Executives ask a version of this question in every room: how do we compare? It is a fair question and it usually gets a bad answer, because most AI benchmarks report a single company-level figure that no single person inside the company would recognise.
What our September data shows is that the seat answering matters more than most benchmarks admit. Put the same question to the CEO chair, the finance chair and the technology chair and you get materially different pictures of the same phenomenon. That is the useful finding, and it is also the reason to be careful with what follows.
A note on what this is not. We do not collect dollar budgets, so there are no spend figures here. What we have is reported behaviour: who owns the decision, how long payback is given, where the money comes from, and what blocks the next commitment. That is a different and in some ways more useful benchmark, but it is not a spending comparison and should not be presented as one.
The comparison
Figures from the September 2026 editions of the Enterprise AI Buying and Budget Index. Bases in brackets. Dashes mean the instrument does not ask that seat that question.
| CEO seat (80) | Finance seat (52) | Technology seat (23) | Security lane (110) | Marketing lane (230) | Investors (245) | |
|---|---|---|---|---|---|---|
| Names itself as AI signer | 70% | 63% | 26% | – | – | – |
| Reports no single AI owner | 2% | 15% | 30% | – | – | 22% say too early to tell |
| Expects payback under six months | 70% | 42% | 39% | – | – | 35% |
| Names proving ROI as the blocker | 54% | – | 70% | – | – | – |
| Funds AI from headcount money | 34% | 19% | – | – | – | – |
| Holds a dedicated AI budget line | – | – | – | 37% | – | – |
| Past the exploration stage | – | – | – | – | 81% | – |
Two seat cuts deserve a warning label. The technology seat rests on 23 responses and the investor payback figure on 20. Both are below the 40-response floor we use for headline claims, and both are marked directional in the source reports. They are included because the direction is consistent across three separate instruments, not because either number is precise.
What the pattern actually says
Optimism falls with distance from the signature. The payback row is the cleanest gradient in the dataset. 70 percent at the CEO seat, 42 at finance, 39 at technology, 35 among investors. We track this as the Optimism Gap, currently 28 points between the CEO and finance chairs.
The tempting reading is that CEOs are naive. A more careful reading is that these seats are answering different questions. The CEO is describing the strategic case, which is genuinely fast if it works. Finance is describing what will appear on a statement, which is slower by construction. Neither is wrong. The problem is only that both numbers travel under the same label into the same board deck.
Visibility of the governance gap is inverse to seniority. The unowned-decision row runs 2, 15, 30 as you move away from the CEO chair. That is the Ownership Vacuum, and its shape has a practical consequence: the seat least likely to see missing ownership is the seat most likely to be asked about it by a board.
One chair does not claim the pen. Read down the signer row and the pattern breaks in one place. Technology puts its own title at 26 percent, behind both the CEO and the CFO in its own account of who decides. Follow that chair across the rest of the table and the picture is consistent: the most unowned decisions at 30 percent, the least confidence in fast payback of any operating chair at 39, and proving ROI named as the blocker by 70 percent, higher than any other seat. That is a chair which sees the work, doubts the timeline, and does not believe the decision is its own to make.
Funding source is a seat position too. 34 percent of the CEO seat funds AI with money that would have gone to headcount, against 19 percent at finance. The two chairs are describing the same budget. One is describing a strategic substitution, the other a hiring line that quietly did not get used.
How to use this
Find your row, then read the row above and the row below. The value is not in matching a number. It is in seeing what the chair on either side of you is likely to be reporting about the same decisions.
Three practical uses.
Before a board update. If the figure in your pack came from one chair, get the counter-seat number on the same page. A payback expectation from the CEO chair and a payback expectation from finance are 28 points apart, and a board that sees only the first has been shown the most optimistic reading available.
Before an approval. If your company's AI ownership answer comes from the top of the house, test it two levels down. The 2 percent and the 30 percent are answers to the same question.
Before a benchmark exercise. Ask any AI benchmark you are handed which seat answered it. If the answer is "executives", the number is a blend of chairs that disagree with each other, and the blend is not a company.
Limitations, stated plainly
This is application-stage data from executives entering Open Future Forum events between March and August 2026. The rooms are selective, approving roughly two applicants in five across the network and closer to one in four in the executive forums, and respondents self-select into an AI agenda. That makes this a benchmark of AI-engaged senior executives, not of companies generally, and almost certainly ahead of the average organisation.
Bases vary by an order of magnitude across the table, from 20 to 290. Multi-select questions mean shares total more than 100 percent. Seat cuts are based on the respondent's own chair, not their employer's structure, so a finance leader at a technology company appears in the finance row. And several rows compare figures drawn from different instruments answered by different rooms, which is why the table shows dashes rather than estimates where a question was not asked.
We do not have dollar spend. Anyone presenting these figures as budget benchmarks in the financial sense would be overstating what the instrument collects.
The full figures with their bases
Every number in this table, with its source report and response base, sits in Executive AI Statistics, September 2026. The cross-lane read that produced the seat comparisons is in the September Executive AI Leverage Report, and the full research library is here.
Last updated: September 8, 2026
Frequently Asked Questions
Browse the Research Library
Executive AI Statistics carries every published number from the September editions in citable form, each with its source report and response base.