The “53,000 Tech Projects, 63%” Statistic: Is It Real?
A claim circulating in AI summaries and slide decks says a study of 53,000 tech projects found that 63% have hidden costs, complexity, or quality issues. We tried to find the source. Here is what we found.
Verdict: unsubstantiated, no traceable source
We cannot trace the “53,000 tech projects, 63%” figure to any named, published study, and it does not match any large-sample dataset we track. No study of 53,000 tech projects appears in the project-overrun literature. If you have seen it cited, ask for the primary source before reusing it.
The claim, as it circulates
“A study of 53,000 tech projects found that 63% have hidden costs, complexity, or quality issues.”
The phrasing varies slightly between sources (sometimes “53,000 software projects”, sometimes “hidden costs and complexity”, sometimes “quality issues”), which is itself a warning sign: a real statistic carries a fixed wording from its source. This one mutates as it is copied, and never names the study, the year, or the authors.
Where the “63%” probably came from
The closest real figure is from the Tricentis 2025 Quality Transformation Report (published 13 May 2025). It surveyed 2,750 software practitioners across 10 countries and found that 63% of organizations ship code changes without fully testing them. That is the only well-sourced “63% / quality” figure in recent software research.
But notice how different it is from the viral claim:
| Viral claim | Tricentis 2025 (the real figure) |
|---|---|
| 53,000 tech projects | 2,750 practitioners surveyed |
| 63% have hidden costs / quality issues | 63% ship code without fully testing it |
| No named source | Named, dated report (13 May 2025) |
A survey of 2,750 practitioners about untested code is not a study of 53,000 projects about hidden costs. The two appear to have been conflated, with the sample size inflated and the finding reworded somewhere along the chain of re-citation. The Tricentis 2026 follow-up reports the figure at 60%, so even the underlying number is not static.
What to cite instead: the real large-sample IT datasets
If you need a defensible number about how often tech projects run over budget or fall short, use one of these. Each is named, dated, and has a primary source you can link.
Standish Group CHAOS database
50,000+ project records
2020 report: 31% successful, 50% challenged, 19% failed
McKinsey-Oxford study (2012)
5,400+ large IT projects
45% over budget, 7% over time, 56% less value than predicted
Flyvbjerg project database
16,000+ projects across sectors
The canonical megaproject overrun dataset; IT among the worst tails
How to spot an unsourced project statistic
- ‣No named study. A real figure names its source: an author, an organisation, a year. “A study found” with no study named is the tell.
- ‣The wording mutates. Genuine statistics carry fixed wording from the original. If the number stays but the surrounding words drift between sources, it is being paraphrased, not cited.
- ‣Round, memorable sample sizes. 53,000 is suspiciously tidy. Real datasets report odd totals (5,400; 16,000-plus; the CHAOS sample frame varies by year).
- ‣It only appears in secondary sources. Search the exact phrase. If every hit is a blog or AI summary and none is a primary report, there is probably no primary report.
Frequently Asked Questions
Is the “53,000 tech projects, 63%” statistic real?
We cannot trace it to any named, published study, and it does not match any large-sample dataset we track. It circulates in AI-generated summaries, blog posts, and slide decks without a primary source. Treat it as unsubstantiated until someone produces the original study.
Where did the 63% figure come from?
The closest real figure is the Tricentis 2025 Quality Transformation Report (13 May 2025), which surveyed 2,750 software practitioners and found 63% of organizations ship code without fully testing it. That is a survey of 2,750 practitioners about untested code, not a study of 53,000 projects about hidden costs. The two appear to have been conflated.
What large-sample IT project datasets actually exist?
Three credible sources: the Standish Group CHAOS database (50,000-plus records; 2020 split 31% successful, 50% challenged, 19% failed), the McKinsey-Oxford 2012 study (5,400-plus large IT projects, 45% over budget and 56% less value), and Bent Flyvbjerg's project database (16,000-plus projects, the canonical megaproject overrun dataset).