The Revenue AI Report

The same question, four different answers

Enterprise AI is paying off for 5% of companies, or 74%, depending on the question asked. None of the four publishers is wrong.

The short answer

What does the research show about The same question, four different answers?

Enterprise AI is paying off for 5% of companies, or 74%, depending on the question asked. None of the four publishers is wrong. Four credible publishers measured something adjacent to is enterprise AI paying off and produced answers that cannot all be right. They are not measuring the same thing.

Evidence

  • Before you quote an AI ROI number, name the population, the question wording, and the field dates. Three numbers in this chart are defensible and none of them is interchangeable with another.
  • Fortune's widely republished description of the MIT methodology does not match the report's own description of it.

Supporting pages

Four credible publishers measured something adjacent to is enterprise AI paying off and produced answers that cannot all be right. They are not measuring the same thing.

MIT looked at pilots and required measurable P&L impact. Wharton and Deloitte asked leaders about their most advanced initiative. McKinsey required attribution to EBIT. Each answer follows from its question, which is exactly why a single quoted number in a board deck is close to meaningless without its second line.

What this page is

Four credible publishers answering a version of is enterprise AI paying off, with answers spanning 5 percent to 74 percent, and the reason each answer follows from its question.

The argument

There is no single AI ROI number. The spread between defensible answers is 69 points and it is produced entirely by population, question wording, and success definition.

How to read it

  • None of these four publishers is wrong. Each answer is correct for the question asked.
  • The numbers are not interchangeable and averaging them produces a figure that measures nothing.
  • A quoted ROI number without its population, wording, and field dates is not evidence.

Enterprise AI is paying off for 5% of companies, or 74%, depending on the question asked.

None of these four publishers is wrong. They asked four different questions.

  • MIT NANDA, The GenAI Divide

    Share of pilots with measurable P&L impact. 52 organizations interviewed, 153 leaders surveyed, more than 300 public initiatives reviewed. Preliminary, not peer-reviewed

    5% seeing return

    Low confidence

  • McKinsey, State of AI 2026

    Share attributing any EBIT impact to AI. n=1,719, fielded May to June 2026. 6% attribute 5% or more of EBIT

    37%

    High confidence

  • Deloitte, State of Generative AI Q4 wave

    Share whose most advanced initiative met or exceeded ROI expectations, with 20% reporting ROI of 31% or more. n=2,773, fielded July to September 2024

    ~74%

    Medium confidence

  • Wharton Human-AI Research with GBK Collective

    Share of leaders reporting positive ROI to date, with 72% formally measuring it. n=801 US enterprises with 1,000+ employees, fielded June 26 to July 11, 2025

    74%

    High confidence

What this does not say

The four answers are not four measurements of one quantity. Each source asked a different question of a different population.

Publisher
MIT NANDA, McKinsey, Deloitte, Wharton and GBK
Sample and method
Four sources, one per row, each with its own sample and field dates. The MIT PDF was reachable at research time only via a third-party mirror
Field dates
Not published by the source.

High confidence in the existence of the disagreement. Each row carries its own confidence tag.

High confidence

Also in the record

Figures that sit alongside these charts.

  • Before you quote an AI ROI number, name the population, the question wording, and the field dates. Three numbers in this chart are defensible and none of them is interchangeable with another.
  • Fortune's widely republished description of the MIT methodology does not match the report's own description of it.

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.

01

Why the answers differ

MIT looked at pilots and required measurable P&L impact, which is the strictest test and produces the lowest number. Wharton and Deloitte asked leaders about their most advanced initiative, which selects for the best case inside each organization and produces the highest numbers. McKinsey required attribution to EBIT, which sits in between because it accepts any attributable effect but demands attribution.

Each design is defensible for its purpose. A CFO deciding whether the portfolio pays back wants the MIT question. A strategist asking whether the technology can work anywhere wants the Wharton and Deloitte question. Confusing the two is how a board gets told the same program is a 5 percent story and a 74 percent story in the same month.

02

How the confusion propagates

The number travels and the second line does not. Fortune's widely republished description of the MIT methodology does not match the report's own description of it, and that misdescription is the version most people encountered. Once a number is separated from its question it becomes usable in any argument, which is precisely why it spreads faster than its caveats.

This is the mechanism documented in detail in the citation decay theme. The four-answers chart is the diagnostic; citation decay is the pathology.

03

The usable rule

Before quoting an AI ROI number, name three things: the population, the question wording, and the field dates. If you cannot supply all three, you are quoting a mood rather than a measurement.

Inside your own organization, pick one definition and hold it for at least four quarters. The value of a consistent internal definition exceeds the value of a favorable external benchmark, because only the internal one can tell you whether anything changed.

What to do with it

The move, by seat.

Board reporting
Attach population, wording, and field dates to every external number on the slide. Numbers without them get challenged and should.
CFO
Use the strictest definition internally. It is the only one that supports a funding decision.
Comms and content
Stop averaging ROI figures across publishers. The spread is a methodology artifact, not a range.

Questions this page answers

What the data says, in plain language.

What does the research show about The same question, four different answers?
Enterprise AI is paying off for 5% of companies, or 74%, depending on the question asked. None of the four publishers is wrong. Four credible publishers measured something adjacent to is enterprise AI paying off and produced answers that cannot all be right. They are not measuring the same thing.
What does the figure "Enterprise AI is paying off for 5% of companies, or 74%, depending on the question asked" show?
None of these four publishers is wrong. They asked four different questions. Source: MIT NANDA, McKinsey, Deloitte, Wharton and GBK. Four sources, one per row, each with its own sample and field dates. The MIT PDF was reachable at research time only via a third-party mirror Confidence: High. Caveat: High confidence in the existence of the disagreement. Each row carries its own confidence tag.
What else sits alongside these figures?
Before you quote an AI ROI number, name the population, the question wording, and the field dates. Three numbers in this chart are defensible and none of them is interchangeable with another. Fortune's widely republished description of the MIT methodology does not match the report's own description of it.
Where does this data come from?
Every figure is reproduced from a named publisher: MIT NANDA, The GenAI Divide: State of AI in Business 2025, Wharton Human-AI Research with GBK Collective, 2025 AI Adoption Report, McKinsey, The State of AI: Global Survey 2026, Deloitte, State of Generative AI in the Enterprise, Q4 wave. 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/four-answers

Kvarfordt, Jonathan. "The same question, four different answers." The Revenue AI Report, Research Library. https://www.therevenueaireport.com/research/four-answers

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.

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.

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 same question, four different answers. The Revenue AI Report. Retrieved from https://www.therevenueaireport.com/research/four-answers

MLA

Kvarfordt, Jonathan. "The same question, four different answers." The Revenue AI Report, 31 Aug. 2026, www.therevenueaireport.com/research/four-answers.

BibTeX

@misc{kvarfordt2026fouranswers,
  author = {Kvarfordt, Jonathan},
  title = {The same question, four different answers},
  year = {2026},
  publisher = {The Revenue AI Report},
  url = {https://www.therevenueaireport.com/research/four-answers}
}

Share this research

Posting to Instagram or TikTok? Copy the link, it carries the title, summary and share image.

Next theme

Nobody in revenue publishes research with nothing to sell

Sixty publishers who write about AI, GTM, or both, assessed against four tests. Three pass all four. None of the three cover revenue.

Subscribe

Get the weekly issue built on this data.

Arrives weekly by email. Free. Unsubscribe anytime. By subscribing you agree to our Privacy policy and Terms. We never sell or share the list.