Startup, MVP and No-Code Statistics, Fact-Checked (2026)

Some statistics get quoted so often that nobody checks them any more. "90% of startups fail." "42% fail because nobody wants the product." "64% of features are never used." "70% of new apps will be low-code by 2025." They show up in pitch decks, MVP guides and blog posts, often years after the source changed, the forecast year passed, or it turned out the number came from a very small study.

This page checks the ones we see most in articles about startups, MVPs, no-code and building apps with AI. For each one: where it comes from, how old it is, what it was based on, and what to cite instead. We checked every source on the date at the top of this page and will keep it updated.

The short version

"90% of startups fail"

This is the most quoted startup statistic, and the hardest to trace. There is no single study behind it. Articles that link a source often point to a 2015 Forbes column, not to a dataset. We have also seen it credited to CB Insights, but CB Insights' report on why startups fail doesn't give an overall failure rate at all.

The bigger problem is that "fail" means different things: shut down, never raised money, never became profitable, or sold for less than was invested. Depending on the definition and the type of company, you get very different numbers.

What to do instead: if you want to say that many startups don't make it, say it in words, or cite a specific dataset and say what it measures. "Many startups fail" is honest. "90% of startups fail" with no source isn't.

"42% of startups fail because there's no market need"

This one comes from CB Insights. The version most articles still quote is based on 101 startup post-mortems, and put "no market need" at the top with 42%.

CB Insights has since replaced it. Its current analysis looks at 431 VC-backed companies (385 of them with failure reasons it could identify). The top reasons are now:

A company can have more than one reason, so these add up to more than 100%.

Two things change if you use the new numbers. Running out of money is now the most common reason, not lack of demand. But product-market fit is still a big one, and money often runs out because the product didn't find enough customers. The lesson for MVP builders hasn't changed: check that people want it before you build everything. Our what is an MVP guide shows how.

Watch out: many articles link to the current CB Insights page while still quoting the old 42%. If you link it, quote what it says now.

Source: CB Insights, The top reasons startups fail

"64% of features are rarely or never used"

This statistic is usually credited to the Standish Group, often split as "45% never used, 19% rarely used". Mike Cohn of Mountain Goat Software traced where it came from: Jim Johnson, chairman of the Standish Group, presented it in a keynote at the XP 2002 conference. The results were based on a study of four internal applications.

So it's more than 20 years old, it's based on four apps, and those were internal company tools, not products people chose to buy. That doesn't make the idea wrong. Lots of software does have features nobody uses. But "64% of features are never used" stated as a current fact about all software goes far beyond what the study showed.

A newer number is often quoted next to it: Pendo's 2019 Feature Adoption Report found that around 80% of features in the products it measured were rarely or never used. That's more recent and based on real product usage, but it's also from 2019, and it measured Pendo's own customers' software.

What to do instead: use these as "an often-quoted early finding" with the year and sample size, or skip them. The best evidence is your own: after launch, look at which features people actually use. Planning a short MVP feature list up front means fewer unused features to begin with.

Sources: Mike Cohn, Are 64% of Features Really Rarely or Never Used? and Pendo, The 2019 Feature Adoption Report

Gartner's low-code forecasts

Three Gartner predictions are still quoted everywhere, but all three have now passed their target year:

• "By 2024, low-code application development will be responsible for more than 65% of application development activity." From Gartner's 2019 research.
• "By 2025, 70% of new applications developed by organizations will use low-code or no-code technologies, up from less than 25% in 2020." From a Gartner press release in November 2021.
• "By 2023, there will be at least four times as many active citizen developers at large enterprises as professional developers." A Gartner prediction from 2021.

Writing "Gartner predicts that by 2025…" in 2026 makes an article look out of date, even if the rest of it is current. Forecasts are guesses about the future, so once the year has passed they're no longer evidence of anything.

The same goes for market-size forecasts. Gartner forecast the low-code market at $13.8 billion for 2021 (February 2021), then $26.9 billion for 2023 and $44.5 billion by 2026 (December 2022).

What to cite instead: Forrester's January 2024 analysis is one of the most recent public ones. It put the low-code and digital process automation market at $13.2 billion at the end of 2023. It expects around $30 billion by 2028, and says the market could approach $50 billion if AI speeds things up. It also found that 87% of enterprise developers use low-code platforms for at least some of their work.

Note that analysts measure different things (some include process automation, some don't), so their market sizes don't compare one to one. Say whose number it is and what year it's for.

Source: Forrester, The Low-Code Market Could Approach $50 Billion By 2028

"92% of developers use AI coding tools"

This comes from a GitHub survey published in June 2023. It was run by Wakefield Research and asked 500 US-based developers (not managers) at companies with more than 1,000 employees. It's a real survey, but it's from 2023 and covers only large US companies.

What to cite instead: the Stack Overflow Developer Survey 2025 is much broader. It found:

• 84% of respondents use or plan to use AI tools in their development process, up from 76% the year before.
• 51% of professional developers use AI tools every day.
• More developers distrust the accuracy of AI tools (46%) than trust it (33%), and only 3% "highly trust" the output.

The trust numbers matter as much as the adoption numbers. Almost everyone uses AI now, but most people check its work. That's why planning and testing each feature, as described in how to build an app without coding, still matters when AI writes the code.

Source: GitHub, Survey reveals AI's impact on the developer experience

"Developers are 55% faster with GitHub Copilot"

This comes from a GitHub experiment published in September 2022. GitHub recruited 95 professional developers, split them into two groups at random, and timed how long it took them to write an HTTP server in JavaScript. The group with Copilot took 1 hour and 11 minutes on average. The group without it took 2 hours and 41 minutes. That's the 55%.

It was a fair experiment, but it measured one well-defined task, with an early version of Copilot. You'll now see it quoted as "developers complete complex tasks 55% faster", which it never claimed.

Newer research tells a more mixed story. In July 2025, the research group METR ran a randomized study with 16 experienced open-source developers working on 246 real issues in their own projects. When they were allowed to use AI tools, they took 19% longer to finish.

Neither study settles the question. One is a small, clean task; the other is experienced people in large, familiar codebases. The honest summary for 2026: AI can speed up some work a lot, and slow down other work, and it depends heavily on the task and how you use it. For beginners, the biggest gains come from small, clearly described tasks, which is why we recommend building one feature at a time.

Sources: GitHub, Research: quantifying GitHub Copilot's impact and METR, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity

How to use statistics without getting caught out

1. Click through to the original. If the link goes to another blog, keep going until you reach the study, survey or press release.
2. Write the year next to the number. "(Stack Overflow, 2025)" tells readers how fresh it is.
3. Check the sample. Four apps, 95 developers or 500 US developers can still be useful, but say so.
4. Treat forecasts as forecasts. Once the target year has passed, don't quote the prediction as if it's still coming.
5. Quote what the page says now. Reports get updated. If you link the current page, use its current numbers.

Plan from your own data, not someone else's statistics

The best numbers for your app are the ones you collect yourself: who signs up, which features they use, and which ones they ignore. Flowmaps keeps your whole plan on one map, so when your first users show you what they actually need, you can move those features up and leave the rest for later.

Frequently asked questions

Is it true that 90% of startups fail?

There's no single study behind the 90% figure, and it's often linked to a 2015 Forbes column rather than to data. Many startups do fail, but the exact rate depends on how you define failure and what kind of companies you count. Say what you mean, or cite a specific dataset.

What is the number one reason startups fail?

In CB Insights' current analysis of 431 VC-backed companies, the most common reason is running out of capital (70%), followed by poor product-market fit (43%). The older and still widely quoted version, based on 101 post-mortems, put "no market need" first at 42%.

Is it true that 64% of software features are never used?

The 64% figure (45% never used, 19% rarely used) comes from a 2002 keynote by Jim Johnson of the Standish Group, based on four internal applications. It's often repeated as a general fact, but the original sample was very small. Pendo's 2019 report found around 80% of features rarely or never used in the products it measured.

Did Gartner's prediction that 70% of new apps would use low-code by 2025 come true?

Gartner made that forecast in 2021, and the target year has now passed. It shouldn't be quoted as a prediction any more. For current figures, Forrester's 2024 analysis put the low-code market at $13.2 billion in 2023 and found that 87% of enterprise developers use low-code for at least some of their work.

How many developers use AI coding tools in 2026?

The most recent large survey is the Stack Overflow Developer Survey 2025: 84% of respondents use or plan to use AI tools, and 51% of professional developers use them daily. But 46% distrust the accuracy of AI output, compared with 33% who trust it.

Does AI really make developers 55% faster?

The 55% comes from a 2022 GitHub experiment where 95 developers wrote one HTTP server in JavaScript. A 2025 randomized study by METR found experienced developers working on real issues in their own projects took 19% longer with AI tools. The effect depends a lot on the task.