer is often some version of the same thing: “I can't tell you the risk reduction. I can tell you the technologies I put in place.” The CFO and CEO keep sending questions to the CISO as to who's using AI, what they are using it for and where the spend is going, because they don't know where else to go. It’s a language problem as much as anything; CISOs talk technology, boards want risk.​I presented to a room of 20 CIOs recently. They know they're blind, and they're worried about it. And when organizations can't see what AI is being used for, you end up spending millions of dollars in token costs to get people making cat videos and planning vacations. With agents, it’s easy to spend thousands on tokens in just a couple of days of coding.​What Most CFOs Are Missing​Based on conversations I've had with finance and engineering leaders, the gap between what AI costs and what it produces is consistently misread. A CFO who sees a large AI token bill and cuts it may have just canceled the most productive work their engineering team has done. The same output, built by human engineers, would have cost orders of magnitude more and taken months longer. The bill looked alarming because the output was invisible. Most cost frameworks assume a fixed relationship between what you spend and what you get. AI breaks that, and most organizations haven't adjusted for it.​I saw this play out with a C-level team at one of the largest mutual fund companies in the country. We walked them through a framework covering risk, cost and business value. They were confident going in. When we started asking specific questions, that confidence evaporated. They had no visibility into business value or cost attribution. On risk, they could see only the security layer. They estimated they had roughly 5% visibility into what was actually happening across their AI environment. They were stunned.​Agents Make The Problem More Expensive​Agentic AI is dramatically more powerful, more risky and more expensive than the chat tools that preceded it. Yet most enterprises budget for both the same way.​Nobody decided to start using agents. They just showed up inside tools organizations already owned. If your organization had a Claude or Copilot subscription last year, developers may already be building and running agents inside it. Fifty ServiceNow agents running on Tuesday that weren't there on Monday. No security review. No vendor assessment. A chat tool generates a predictable bill. An agent running autonomously can generate costs no one in finance saw coming, before anyone knows it's running. ​What Needs To Change​Finance needs detailed insight into actual spend, by team, project, tool and use type. Organizations need the ability to act on what they see, distinguishing productive use from personal use and high-return investment from spend that produces nothing.​Agentic AI is a different budget problem from a copilot drafting emails. The cost and risk profile is categorically different. Treating them as one line item means making decisions that are wrong for both. ​Splitting cost visibility and outcome accountability across three executives means the full picture belongs to none of them. Put both in a single role, with someone who can connect AI spend to AI output and make the case to the board. ​The companies that get this right now will come out ahead. Every quarter without the right information infrastructure, the gap widens. The AI ROI problem is solvable. Start with the information problem first.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?