aibrevoGet a quote

Question9 min read

Why Do Software Projects Fail?

Quick answer

Software projects usually fail for business reasons, not because the code is hard: the scope is unclear and keeps changing, nobody has the final say, the estimate came before anyone understood the work, too much gets built before real users see it, and progress is reported in words instead of working demos. Integration and data surprises, team changes and a missing handover finish off the rest. Each one shows up early, and each one has a fix.

Reviewed by Alpit Patel, Founder and CEO · Updated November 11, 2026

On this page 9 sections
  1. Why do software projects fail?
  2. How often do software projects fail?
  3. Why do most software projects fail before the build even starts?
  4. Why do software development projects fail even with agile?
  5. Why do AI agents fail in production?
  6. Is your project already slipping? Symptom, cause, fix
  7. What should you ask a developer before you sign?
  8. How do we run builds to avoid these failures?
  9. Sources

Key takeaways

  1. Most software projects go wrong for business reasons: unclear goals, shifting requirements, no single decision maker, estimates made too early and too much built before users see it.
  2. The best-known research is on large projects. McKinsey and Oxford found large IT projects ran 45% over budget on average and delivered 56% less value than predicted.
  3. The often-quoted claim that most projects fail comes largely from the Standish CHAOS reports, whose definitions researchers have challenged. Treat any single failure rate as directional.
  4. Ask for working software you can click at every milestone. A status report saying "90% done" is not progress.
  5. Keep the code, hosting and app store accounts in your own name from day one, so a failing project can still be rescued.

Why do software projects fail?

Most software projects fail for business reasons, not because the code is too hard. These are the causes we see most, and the ones the research backs up:

  1. The scope is unclear, and it keeps changing. Nobody wrote down what the first version must do, so every meeting adds something.
  2. Nobody on the client side has the final say. Three people give the developer three different answers, and work waits or gets redone.
  3. The estimate came before the work was understood. A price and a date were promised on a one-page brief, and reality arrived later.
  4. Too much gets built before real users see it. Months of work go into features nobody has tried.
  5. Progress is reported in words, not working demos. “Going well” and “90% done” replace software you can click.
  6. Integrations and data are harder than they looked. The payment provider, the old database or the partner’s system behaves differently than anyone assumed.
  7. The team changes. The developer who understood the system leaves, and the next one starts from scratch.
  8. There are no tests and no handover. Nothing proves old features still work, and nobody but the original developer knows how to run the thing.

McKinsey and Oxford sorted the causes of overruns on large IT projects into four groups that line up with this list: missing focus (unclear objectives, lack of business focus), content issues (shifting requirements, technical complexity), skill issues (an unaligned team, lack of skills) and execution issues (an unrealistic schedule, reactive planning).

How often do software projects fail?

Often enough that you should plan for it, but no single failure rate is reliable. The careful studies measure how far projects overrun, not a pass or fail grade:

Study What it looked at What it found
McKinsey and the BT Centre for Major Programme Management, University of Oxford (2012) More than 5,400 IT projects; “large” means an initial price above $15 million Large IT projects ran 45% over budget and 7% over time on average, and delivered 56% less value than predicted. 17% went so badly they could threaten the company.
Flyvbjerg and Budzier, Harvard Business Review (Sept 2011) 1,471 IT projects, average cost $167 million Average cost overrun 27%. One in six was a “black swan” with a 200% cost overrun on average and a schedule overrun of almost 70%.
Standish Group CHAOS report (1994), as summarized by Eveleens and Verhoef Standish’s own survey 16% of projects counted as successful. Eveleens and Verhoef tested Standish’s definitions on 5,457 forecasts for 1,211 real projects and called them misleading and one-sided.

Sources: McKinsey, HBR (figures read in the authors’ arXiv version), IEEE Software. All checked Nov 11, 2026.

The headline numbers come from very large projects, mostly well above $15 million, so the percentages do not transfer to a $20,000 app. The causes do. Flyvbjerg and Budzier’s point is that the average hides the danger: most projects overrun a little, and a few overrun so badly they sink the budget. Their example is Levi Strauss, where an SAP move budgeted at about $5 million ran into a Walmart requirement to connect with its supply chain system, the company closed its three US distribution centers for a week, and it took a $192.5 million charge in 2008.

Why do most software projects fail before the build even starts?

A lot of the damage is done before anyone writes code. Three habits cause it.

Pricing a project nobody has scoped. A quote on a one-page brief is a guess. The fix is to write down the users, the workflows, what is out of scope and what “done” means for each feature. Our guide on how to write a software requirements document has a template you can fill in before you talk to anyone.

Building everything at once. Flyvbjerg and Budzier’s advice to managers is to break big projects into pieces of limited size, complexity and duration. For a business app, that means a first version that does one job well and ships to real users. That is what an MVP is for.

No one person who decides. Developers need fast answers. When every question waits for a committee, work stalls or gets built on a guess. Name one person on your side who can say yes or no within a day.

Why do software development projects fail even with agile?

Usually because the team kept the meetings and dropped the working software. The Agile Manifesto’s principles say to “deliver working software frequently, from a couple of weeks to a couple of months,” that “working software is the primary measure of progress,” and that “business people and developers must work together daily throughout the project.” Most agile failures break one of those three. Months of sprints with nothing you can click, or a client who is never available, means the project has no plan, whatever the process is called.

The same principles say to “welcome changing requirements, even late in development.” That works when each change is written down and priced. Without that, welcoming change turns into scope creep with a nicer name.

Why do AI agents fail in production?

On top of the usual reasons, AI features have two of their own: the data underneath is not ready, and the agent is asked to act without enough checks. Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, because of escalating costs, unclear business value or inadequate risk controls. Gartner also said many vendors are “agent washing,” rebranding chatbots and assistants, and estimated only about 130 of the thousands of agentic AI vendors are real. Separately, Gartner predicted in February 2025 that through 2026 organizations would abandon 60% of AI projects unsupported by AI-ready data.

These are forecasts rather than measured results, so treat the percentages as a warning. The practical lesson holds either way: give an AI feature one narrow job with a measurable result, check that the data it needs is clean and reachable, and keep a person approving its actions until it has earned trust.

Is your project already slipping? Symptom, cause, fix

Find your symptom, then ask the question in the last column this week.

What you are seeing Likely cause What to do now
Demo dates keep moving Scope was never fixed, or the work is not split into small releases Ask for the current build on a link you can open. Cut scope to hit the next date rather than moving it again.
“90% done” for weeks The last 10% is integrations, edge cases and testing nobody estimated Ask for a written list of every remaining item, each marked done or not done, and test the done ones yourself.
Invoices rising faster than features Changes are not written down or priced, or hourly billing has no cap Start a change list with a price per change. Agree a cap or a fixed price for the next phase.
You cannot get the code The repository sits in the developer’s account Ask for admin access to the repository today, and check the contract assigns the code to you.
Every change is “a big job” Rushed structure, no tests, or the developer is out of their depth Get an independent review of the code before paying for more features.
Fixed bugs come back No automated tests and no separate test environment Ask for a test environment and a test for each bug fixed, so it stays fixed.
The developer stops answering One person holds everything, with no documentation Secure every login and account first, then read what to do if your developer disappears.

Our judgment from building and reviewing software, organized against the McKinsey and Oxford cause groups above. Not a published study.

The fourth row matters most. If you do not control the code and the accounts, every other fix gets harder. Our page on who owns the code when you hire a developer explains why a signed written assignment is the document that settles it in the US.

What should you ask a developer before you sign?

Ask these eight questions. A developer who answers all of them plainly has already avoided most of the causes above:

  1. What exactly is in the first version, and what is out of scope, in writing?
  2. Who on your team will write the code, and what happens if they leave?
  3. When will I first see working software I can click, and how often after that?
  4. How do you price a change I ask for halfway through?
  5. Which integrations and data imports are included, and have you worked with these systems before?
  6. Will the code be in a repository I own from day one?
  7. How do you test, and will I have a separate test version to try changes on?
  8. What do I get at handover so another developer could take over?

For the longer version, with good and bad answers and what to look for in the contract, see how to choose a software development company.

How do we run builds to avoid these failures?

We start with a free 30-minute call about your users, your workflow and the systems the software has to connect to. Then we write the scope down and send a fixed-price quote in writing, so the price is tied to a defined first version instead of a guess. Our pricing page shows the ranges.

The code is yours, and apps are published under your own Apple and Google developer accounts. At handover you get a recorded video walkthrough and written documentation. aibrevo was founded in San Francisco in 2021 by Alpit Patel, and our in-house 16-person engineering team is led by CTO Shivam.

If someone else built your software and the project has stalled, we can take it over after a review of the code and accounts; our maintenance and support page explains how that review works.

If your project is slipping, book the free 30-minute call. Bring the latest demo or the repository link if you have one, and we will tell you plainly what to secure first and whether it looks fixable. Any deeper review or rescue work is quoted on scope.

Sources

Related questions

Short, direct answers. Ask an engineer if yours is not here.

Ask on a free call

Why do most software projects fail?

Most fail for reasons that have little to do with code. The goals are vague, the requirements keep changing, nobody on the client side has the final say, and the price or deadline was set before anyone understood the work. Then too much gets built before real users try it. McKinsey and Oxford sort the causes of overruns on large IT projects into similar groups: missing focus, shifting content, skill gaps and poor execution.

What percentage of software projects fail?

There is no reliable single number. The Standish Group's 1994 CHAOS report put project success at 16%, but researchers Eveleens and Verhoef showed its definitions were misleading. Better studies measure overruns instead. McKinsey and Oxford found large IT projects ran 45% over budget on average, and Flyvbjerg and Budzier found one in six projects in their sample overran cost by 200% on average.

Why do software projects go over budget?

Usually because the price was fixed before the work was understood, and then the scope grew without anyone repricing it. Integrations, data cleanup and testing are the parts most often underestimated. Every new request should go on a written change list with its cost before work starts, so the budget moves in the open and you can decide what to cut.

When does agile fail?

Agile fails when a team keeps the meetings and drops the working software. The Agile Manifesto says working software is the primary measure of progress and should be delivered frequently. If months pass without a version you can click, or nobody on your side is free to answer questions and make decisions, agile becomes an excuse for having no plan.

Why do AI agents fail in production?

Mostly for the same reasons other software fails, plus two of their own: the data underneath is not ready, and the agent is asked to act without enough checks. Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027, with escalating costs and unclear business value among the reasons. Start with one narrow, measurable job.

How do you know if a software project is failing?

Watch for demo dates that keep moving, a project that has been nearly done for weeks, invoices rising faster than features, a developer who will not give you access to the code, simple changes described as big jobs, fixed bugs coming back, and slow or missing replies. Any one of these is worth a direct conversation. Two or more at once means act now.

Can a failing software project be saved?

Often, yes, if you can get the code and the accounts. First secure access to the repository, hosting, domains, app store accounts and any paid services. Then have someone review the code, tests and deployment before deciding whether to fix, finish or rebuild. Sometimes rebuilding part of it is cheaper than patching, and a good reviewer will tell you so.

Start a project

Thinking about a build?

A free 30-minute call with one of our engineers. You leave with a first-release scope, a realistic timeline and a price range, whether or not we work together.

  • A first-release scope
  • A realistic timeline
  • A written quote, fixed-price or hourly

Written scope and quote after the call. NDA on request. You own the code.

Prefer email?

Send one question about your app, software or AI idea. An engineer replies within one business day.