Every Executive Claims to Sign Off on AI Purchases

← Back · ACT III · Chapter 17: VIGOR: Governance and Realization

For two years the question of who owns AI buying has been answered with opinion. This month it acquired an instrument. Open Future Forum, an executive community that runs first-party survey questions through its event network, published its August 2026 Enterprise AI Buying and Budget Index with a finding its own authors named the Self-Attribution Effect: 84 percent of chief executive respondents say the CEO signs AI purchases. Finance respondents say finance signs. Technology respondents put the CIO or CTO first, and every chair reports the pen on its own desk.

The instrument is candid about its limits. The chief executive cut rests on a base of 43; the finance and technology cuts are smaller, and the survey itself labels bases under 40 directional. The decimal points are not the finding; the pattern is, and it means the question “who owns this” has been returning confident answers that cannot all be true.

The Answer Depends On Who You Ask

Across the whole network the CEO is the most-named signer, at 44 percent of late-July answers on a base of 115. Investors, asked separately about their portfolios, name the CEO at 51 percent of 245 responses. The outside views agree with the corner office in direction and disagree in magnitude, and the survey draws the honest conclusion: the CEO’s lead is real, and it is smaller than the CEO believes.

Consider what this instrument actually measured. It asked the same question in room after room and received a different confident answer in each one. That result measures confidence rather than ownership, distributed evenly across seats that cannot all hold the pen.

Any organization can run the same experiment internally, and many now have the survey’s permission to expect the result: the signature will be claimed everywhere it is plausible to claim it.

The Vacuum Is Invisible From The Top

The companion figure is the one the survey names the Ownership Vacuum. Not one chief executive or founder respondent reports an unowned AI purchase decision at their company. Network-wide, “no single owner yet” doubled inside July, from 8 percent to 14 percent. And among technology respondents, the seat closest to the deployed work, roughly three in ten report that nobody owns the decision at all, on a small base the survey flags as directional. The September edition, on bases roughly twice the size, reports the same shape: the chief executive seat sees an unowned decision at 2 percent, the technology seat at 30.

Both readings live inside the same organizations: total attribution at the top of the chart, and a growing vacancy one level down where the work is. The places with the most confident claims are the places with the least written down.

I treat the ownership vacancy as a symptom rather than a thesis, and this is why. The vacancy and the over-claiming are the same condition observed from different chairs. Where nothing is written down, ownership is whatever the person asked believes it to be, and everyone asked believes something.

Assignment Does Not Settle It

The remedies now circulating all operate on the claim layer. Convene the seats and talk it through. Draw a decision-rights map and define escalation tiers for contested calls. Or take the survey’s own advice and name the owner and the metric in the same breath.

Each of these can end the argument about who may claim the signature. None of them makes the claim verifiable. An owner named in January is a fact about January. A signature with a number attached is still an assertion about the future, held by the seat with the strongest incentive to remember it generously. And the contest’s most common institutional product, a newly created seat to hold the contested thing, adds a claimant rather than an answer. Value does not need a new seat, and neither does the signature.

The survey’s advice to boards is two questions: who owns the AI purchase, and what number vindicates it. Both are the right questions. Neither can be answered by asking a seat to referee its own claim.

The Record Ends The Argument

Everywhere else in enterprise life, contested ownership is settled the same way: by an entry, not a recollection. Title runs with the deed. Spending answers to the ledger. Nobody surveys the executive team to establish who owns a building.

An AI purchase becomes owned, in that sense, when three things are written down by the buyer before the money moves. The definition of the value the purchase is for, declared by the organization that will live with it, because the buyer sets the definition. The evidence that will be admissible against that definition, fixed before the approval rather than reconstructed after it. And the entry made when the value arrives, or does not.

Against a record like that, the Self-Attribution Effect becomes harmless. The seats can believe what they like about the pen; the declaration says whose name is on it, and the vindication a board asks for is read, not recalled. Without such a record, the survey has shown what the alternative produces at scale: a signature claimed in every room and owned in none, over spending whose top blocker, by the same instrument, is still proving that the return exists. This is the custodian question arriving at the level of a single purchase.

Every seat has an answer. The buyer needs a record.

The same question now arrives for software acting on delegated authority. A federal bill, S. 5051, has begun specifying what a record of that authority has to contain, and most enterprises cannot answer it for their own agents.

Sources: Open Future Forum, CEO AI Leverage Report, Edition 2 (August 2026) and CFO AI Leverage Report, Edition 2 (August 2026), both at openfutureforum.com/research, opened and read 15 August 2026 and again 8 September 2026. The Executive AI Statistics page at the same address is refreshed monthly and now carries the September 2026 editions; the August figures cited here remain on the two Edition 2 report pages. Sign-off figures use bases 91 and 115; the seat-level cuts use bases 43 (chief executives and founders), 33 (finance) and 17 (technology), which the instrument itself labels directional below 40; the underlying dataset is archived with a permanent DOI. The names Self-Attribution Effect and Ownership Vacuum are the instrument’s own. The 51 percent proving-ROI blocker figure is from the same index’s CFO instrument.

Keep reading: ACT III · Chapter 17: VIGOR: Governance and Realization