The Standish CHAOS Report by Year: Every Edition from 1994 to 2020
The CHAOS Report has tracked IT project outcomes since 1994. This page dates every landmark edition, gives the original 1994 figures straight from the primary report, and explains the 2015 definition change that makes the years hard to compare.
Which CHAOS Report editions exist, and what did each say?
The Standish Group has published a CHAOS Report roughly every two years since 1994. The original 1994 report found just 16% of projects successful; by the CHAOS Summary 2009 it was about 32%; and under the Modern Resolution definition introduced in 2015 it was 29%, rising to 31% in CHAOS 2020: Beyond Infinity. CHAOS 2020 is the latest and, by Standish's own account, the last full edition, so there is no 2021 to 2026 report.
The original 1994 CHAOS Report, from the primary source
The report that started it all is the 1994 CHAOS study (the widely-circulated reprint carries a 1995 Standish Group copyright). It surveyed 365 IT executive managers covering 8,380 applications, and sorted every project into three "resolution types". These are the exact figures from the report itself:
| Resolution type | Definition | 1994 share |
|---|---|---|
| Type 1 – Successful | On time, on budget, with all features and functions as initially specified | 16.2% |
| Type 2 – Challenged | Completed and operational, but over budget, over time, and with fewer features than specified | 52.7% |
| Type 3 – Impaired | Cancelled at some point during the development cycle | 31.1% |
The report is also the origin of two numbers still quoted today: challenged and impaired projects ran, on average, 189% of their original cost estimate and 222% of their original time estimate. Size mattered even then: only 9% of large-company projects succeeded, against 16.2% for medium and 28% for small companies. And the top three success factors it named (user involvement, executive management support, and a clear statement of requirements) are still the standard advice thirty years later.
The Standish Group, CHAOS (1994; reprinted with permission, copyright 1995), pages 2–3. Figures read directly from the primary report PDF.
Success rate by edition, 1994 to 2020
Standish issued a new CHAOS Report roughly every two years. The table below gives the landmark editions. Read it with one caveat above all: the definition of "successful" changed in 2015, so the 1994–2009 rows and the 2015–2020 rows are measured on different yardsticks and should not be compared point to point.
| Edition | Successful | Challenged | Failed | Definition |
|---|---|---|---|---|
| 1994 (original) | 16% | 53% | 31% | Traditional |
| 2009 (CHAOS Summary) | 32% | 44% | 24% | Traditional |
| 2015 | 29% | 52% | 19% | Modern Resolution |
| 2020 (Beyond Infinity) | 31% | 50% | 19% | Modern Resolution |
1994: primary report (see above). 2009: Standish CHAOS Summary 2009, as widely reported. 2015: Standish Group CHAOS Report 2015, Modern Resolution, FY2011–2015 (read from the primary report). 2020: CHAOS 2020 "Beyond Infinity" as reported in open secondary sources. Standish also published editions in the intervening years (1996, 1998, 2000, 2002, 2004, 2006, 2011, 2013); on the traditional definition their reported success rate hovered between roughly a quarter and a third, but those figures come from paywalled reports and are treated with caution here.
The 2015 definition change: why the years do not line up
From 1994 to the early 2010s, CHAOS scored a project "successful" only if it was on time, on budget, and delivered every originally-specified feature: the traditional resolution. In the CHAOS Report 2015, Standish switched its headline metric to the Modern Resolution: on time, on budget, with a satisfactory result. The new definition drops the strict all-features test in favour of a satisfaction judgement, so a project delivered with fewer features but a happy user can now count as successful where it would once have been "challenged".
This is why a jump from 16% (1994) to 31% (2020) overstates real improvement: some of it is a looser scope test. It is also why the same 2020 report is quoted with different success percentages depending on which definition a source uses. The 29% and 31% figures on this page are the Modern Resolution numbers, the version most widely cited.
Is there a newer CHAOS Report?
No. CHAOS 2020: Beyond Infinity, written by the Standish Group's Jim Johnson, is the most recent full CHAOS Report and, by the Standish Group's own account, the last: the firm has said it is turning its research toward a "Flow" model rather than another complete edition. There is no CHAOS Report for 2021, 2022, 2023, 2024, 2025, or 2026. If you are looking for "the latest CHAOS Report", it is the 2020 edition, and its 31% successful / 50% challenged / 19% failed split remains the current reference point.
Beyond Infinity also moved past the raw success/challenged/failed split, arguing that outcomes are driven by three Factors of Success (a Good Sponsor, a Good Team, and a Good Place) and naming decision latency as the root cause of project failure.
Full detail on CHAOS 2020, the classification, and the methodology ›
A note on reliability
CHAOS is a self-reported survey of IT executives, not an audited dataset, and the raw data is proprietary. The most cited academic critique is Eveleens and Verhoef's 2010 IEEE Software paper, The Rise and Fall of the CHAOS Report Figures, which argued that Standish's definitions can distort the numbers and inflate apparent failure. Treat the year-by-year figures as directional: the broad shape (only about a third of projects fully succeed, large projects fare far worse than small ones) is robust and echoed by independent datasets, while any single percentage carries that methodological caveat.
How to cite
Document: CHAOS Report 2015 (PDF, copy hosted by Info-Tech Research Group), read 8 October 2026. CHAOS 2020 is sold by the Standish Group (product listing); we have not read it, so its row is as reported.
The full CHAOS reports are paywalled. The original 1994 report is reproduced in academic course archives; the 2015 tables are widely reproduced in openly accessible secondary sources (Scrum.org, InfoQ, PM World Journal).
Related references on this site
Frequently Asked Questions
What did the original 1994 CHAOS Report find?
The first Standish Group CHAOS Report, published in 1994, found that only 16.2% of software projects were successful (delivered on time, on budget, with all specified features), 52.7% were challenged (completed but over budget, over time, or with fewer features), and 31.1% were impaired and cancelled outright. It was based on 365 IT-executive respondents covering 8,380 applications. Challenged and impaired projects ran 189% over their original cost estimate and 222% over their original time estimate on average. It is the source of the famous claim that most software projects fail.
How has the CHAOS Report success rate changed by year?
On the traditional definition (on time, on budget, all features), the reported success rate rose from 16% in 1994 to about 35% in 2006 and 32% in the CHAOS Summary 2009. Standish changed its yardstick in 2015 to the Modern Resolution (on time, on budget, with a satisfactory result), under which 29% of projects were successful in 2015 and 31% in CHAOS 2020. Because the definition changed in 2015, the pre-2015 and post-2015 percentages are not directly comparable, but the headline that only about a third of projects fully succeed has held for two decades.
Is there a CHAOS Report for 2026 (or any year after 2020)?
No. CHAOS 2020, subtitled Beyond Infinity and written by the Standish Group's Jim Johnson, is the most recent full CHAOS Report and, by the Standish Group's own account, the last complete edition; the firm has said it is shifting its research toward a Flow model. There is no 2021, 2022, 2023, 2024, 2025, or 2026 CHAOS Report. When people search for a 'CHAOS Report 2026' they are looking for the current reference point, which remains CHAOS 2020.
Why did the CHAOS success rate jump between the 1994 and later reports?
Partly real improvement (smaller projects, iterative delivery, better tooling) and partly definitional. Academic critics, notably J. Laurenz Eveleens and Chris Verhoef in their 2010 IEEE Software paper 'The Rise and Fall of the CHAOS Report Figures', argued that the way Standish defines and measures success can distort the numbers and inflate apparent failure. Standish's data is a self-reported survey, not an audited dataset, and the raw data is proprietary, so the year-to-year figures should be read as directional rather than precise.