The first AI budget question:
who signs, and which line pays

In August, sign-off on AI purchases moved toward finance: CFO or finance was named on 43% of sign-off answers, up from 33%, while business-unit sign-off fell from 12% to 4% and proving ROI as the top blocker jumped from 53% to 65%. Funding clarity did not move at all — 26% still have no clear AI budget and 24% are paying for AI with money that would have gone to headcount. Our judgment: accountability and proof tightened first and the money never caught up, which is the wrong order; and a 28-point gap between the CEO seat and the finance seat on the same money says the missing asset is a measurement basis both seats accept.

1. The facts: sign-off moved, the money did not

Open Future Forum published Edition 3 of its CFO AI Leverage Report on 6 September 2026, with data collected through 31 August. In the August reading, the CFO or finance is named on 43% of AI sign-off answers against 33% in the cohort through July (bases 54 and 238), which the report calls the largest single move in the edition. The CEO remains the most-named signer at 50% in August and 47% on the full base of 290; business-unit sign-off fell from 12% to 4%, and no single owner yet fell from 12% to 7%. Source: openfutureforum.com/research/cfo-ai-leverage-report-september-2026.

The rise in the proof burden is steeper still: naming proving ROI as the main blocker climbed from 53% to 65% inside August, while sub-six-month payback expectations eased from 55% to 48% and six-to-twelve months rose from 22% to 28%. Funding structure did not follow: 26% of the August cohort has no clear AI budget (28% on the full base), and 24% fund AI with money that would have gone to headcount.

Our judgment: what moved in August was accountability and proof; what did not move was funding ownership. Sign-off is tightening and payback demands are rising while a quarter of organisations do not yet have a budget line — not weak execution, but an inverted sequence.

2. Four funding archetypes: decide which one you are

The edition adds a funding typology (base 290): Funders 37% (net-new money with a dedicated budget line), Unbudgeted 25% (no clear AI budget), Substituters 21% (funded with headcount money) and Reallocators 18% (moved from other software budgets). The companion AI Transformation Report, Edition 2, gives the finance-lane view: net-new money is still the largest source at 41%, 28% have no clear budget, about a fifth reallocate from other software, and the report states plainly that the money has not moved into the core budget. Source: openfutureforum.com/research/ai-transformation-report-september-2026.

The value of the cut is not the percentages but the fact that it turns the AI budget from a number into four different governance situations: organisations with a budget line are arguing about the amount, organisations without one are arguing about ownership, organisations using headcount money are arguing about the accounting, and organisations that reallocated are arguing about renewal. Four situations, four distinct failure modes.

Our judgment: classify yourself before debating the AI budget. Applying Funder metrics to a Substituter produces the wrong conclusion — that ROI is insufficient so spending should stop — when the real problem is that the money never had a budget line and the payback claim has nowhere to land.

3. A 28-point Optimism Gap: same money, two seats

The most quotable figures in the report are the seat split: 70% of CEO-seat respondents expect payback inside six months against 42% of the finance seat, 28 points apart (bases 80 and 52); the report states this is the only seat cut that clears its sample floor, with other cuts directional. A related line: 34% of CEO-seat respondents fund AI with money that would have gone to headcount, against 19% at the finance seat.

Our judgment: a 28-point gap is not optimism versus caution as a personality trait; it is one pool of money without a shared measurement basis. When payback is decided by the seat rather than by data, the budget review becomes a contest of voice — which is precisely where independently verifiable execution evidence earns its place.

4. Our judgment: the sequence is inverted

Read the August moves together: sign-off consolidated (business unit 12% to 4%), the security budget line grew from 33% to 41%, and proving ROI rose from 53% to 65% — three governance markers moved at once while budget clarity stood still. The report’s phrasing is that the accountability structure is being built after the deployment. Our reading is blunter: name the owner first, demand the proof second, and discuss where the money sits last.

That order has two costs. First, responsibility lands on an individual while funding has no owner, so the accountable person cannot defend the decision on the books. Second, the proof burden lands on the party least equipped to document it: business and engineering teams are asked to demonstrate payback, usually without verifiable execution evidence in hand.

Our judgment: the correct sequence is funding ownership first, measurement basis second, named owner third. Skip the middle step and the signer becomes a bearer of risk rather than a decision-maker.

5. Governance: agent access is now the top security problem

The AI Transformation Report, Edition 2, adds the governance reading: 67% of security leaders name agent access as their top problem; 37% have a dedicated AI security budget line, rising to 41% in August; about a third fund it case by case (base 110); and 11% of finance-lane respondents report no single AI owner against 30% at the technology seat.

Our judgment: naming agent access as the top problem shows the governance focus shifting from model capability to execution permission — the same shift that turned models and retention into per-model configuration items in the same week. Governance budgets are rising; what they should buy is verifiable access and execution records, not another policy document.

6. Conclusion and this month’s smallest move

Our judgment: the first AI budget question is no longer how much to spend, but who signs, which line pays and what proves the return. August’s answer is that sign-off moved toward finance, the money still has no line of its own, and the proof obligation has just been added. Get the order wrong and every step after it discounts.

Smallest move this month: classify your own funding archetype, then write down which line the money comes from, who signs for it and what basis proves the return. Then put the CEO seat’s payback expectation and the finance seat’s on one page as two numbers. The question to leave with: in your organisation, are the people who say six months and the people who say it is impossible using the same measurement basis?

References: Open Future Forum, CFO AI Leverage Report, Edition 3, published 6 September 2026 with data collected through 31 August 2026, curl HTTP 200; sign-off, funding archetypes, Optimism Gap and blocker figures are that report’s account (self-selected executive community, cohorts are different people rather than a panel; August cohort base 54, full base 290). openfutureforum.com/research/cfo-ai-leverage-report-september-2026 | Open Future Forum, AI Transformation Report, Edition 2, published 6 September 2026, curl HTTP 200; governance figures and the funding archetypes restated there (the report states it measures responses, not spend, headcount or performance, and that seat and vertical cuts are directional). openfutureforum.com/research/ai-transformation-report-september-2026.