Of the 16,000 projects in the largest dataset of its kind in the world, 0.5% delivered on budget, on time and with the benefits promised. IT projects carry the fattest cost-overrun tail of 23 project types. Developers now save 13 hours a week with AI and lose ten to organizational friction. And the most quoted number in software, the 31% success rate, comes from a report whose latest edition appeared in 2020. These software development statistics for 2026 cover the software project failure rate, cost overruns, the average software development cost, how developers spend their week and enterprise software development speed statistics, each figure dated, because most "2026 statistics" in this genre were measured a decade ago.
Key software development statistics 2026
1. 0.5% of the 16,000+ projects in Oxford's Flyvbjerg database delivered on budget, on time and with the promised benefits (Flyvbjerg and Gardner, 2023 dataset). 2. IT projects run 73% over budget on average, and the 18% that go more than 50% over run 447% over (Flyvbjerg database, 2023). 3. 31% of software projects succeeded, 50% were challenged and 19% failed in CHAOS 2020, the latest full Standish report (2020). 4. 30% of companies running large tech programs met timeline, budget and scope expectations; 35% missed all three (BCG, 1,000+ C-suite executives, November 2024). 5. 31% of complex projects fail to achieve the full scope of their intended benefits, more than twice the rate for projects overall (PMI Pulse of the Profession 2026). 6. The average software project on Clutch costs $132,480 and takes about 13 months (Clutch, updated September 2026). 7. 50% of developers lose 10 or more hours a week, largely to organizational inefficiencies, while 68% save more than 10 hours with AI tools (Atlassian, 3,500 respondents, July 2025). 8. 21% of developers spend more than half their week writing new code from scratch (BairesDev Dev Barometer, 705 developers, September 2026). 9. 80% of features in the average software product are rarely or never used (Pendo, 615 subscriptions, 2019, never re-measured). 10. Poor software quality cost the US at least $2.41 trillion in 2022, with technical debt principal of about $1.52 trillion (CISQ, latest edition). 11. 9.4% of teams get a change from commit to production in under an hour; 43.5% need more than a week (DORA 2025, nearly 5,000 technology professionals).
What is the software project failure rate?
Between 19% and 70%, depending on the bar. 19% of software projects failed outright and another 50% were late, over budget or short on features in the latest Standish CHAOS report (2020); 70% of companies running large tech programs missed timeline, budget or scope expectations in BCG's 2024 survey. Of all these software development statistics, the software project failure rate depends most on its definition.
The Standish CHAOS series by edition
Standish CHAOS 2020: Beyond Infinity found 31% of projects successful, 50% challenged and 19% failed. Two facts most pages omit. First, the long-run trend was improvement: the original 1994 report found 16% successful, 53% challenged and 31% failed (the 2015 Modern Resolution definition changed the yardstick, so the rows are not comparable point to point). Second, CHAOS 2020 is the latest full edition, according to BudgetOverrun's edition history; Henny Portman's January 2021 review records Standish's forecast that the Agile period gives way to what it calls the Flow period. Every "Standish 2024" citation is recycled vintage.
Large programs and complex projects
BCG's November 2024 survey of more than 1,000 C-suite executives across 20 sectors found 30% champions, whose programs met timeline, budget and scope expectations, 35% midtiers, which met budget but not timeline and scope, and 35% laggards, which missed all three. BCG's own summary of the decade since its 2015 research: "Ten years is a long time to show no improvement." PMI's Pulse of the Profession 2026 adds the complexity lens: 31% of complex projects fail to achieve the full scope of their intended benefits, more than twice the rate reported for projects overall.
For the never-measured "70% of transformations fail" claim, see our digital transformation statistics; for company-level base rates, our startup failure rate statistics.
How much do IT projects run over budget?
73% on average, with a tail that averages 447%. In Bent Flyvbjerg's Oxford database of more than 16,000 projects, IT projects show a mean cost overrun of 73%, and the 18% of IT projects that go more than 50% over budget overrun by 447% on average, the fattest tail of the 23 project types in the dataset. Across all types, only 0.5% came in on budget, on time and with the promised benefits. The peer-reviewed version, The Uniqueness of IT Cost Risk (Project Management Journal, online July 2025), finds that IT cost risk has a fatter tail than any other project type.
The figure most pages still quote instead, 45% over budget and 56% less value, is the McKinsey and Oxford study of more than 5,400 IT projects as of June 2012: large IT projects ran 45% over budget and 7% over time while delivering 56% less value than predicted. The practical reading of the current software development statistics: the average project hurts, and the tail kills. We unpack what drives the tail in our software project cost overruns analysis.
What is the average software development cost, and how long does it take?
$132,480 and about 13 months for the average project on Clutch, with the typical project far cheaper. Clutch's pricing guide, updated September 21, 2026 from verified project reviews, puts the average software development cost at $132,480.29 and the usual timeline at about 13 months, while the most common band is $10,000 to $49,999. The average is a mean pulled up by large builds.
The vendor side: the GoodFirms Custom Software Development Cost Survey 2026 (March 2026, responses from more than 100 software development companies) found 66% of companies charge $30,000 to $100,000 for small to mid-sized projects, enterprise-grade platforms often exceed $200,000, and scope creep increases development costs by 10 to 25%.
For what teams cost rather than projects, see our developer salary statistics, developer headcount data and IT outsourcing statistics.
How do developers actually spend their time?
Mostly not writing code: 21% of developers spend more than half their week writing new code from scratch, and half lose 10 or more hours a week, largely to organizational inefficiencies. The time-allocation software development statistics, dated:
- 2018: Stripe's Developer Coefficient (September 2018, thousands of C-level executives and developers across six countries) found developers working a 41.1-hour week and spending 17.3 hours of it, 42%, on maintenance such as debugging and refactoring, plus about four hours a week on "bad code", worth nearly $85 billion a year in opportunity cost.
- 2025: Atlassian's State of Developer Experience (July 2025, 3,500 developers and managers in six countries): 68% save more than 10 hours a week with AI tools, while 50% lose 10 or more hours a week, largely to organizational inefficiencies. Atlassian's own summary: developers are saving 10 hours a week with AI and losing 10 hours a week to inefficiencies.
- 2026: BairesDev's Q3 2026 Dev Barometer (September 2026, 705 developers and 41 CTOs, vendor data): 21% of developers spend more than half their week writing new code from scratch, developers save 13 hours a week on coding with AI, up from about 7 a year earlier, and 67% spend more time reviewing AI-generated code. Stack Overflow's 2026 survey of more than 30,000 technologists found 80% using AI at least an hour a day.
The through-line: writing code was never the bottleneck, and AI moved the bottleneck into review and coordination. The AI layer's own numbers are in our AI code generation statistics.
What do unused features, poor quality and technical debt cost?
Up to $29.5 billion of R&D on features that may rarely or never be used, and at least $2.41 trillion a year in poor quality, by the only measurements that exist. These are the oldest software development statistics on the page, so the dates matter.
What percentage of software features get used
About 20%. Pendo's 2019 Feature Adoption Report analyzed feature usage across 615 Pendo subscriptions and found that 12% of features generate 80% of average daily usage volume while 80% of features are rarely or never used. Applying that share to the reported R&D expenses of 54 public software companies on the Bessemer Emerging Cloud Index, Pendo estimated up to $29.5 billion spent on features that may rarely or never be used. No comparable edition has appeared since.
The cost of poor software quality
- CISQ's Cost of Poor Software Quality in the US: A 2022 Report estimates the 2022 cost at at least $2.41 trillion, up from $2.08 trillion in its 2020 report, with accumulated technical debt principal of about $1.52 trillion, up from $1.31 trillion in 2020. No later edition has been published.
- CIOs estimated that tech debt amounts to 20 to 40% of the value of their entire technology estate before depreciation, and that 10 to 20% of the technology budget dedicated to new products is diverted to resolving tech debt issues (McKinsey, 50 CIOs of financial-services and technology companies with revenues above $1 billion, July 2020).
- Myth check for the oldest number in the genre: "fixing a bug in production costs 100x more than in design" is attributed to an IBM Systems Sciences Institute study that, as The Register reported in 2021, nobody has been able to find.
What are the enterprise software development speed statistics?
Slower than the discourse suggests: 9.4% of teams get a change from commit to production in under an hour, and 43.5% need more than a week. Those are the enterprise software development speed statistics from DORA's 2025 report, State of AI-assisted Software Development, surveyed June 13 to July 21, 2025 among nearly 5,000 technology professionals.
Where teams actually sit in 2025
Lead time: 9.4% under an hour, 15% under a day, 31.9% between a day and a week, 28.3% between a week and a month, 15.2% beyond a month. Deployment frequency: 16.2% deploy on demand, 21.9% between daily and weekly, 23.9% less often than monthly. Change failure rate: 8.5% of teams sit at 0 to 2%, while the largest group, 26%, lands at 8 to 16%. DORA 2024 added a fifth metric, rework rate, and DORA 2025 replaced its four performance levels with seven team archetypes, so it publishes no new elite-versus-low table.
How far ahead elite teams are
The last published multipliers are in the DORA 2024 report, its tenth: elite performers, 19% of respondents, realize 127x faster lead time, 182x more deployments per year, 8x lower change failure rate and 2,293x faster failed-deployment recovery than low performers, who are 25% of respondents. The same report found internal platform use associated with 6% higher software delivery and operations performance, alongside about 8% lower throughput and 14% lower change stability. Paired with Atlassian's friction data, the gap between elite and average is not typing speed; it is everything around the typing. More context in our software development trends data.
Methodology: which edition these software development statistics come from
Every figure above is quoted from the page linked, with the year it was measured, because the famous numbers in this genre travel without their dates:
- Standish CHAOS: "31% succeed, 19% fail" is CHAOS 2020, the latest full edition; 1994 to 2009 use the traditional definition and 2015 to 2020 the Modern Resolution definition, so the rows are not comparable.
- McKinsey and Oxford: "45% over budget, 56% less value" was measured on 5,400 projects as of June 2012; the current overrun figures are Flyvbjerg's 2023 dataset.
- Stripe: "42% of developer time on maintenance" is September 2018; Atlassian 2025 and BairesDev 2026 are current, and BairesDev's is vendor data.
- Pendo: "80% of features unused" is 2019 and has not been re-measured.
- CISQ: "$2.41 trillion" is the 2022 report, still the latest; editions use different component sets, so quote the edition.
- DORA: the distributions are DORA 2025 (nearly 5,000 respondents); the multipliers are DORA 2024, because DORA 2025 replaced its four performance levels with seven team archetypes. Clutch and GoodFirms figures are marketplace and vendor data.
What the honest software development statistics add up to: delivery fails in the tail, loses time in coordination and ships features nobody uses, and all three are process choices rather than laws of nature. Beating those base rates is the product we sell in custom software development.