The money does not match the measurement
$1.76 trillion went to AI in 2025. $826 million of it went to AI data, which is the most-cited barrier in every survey in this library.
The short answer
What does the research show about The money does not match the measurement?
Evidence
- IBM Institute for Business Value: generative AI pilot ROI was 31% in 2023 and 7% as those pilots scaled, against a typical cost-of-capital hurdle of about 10%. Sample size and field dates are not stated on that page.
- McKinsey: 93% of respondents exceeded AI budgets. Refinement work of checking, repairing and reverifying is about 60% of an agentic task's cost, and agentic tasks consume roughly 1,000 times more tokens than chat or code reasoning tasks.
- CloudZero: only 51% strongly agree they can track AI ROI effectively against 91% claiming overall confidence, and 15% have no formal cost tracking at all.
Supporting pages
- Do Agentforce Token Counts Equal Flex Credit Bills? analysis
- When Does Agentforce Help Agent Bill 400 Flex Credits for a Resolution? analysis
- Dictionary plain-language definitions
Forecast AI spending is up 47% year over year in 2026 to $2.59 trillion. The share of finance leaders who can tie AI spend to a business outcome is 22%.
The most damaging line in the spend table is the smallest one. AI data was $826M in 2025, which is 0.05% of total AI spending, in the same period that data quality was the most-cited barrier across every survey collected here.
What this page is
Total AI spending against the share of finance leaders who can tie any of it to a business outcome, plus where inside the spend the money is not going.
The argument
Spending is compounding faster than the ability to attribute it, and the least-funded line item is the one every survey names as the top barrier.
How to read it
- Total spend figures include infrastructure and semiconductors. They are not enterprise software budgets and should not be compared to one.
- Attribution shares are self-assessments by finance leaders, which is the closest available proxy for whether a CFO would defend the number.
- Inference price declines are real and mostly irrelevant to the attribution problem. Cheap tokens do not produce an attributable outcome.
$1.76 trillion went to AI in 2025. $826 million of it went to AI data.
Worldwide AI spending forecast, 2025, in millions of US dollars.
2025 spend, $M
- AI infrastructure975,581
- AI services436,351
- AI software282,897
- AI cybersecurity25,920
- AI platforms for data science and ML21,292
- AI models15,494
- AI application development platforms6,587
- AI data826
0.05% of total AI spending
What this does not say
These are forecast totals, not booked spend. A forecast revision is not a market movement.
- Publisher
- Gartner Forecast: AI Spending Worldwide
- Sample and method
- 1Q26 forecast, retrieved via Business Wire on May 19, 2026 because the Gartner newsroom page was not reachable. These are forecasts, not actuals
- Field dates
- Not published by the source.
Worldwide AI spending forecast, 2025, in millions of US dollars.
| 2025 spend, $M | Value | Note |
|---|---|---|
| AI infrastructure | 975581 | |
| AI services | 436351 | |
| AI software | 282897 | |
| AI cybersecurity | 25920 | |
| AI platforms for data science and ML | 21292 | |
| AI models | 15494 | |
| AI application development platforms | 6587 | |
| AI data | 826 | 0.05% of total AI spending |
Source: Gartner Forecast: AI Spending Worldwide. 1Q26 forecast, retrieved via Business Wire on May 19, 2026 because the Gartner newsroom page was not reachable. These are forecasts, not actuals Confidence: Medium.
What this does not say: These are forecast totals, not booked spend. A forecast revision is not a market movement.
The spend is forecast to 47% growth. The attribution is at 22%.
Two different populations, placed side by side for scale, not compared statistically.
22 of 100 finance leaders can tie AI spend to a business outcome, against $2.59 trillion in forecast 2026 AI spend.
What this does not say
An attribution gap is not evidence that the spend was wasted. It is evidence that nobody measured it.
- Publisher
- CloudZero and Gartner via Business Wire · Vendor research
- Sample and method
- CloudZero: n=260 finance executives including 135 CFOs. 78% report a gap, 54% call it serious, 87% must close it within the year
- Field dates
- Not published by the source.
Also in the record
Figures that sit alongside these charts.
- IBM Institute for Business Value: generative AI pilot ROI was 31% in 2023 and 7% as those pilots scaled, against a typical cost-of-capital hurdle of about 10%. Sample size and field dates are not stated on that page.
- McKinsey: 93% of respondents exceeded AI budgets. Refinement work of checking, repairing and reverifying is about 60% of an agentic task's cost, and agentic tasks consume roughly 1,000 times more tokens than chat or code reasoning tasks.
- CloudZero: only 51% strongly agree they can track AI ROI effectively against 91% claiming overall confidence, and 15% have no formal cost tracking at all.
- Epoch AI: inference prices declined at a median of roughly 50x per year across six benchmark thresholds. Price is no longer the constraint. Attribution is.
- Forrester reports 25% of planned AI spend deferred to 2027 while Gartner forecasts 47% growth in the same year. That contradiction is left standing.
The brief
What is going on here, and why it matters.
The charts above are the evidence. This is the read: what the data describes, the mechanism behind it, where the argument could be wrong, and what a revenue team does about it.
The number that matters is the small one
Total AI spending reached $1.76 trillion in 2025, and $826 million of that went to AI data. That is 0.05 percent of the total, in the same period that data quality was the most-cited barrier across every survey collected in this library. The market bought the engine and skipped the fuel.
The consequence shows up downstream in every other theme. Informatica's 67 percent who cannot move half their pilots to production, RAND's persistent data-quality problems raised by 30 of 50 industry interviewees, and Clari's 48 percent who say their data is not AI-ready are all the same line item, priced at five hundredths of one percent.
Attribution is not improving with scale
Forecast spend is up 47 percent year over year in 2026 to $2.59 trillion. The share of finance leaders who can tie AI spend to a business outcome is 22 percent. IBM found generative AI pilot ROI of 31 percent in 2023 falling to 7 percent as those pilots scaled, against a typical cost-of-capital hurdle of about 10 percent. Scaling made the economics worse, not better.
McKinsey names the mechanism: 93 percent of respondents exceeded AI budgets, refinement work of checking, repairing and reverifying is roughly 60 percent of an agentic task's cost, and agentic tasks consume roughly 1,000 times more tokens than chat or code reasoning tasks. Verification cost scales with autonomy, and almost no business case includes it.
Why cheaper tokens do not fix this
Epoch AI measured inference prices declining at a median of roughly 50 times per year across six benchmark thresholds. If unit price were the constraint, the attribution problem would already be solved. It is not, because the binding costs are verification labor, data remediation, and process change, none of which follow a semiconductor curve.
CloudZero found only 51 percent strongly agree they can track AI ROI effectively against 91 percent claiming overall confidence, and 15 percent have no formal cost tracking at all. Confidence is running 40 points ahead of instrumentation.
The correction
Move a visible share of the AI budget to data remediation and to the verification layer, and report it as AI spend rather than hiding it in platform maintenance. That single accounting change makes the real cost of an agent legible before the second year of the contract.
Then pick one workflow and attribute it properly, end to end, including the refinement cost. One defensible attribution is worth more in a board conversation than a portfolio of unmeasured pilots.
What to do with it
The move, by seat.
- CFO
- Require the verification and data cost line in every AI business case. It is roughly 60 percent of an agentic task's cost and it is routinely omitted.
- CRO
- Fund one fully attributed workflow instead of a portfolio. Attribution, not spend, is the constraint on the next budget.
- RevOps
- Stand up cost tracking per use case now. Fifteen percent of organizations have none at all and cannot answer the first CFO question.
Questions this page answers
What the data says, in plain language.
- What does the research show about The money does not match the measurement?
- $1.76 trillion went to AI in 2025. $826 million of it went to AI data, which is the most-cited barrier in every survey in this library. Forecast AI spending is up 47% year over year in 2026 to $2.59 trillion. The share of finance leaders who can tie AI spend to a business outcome is 22%.
- What does the figure "$1.76 trillion went to AI in 2025. $826 million of it went to AI data" show?
- Worldwide AI spending forecast, 2025, in millions of US dollars. Source: Gartner Forecast: AI Spending Worldwide. 1Q26 forecast, retrieved via Business Wire on May 19, 2026 because the Gartner newsroom page was not reachable. These are forecasts, not actuals Confidence: Medium.
- What does the figure "The spend is forecast to 47% growth. The attribution is at 22%" show?
- Two different populations, placed side by side for scale, not compared statistically. Source: CloudZero and Gartner via Business Wire. CloudZero: n=260 finance executives including 135 CFOs. 78% report a gap, 54% call it serious, 87% must close it within the year Confidence: Medium. This figure comes from vendor research, so read it as a vendor claim.
- What else sits alongside these figures?
- IBM Institute for Business Value: generative AI pilot ROI was 31% in 2023 and 7% as those pilots scaled, against a typical cost-of-capital hurdle of about 10%. Sample size and field dates are not stated on that page. McKinsey: 93% of respondents exceeded AI budgets. Refinement work of checking, repairing and reverifying is about 60% of an agentic task's cost, and agentic tasks consume roughly 1,000 times more tokens than chat or code reasoning tasks. CloudZero: only 51% strongly agree they can track AI ROI effectively against 91% claiming overall confidence, and 15% have no formal cost tracking at all. Epoch AI: inference prices declined at a median of roughly 50x per year across six benchmark thresholds. Price is no longer the constraint. Attribution is.
- Where does this data come from?
- Every figure is reproduced from a named publisher: Gartner worldwide AI spending forecast, CloudZero, Finance needs AI ROI 2026 survey, IBM Institute for Business Value, agentic AI and profit, McKinsey, agentic economics and the modern operating model. Sample, field date, and confidence are shown on each chart. Sources marked as vendor research are labelled on the page.
Cite this page
Permanent URL and suggested citation.
https://www.therevenueaireport.com/research/spend-vs-attribution
Kvarfordt, Jonathan. "The money does not match the measurement." The Revenue AI Report, Research Library. https://www.therevenueaireport.com/research/spend-vs-attribution
Figures on this page are reproduced from the publishers listed below. Cite the original publisher for the underlying data, and this page for the compilation and framing.
Sources
Every publisher used on this page.
If a metric, model term, or method on this page is unfamiliar, every one of them is defined in The AI and Revenue Dictionary. Sample size, field date, and confidence tags are explained there too.
Gartner worldwide AI spending forecast
Retrieved via Business Wire, May 19, 2026
https://www.businesswire.com/news/home/20260519405832/en/Gartner-Forecasts-Worldwide-AI-Spending-to-Grow-47-in-2026CloudZero, Finance needs AI ROI 2026 surveyVendor research
n=260 finance executives including 135 CFOs
https://www.cloudzero.com/finance-needs-ai-roi-2026-survey-report/IBM Institute for Business Value, agentic AI and profit
Sample size and field dates not stated
https://www.ibm.com/thought-leadership/institute-business-value/report/agentic-ai-profitsMcKinsey, agentic economics and the modern operating model
Token and refinement cost analysis
https://www.mckinsey.com/capabilities/quantumblack/our-insights/is-that-ai-agent-worth-it-agentic-economics-and-the-modern-operating-model
Could not confirm
What we looked for and did not find.
Claims found during research and not charted
Nothing on this theme was dropped for sourcing. Every claim we found that met the standards on the Research hub is charted above, and anything that failed them would be listed here by name.
Read the analysis
Issues built on this theme.
Research on this site is the evidence layer. These essays take the numbers above and apply them to real decisions, so you can see how the data reads in practice.
- Do Agentforce Token Counts Equal Flex Credit Bills?
Salesforce's own August 18, 2026 post says the AI Usage Data Model provides token counts, not Flex Credits directly, and that Digital Wallet is the single source of truth for exact consumption, billing verification, and contractual overage calculations. Use the data model for near-real-time dashboards. Use Digital Wallet for the invoice.
- When Does Agentforce Help Agent Bill 400 Flex Credits for a Resolution?
A billed Help Agent resolution costs 400 Flex Credits at Salesforce list ($2). Escalation, negative feedback, or walk-away is not billed. Sandbox resolutions are 0 on the rate card. Data 360 and Agentforce actions stay unmetered during the Help Agent session.
- Do Agentforce and Data 360 Share One Flex Credit Pool With No Upfront Allocation?
Flex Credits are fungible across Agentforce and Data 360. Salesforce says you do not allocate upfront. Data work can burn the same pool your Agentforce forecast assumed was yours.
How to cite this research
Written by Jonathan Kvarfordt, Founder and Principal Analyst, The Revenue AI Report. Published under CC BY 4.0.
APA
Kvarfordt, J. (2026). The money does not match the measurement. The Revenue AI Report. Retrieved from https://www.therevenueaireport.com/research/spend-vs-attribution
MLA
Kvarfordt, Jonathan. "The money does not match the measurement." The Revenue AI Report, 31 Aug. 2026, www.therevenueaireport.com/research/spend-vs-attribution.
BibTeX
@misc{kvarfordt2026spendvsattribution,
author = {Kvarfordt, Jonathan},
title = {The money does not match the measurement},
year = {2026},
publisher = {The Revenue AI Report},
url = {https://www.therevenueaireport.com/research/spend-vs-attribution}
}Next theme
Trust went down as capability went upSeven trust statements, asked twice ten months apart. All seven fell. Expected autonomy is far lower than the market assumes.
Subscribe