References

Every figure we publish carries a source you can open. This page lists them — what each source is, what it supports on this site, and a link to the original — so a number you read here can be checked without taking our word for it.

How to read this page

Each entry names a source, the figure it supports on this site, and the elephant that figure belongs to; the title opens the original in a new window. An entry marked withdrawn stays listed, with the reason — a figure we stop using is retired in the open, never quietly removed.

[1] World Economic Forum — Future of Jobs Report 2025

The 63%: employers naming skills gaps the biggest barrier to transformation. Also the top-ten skills for 2030.

How we vetted this source3kept · 1 withdrawn

Primary source
read, not summarized
4 figures proposed
each one a claim in the ledger
Verified
quote, type, tier — by an AI verifier
Steward decides
3 kept · 1 withdrawn
Published
the homepage

Kept

Employers expect 39% of workers' core skills to change by 2030 (down from 44% in 2023).
Their words: “employers expect 39% of workers' core skills to change by 2030”
Evidence:

Survey — self-reported · Tier 2 — research institute or official statistics
Verified:

September 3, 2026 — Sylvain Dufour
Cleared:

August 19, 2026 — Sylvain Dufour
Re-check by:

August 19, 2027
If the world's workforce were 100 people, 59 would need training by 2030.
Their words: “if the world's workforce was made up of 100 people, 59 would need training by 2030”
Evidence:

Survey — self-reported · Tier 2 — research institute or official statistics
Verified:

September 3, 2026 — Sylvain Dufour
Cleared:

August 19, 2026 — Sylvain Dufour
Re-check by:

August 19, 2027
Zero of 2,800+ assessed skills show very-high capacity for GenAI substitution; 69% show low or very low capacity (GPT-4o assessment method, stated).
Their words: “Zero of the more than 2,800 skills assessed were determined to exhibit 'very high capacity' to be replaced”
Evidence:

Modeled estimate · Tier 2 — research institute or official statistics
Verified:

September 3, 2026 — Sylvain Dufour
Cleared:

August 19, 2026 — Sylvain Dufour
Re-check by:

August 19, 2027

Withdrawn

63% of employers identify skills gaps as the biggest barrier to business transformation, 2025-2030.
Why: The source measures something narrower or different from what the statement claimed — August 19, 2026

[3] McKinsey Global Institute — Agents, robots, and us (2025)

57% of U.S. work hours technically automatable with today’s technologies — the Robot elephant’s number.

How we vetted this source1kept · 5 refused

Primary source
read, not summarized
6 figures proposed
each one a claim in the ledger
Verified
quote, type, tier — by an AI verifier
Steward decides
1 kept · 5 refused
Published
the “robot” article, this page

Kept

57% of U.S. work hours could technically be automated with already-existing technologies (hours and technical potential, not jobs or likelihood).
Their words: “We estimate that today’s technology could, in theory, automate about 57 percent of current US work hours”
Evidence:

Modeled estimate · Tier 2 — research institute or official statistics
Verified:

September 3, 2026 — Sylvain Dufour
Cleared:

August 22, 2026 — Sylvain Dufour
Re-check by:

August 22, 2027

Refused — from the same source

More than 70% of today's skills can be applied in both automatable and non-automatable work.
Why: The source measures something narrower or different from what the statement claimed
Nearly 90% of companies say they have invested in AI, but fewer than 40% report measurable gains.
Why: The primary could not be read and nothing verifies it one hop away
In McKinsey's midpoint adoption scenario, roughly 27% of current US work hours are automated by 2030.
Why: A projection stated as a measurement
About 60% of potential productivity gains are concentrated in sector-specific workflows — the activities at the core of each industry.
Why: A projection stated as a measurement
In McKinsey's midpoint adoption scenario, AI agents and robots could generate about $2.9 trillion in US economic value a year by 2030, conditional on organizations redesigning workflows rather than automating individual tasks.
Why: The source measures something narrower or different from what the statement claimed

[4] Asana — Anatomy of Work Index

209 hours a year per knowledge worker on duplicated work — the Duplicate elephant’s number.

How we vetted this source1kept

Primary source
read, not summarized
1 figure proposed
each one a claim in the ledger
Verified
quote, type, tier — by an AI verifier
Steward decides
1 kept
Published
the “duplicate” article, this page

Kept

Knowledge workers spend 209 hours a year on work that has already been done.
Their words: “over the course of a year, the average knowledge worker spends 103 hours in unnecessary meetings, 209 hours on duplicative work”
Evidence:

Survey — self-reported · Tier 3 — large-sample industry study
Verified:

September 3, 2026 — Sylvain Dufour
Cleared:

August 22, 2026 — Sylvain Dufour
Re-check by:

August 22, 2027

[5] Oracle & Seth Stephens-Davidowitz — The Decision Dilemma (2023)

78% of leaders say decisions come first and data is found to justify them — the Bias elephant’s number. n=14,250, 17 countries.

How we vetted this source1kept · 1 refused · 1 withdrawn

Primary source
read, not summarized
3 figures proposed
each one a claim in the ledger
Verified
quote, type, tier — by an AI verifier
Steward decides
1 kept · 1 refused · 1 withdrawn
Published
the “bias” article

Kept

78% of business leaders say people often make decisions and then look for the data to justify them.
Their words: “78 percent of business leaders say people often make decisions and then look for the data to justify them”
Evidence:

Survey — self-reported · Tier 3 — large-sample industry study
Verified:

September 16, 2026 — an AI verifier
Cleared:

September 15, 2026 — Sylvain Dufour
Re-check by:

September 15, 2027

Refused — from the same source

78% of business leaders say people often make decisions and then look for the data to justify them (n=14,250, 17 countries).
Why: The source measures something narrower or different from what the statement claimed

Withdrawn

78% of business leaders say the decision comes first and data is found to justify it (n=14,250, 17 countries).
Why: The source’s own words do not contain the figure — August 22, 2026

[6] RAND Corporation — The Root Causes of Failure for AI Projects (2024)

The #1 root cause of AI project failure: misunderstanding the problem to be solved — the Mirage elephant’s evidence.

How we vetted this source3kept

Primary source
read, not summarized
3 figures proposed
each one a claim in the ledger
Verified
quote, type, tier — by an AI verifier
Steward decides
3 kept
Published
the “mirage” article, this page

Kept

The most common root cause of AI project failure is misunderstanding the problem to be solved (practitioner-interview study).
Their words: “To investigate why AI projects fail, we interviewed 65 experienced data scientists and engineers.”
Evidence:

Expert opinion · Tier 2 — research institute or official statistics
Verified:

September 3, 2026 — Sylvain Dufour
Cleared:

August 22, 2026 — Sylvain Dufour
Re-check by:

August 22, 2027
Misunderstanding or miscommunicating the intent and purpose of the problem to be solved is the most common reason AI projects fail.
Their words: “Misunderstandings and miscommunications about the intent and purpose of the project are the most common reasons for AI project failure.”
Evidence:

Survey — self-reported · Tier 2 — research institute or official statistics
Verified:

September 4, 2026 — an AI verifier
Cleared:

September 4, 2026 — Sylvain Dufour
Re-check by:

September 4, 2027
By some estimates, more than 80 percent of AI projects fail — twice the rate of failure for information technology projects that do not involve AI, according to a RAND research report.
Their words: “By some estimates, more than 80 percent of AI projects fail—twice the rate of failure for information technology projects that do not involve AI.”
Evidence:

Expert opinion · Tier 2 — research institute or official statistics
Verified:

September 4, 2026 — an AI verifier
Cleared:

September 4, 2026 — Sylvain Dufour
Re-check by:

September 4, 2027

[7] IBM Institute for Business Value (2025)

More than one organization in four estimates losing $5M+ a year to poor data quality — the Fuzzy elephant’s number.

How we vetted this source1kept

Primary source
read, not summarized
1 figure proposed
each one a claim in the ledger
Verified
quote, type, tier — by an AI verifier
Steward decides
1 kept
Published
the “fuzzy” article, this page

Kept

More than one organization in four estimates losing $5M+ a year to poor data quality; 7% put the figure above $25M.
Their words: “over a quarter of organizations estimate they lose more than USD 5 million annually due to poor data quality”
Evidence:

Survey — self-reported · Tier 2 — research institute or official statistics
Verified:

September 3, 2026 — Sylvain Dufour
Cleared:

August 22, 2026 — Sylvain Dufour
Re-check by:

August 22, 2027

[8] Atlassian — State of Teams 2025

25% of teams’ time goes to just searching for answers — the Silo elephant’s number. n=12,000.

How we vetted this source1kept

Primary source
read, not summarized
1 figure proposed
each one a claim in the ledger
Verified
quote, type, tier — by an AI verifier
Steward decides
1 kept
Published
the “silo” article, this page

Kept

25% of teams' time goes to just searching for answers (n=12,000).
Their words: “survey of 12,000 knowledge workers and 200 executives found that leaders and teams waste 25% of their time just searching for answers”
Evidence:

Survey — self-reported · Tier 3 — large-sample industry study
Verified:

September 3, 2026 — Sylvain Dufour
Cleared:

August 22, 2026 — Sylvain Dufour
Re-check by:

August 22, 2027

[withdrawn] Panopto — Workplace Knowledge & Productivity Report (2018)

Withdrawn 2026-08-23: “42% of a role’s knowledge is held by one person only”, the Guru elephant’s former number. Its only support was a 2018 vendor survey with no quotable sentence carrying the figure, and a search for a stronger measurement refused every candidate — the closest was peer-reviewed but measured software projects, not roles. The elephant stands; the statistic does not.

How we vetted this source0kept · 1 withdrawn

Primary source
read, not summarized
1 figure proposed
each one a claim in the ledger
Verified
quote, type, tier — by an AI verifier
Steward decides
1 withdrawn
Published
nowhere — nothing from this source is in use

Withdrawn

42% of the knowledge a role depends on is held by one person only.
Why: Only a self-interested or vendor source supports it — August 22, 2026

[10] Microsoft & LinkedIn — Work Trend Index 2024

78% of AI users bring their own AI tools to work — the Shadow elephant’s number. n=31,000.

How we vetted this source1kept · 1 withdrawn

Primary source
read, not summarized
2 figures proposed
each one a claim in the ledger
Verified
quote, type, tier — by an AI verifier
Steward decides
1 kept · 1 withdrawn
Published
the “shadow” article

Kept

78% of AI users bring their own AI tools to work (n=31,000, 31 countries).
Their words: “78% of AI users are bringing their own tools to work”
Evidence:

Survey — self-reported · Tier 3 — large-sample industry study
Verified:

September 16, 2026 — an AI verifier
Cleared:

September 15, 2026 — Sylvain Dufour
Re-check by:

September 15, 2027

Withdrawn

78% of AI users bring their own AI tools to work, outside governance (n=31,000, 31 countries).
Why: The source measures something narrower or different from what the statement claimed — August 22, 2026

[11] Lee et al. (Microsoft & Carnegie Mellon) — The Impact of Generative AI on Critical Thinking, CHI 2025

Across 936 real GenAI tasks: more trust in the AI, less critical thinking reported — the Zombie elephant’s evidence.

How we vetted this source1kept

Primary source
read, not summarized
1 figure proposed
each one a claim in the ledger
Verified
quote, type, tier — by an AI verifier
Steward decides
1 kept
Published
the “zombie” article, this page

Kept

Across 936 real GenAI tasks, higher confidence in the AI correlates with less self-reported critical thinking.
Their words: “Participants shared 936 first-hand examples of using GenAI in work tasks.”
Evidence:

Survey — self-reported · Tier 1 — peer-reviewed
Verified:

September 3, 2026 — Sylvain Dufour
Cleared:

August 22, 2026 — Sylvain Dufour
Re-check by:

August 22, 2029

[2] One more, unlinked on purpose

S&P Global Market Intelligence, 2025 — the 42%/17% AI-abandonment trajectory and the 46% of proofs-of-concept scrapped before production. The primary report could not be retrieved (spglobal.com blocks automated access); we cite it through CIO Dive, a named secondary, carrying a one-hop rate-down.

Where to go from here

Read the articles

The stories these sources hold up — one per elephant, each naming the figures it rests on.

How we decide what to keep

The methodology behind every entry here: what a source has to show before a figure reaches a page, and what happens when it no longer does.

Have a source we should look at?

Stronger evidence, a correction, or a figure you think we got wrong — we read it, and we say what we did with it.

Talk with us