Companies Track AI Spending but Not the Value It Creates

← Back · ACT II · Chapter 14: Who Controls Value Engagement?

On July 29, Accenture published a perspective whose title does most of the work: “AI is on your P&L. Most companies are only reading half of it.” Its argument is that AI spend lands on the cost line while the value it produces, in retention, freed capacity and working capital, mostly goes uncaptured and untracked. The report cites Goldman Sachs putting AI-related spending above $800 billion in 2026, and Accenture’s own executive survey finds only 23 percent of C-suite leaders reporting widespread and sustained business value from AI across their organization.

The report is worth a close read for what it confirms. A firm of Accenture’s size, drawing on its own client base, describes the condition this standard is about: spend that everyone can see, value that several functions claim and none of them owns, and a C-suite reading different pages of the same ledger. Its proposed fix is a different matter, and the reasons are below.

The Bill Arrives Without a Name

Fortune’s CFO Daily ran a companion piece the same morning. Lan Guan, Accenture’s Chief AI and Data Officer and one of the report’s authors, told Fortune that token cost is “a CFO conversation that I’m having constantly.” Finance chiefs at large companies often cannot say where the costs come from; they just get the bill. The questions the article lists are basic ones: who is using tokens, who is building agents, and which product is driving the consumption. Her clients, Guan said, are “walking in the dark.”

The report puts numbers under that. In nearly every deployment Accenture has examined, its own included, fewer than 10 percent of users and workflows drive the vast majority of the bill. It also offers a sentence any finance team will recognize: “A tool enabled without a defined job is a cost with no owner.”

Usage-based pricing changed how fast that shows up. With traditional software, a tool nobody should have bought could sit in the run rate for years, because licenses were fixed and seats were counted once, until a write-off finally collected it. AI is metered, so every prompt, agent handoff and retrieval is a priced transaction. A misalignment that used to surface in an annual impairment review now turns up in the monthly invoice, itemized, on the desk of a CFO who cannot say whose it is.

What the Report Says Causes It

When Accenture asks why companies cannot link AI spend to business value, 57 percent cite weak business-technology alignment and 50 percent say they struggle to isolate AI’s contribution from other factors. The report describes the mechanism plainly. AI costs land in the technology budget while value shows up in the business units, and “the CFO sees cost, the CIO sees usage and business leaders see outcomes, but those views rarely connect at the enterprise level.”

That is an accurate drawing of the alignment gap. The 50 percent figure deserves the most attention, because it describes the condition that lets claimed value run ahead of realized value for years. If nobody can separate one contribution from another, every function’s number survives review, and the total of the benefits being credited can grow well past anything the financial statements show.

The fix the report proposes is to ask the CFO and CIO to jointly own one enterprise view of AI cost, usage and return, supported by shared metrics, governance forums and decision rights. That is the usual response when a problem sits on the border between two functions: name joint owners and hold a regular meeting. It produces agreement about the numbers, which helps. It does not change where the costs land or who answers for the return, and a forum that meets on a calendar is reviewing a meter that never stops running.

What a Structural Answer Looks Like

Alignment, in the sense this standard uses, is a property of how the operating model is built: incentives and accountability tied to realized value in a way that keeps working when the calendar slips or the org chart changes. A useful test for any fix here is whether it would survive those two things.

Applied to a metered bill, that means consumption is attributed to the work that caused it at the moment it happens, because the systems record it that way. Each claimed return traces back, when the claim is made, to the budget line that paid for it, rather than being pieced together at quarter end. Spend that cannot show what it contributes ends by default instead of running on until someone notices. None of that depends on a named person remembering to ask.

The report gets close to this once. Its most advanced companies, it says, “embed token consumption into how work is scoped and priced, so AI spend sits with the value it creates rather than pooling in a central technology line.” That is the strongest sentence in the document, and it puts the cost where the value is, which a forum cannot do on its own.

A vendor whose tool does a defined job is better off under that arrangement too. Its consumption shows up next to the result it paid for, instead of sitting in a central pool that a CFO under pressure is looking to cut.

Where nobody inside owns the number, the seller’s instrumentation tends to fill the gap, and the vendor computes the return the buyer never set up to compute. A week after this report, Accenture was among the founding supporters of the Linux Foundation’s new Tokenomics Foundation, which Fortune described as focused on standardizing how companies measure and manage AI token costs and connect them to business value. Its membership brings together companies that use tokens and companies that sell them, which is why where that standard starts matters to a buyer.

References: Accenture, “AI is on your P&L. Most companies are only reading half of it” (AI Tokenomics for Enterprise Value), July 29, 2026, the source of the Goldman Sachs spending figure and of the 23, 57 and 50 percent figures from Accenture’s Pulse of Change survey; Sheryl Estrada, “CFOs are hitting a ‘cost wall’ on AI”, Fortune, July 29, 2026; Sheryl Estrada, Fortune, August 4, 2026, on the launch of the Tokenomics Foundation.

Keep reading: ACT II · Chapter 14: Who Controls Value Engagement?