Projects rarely fail suddenly. They fail slowly, in a way that is visible early to anyone looking for the right signals, and then become undeniable at the deadline.

Almost none of it is technical.

The real causes

Nobody agreed what "done" means

The most common cause by a distance. The project starts with a shared sense of direction and no written definition of the outcome. Six weeks in, the client and the team have diverged pictures of the same project, and both are confident they were clear.

Prevention: write it down before anyone builds. Specifically what is included, specifically what is not, and what the first releasable version contains. The excluded list matters as much as the included one.

No working software until the end

A project that delivers nothing usable until month six is a six-month bet on everyone having guessed right. By the time anyone sees it, the cost of being wrong is the whole project.

Prevention: working increments every two weeks, deployed somewhere the client can actually use. Not a demo, not a screenshot — something they can log into and try. The feedback this produces in week four is worth more than any amount of upfront specification.

Decisions nobody owns

More projects stall on an unanswered question than on a technical obstacle. Which payment provider? Who approves the copy? What happens to existing records at migration? Each sits unresolved because no single person has authority, and the project quietly waits.

Prevention: name a decision-maker at the start, with the authority to decide without convening a committee. Keep a visible list of open questions with owners and dates. A question that has been open for two weeks is a project risk, not an administrative detail.

Requirements changed and the plan did not

Scope creep is not the problem — requirements changing is normal and often correct. The problem is when scope grows and the timeline, budget and priorities stay fixed, because nobody wanted the awkward conversation.

Prevention: every addition is explicitly traded. "Yes, we can add that; it adds a week, or it replaces this other thing." Written down, agreed, each time. The discipline is social rather than technical.

Building for an imagined user

Software built from assumptions about what users want, with no contact with actual users, frequently delivers exactly what was specified and still fails.

Prevention: get something in front of real users early, even if it is rough. Five people attempting a real task on a prototype will tell you more than a month of internal discussion.

Wrong people in the room

The stakeholder who could have flagged the fatal problem was not consulted until month four. Common with internal tools, where the team who will actually use the thing is consulted last.

Prevention: at the discovery stage, talk to the people who do the work today, not only to the people commissioning the project. They know where the exceptions are.

Automating a broken process

A process nobody understands gets encoded in software, and now the confusion is faster, more expensive and harder to change.

Prevention: if the current process cannot be described clearly by the people doing it, fix that first. It is valuable regardless of whether the software gets built.

The early warning signs

These all appear in the first six weeks, which is what makes them useful.

Nothing is usable yet. The strongest signal available. If week six arrives with no deployed, clickable increment, ask why directly.

Status reports are percentages. "Seventy percent complete" is meaningless and usually means nobody knows. Working increments are a status report; percentages are a substitute for one.

The same item has been "almost done" for three weeks. Something is wrong with it and nobody wants to say so.

Open questions are accumulating. Count them weekly. A growing list means decisions are not being made, and that compounds.

Scope is growing with no corresponding change. New features keep being agreed and the deadline has not moved. One of those two facts is false.

Communication is slowing. Fewer updates, vaguer updates, meetings postponed. Teams go quiet when they are struggling, not when things are fine.

Testing is scheduled for the end. Testing at the end means every problem is discovered at the point it is most expensive.

What a healthy project looks like

For contrast, the signals that things are going well:

  • Something new is deployed and usable every fortnight
  • The open-questions list stays short because decisions get made
  • Scope changes happen, and the timeline visibly changes with them
  • Problems are raised early and plainly, including awkward ones
  • Someone on the team has told you that something you asked for is a bad idea
  • Testing happens continuously rather than as a phase

That sixth point is a particularly good signal. A supplier who never pushes back is either unusually lucky in their clients or managing the relationship rather than the project.

If it is already going wrong

Stopping to reassess is almost always cheaper than continuing to hope.

  1. Establish what actually works today. Not what is "nearly done" — what is deployed and usable.
  2. Write the smallest useful version. What is the least that could ship and deliver value?
  3. Cut everything else into a later phase. Explicitly, in writing.
  4. Resolve the open decisions, with a named owner and a date.
  5. Ship the reduced version, then reassess from a position of having something.

This conversation is uncomfortable and it is far less costly than the alternative, which is discovering the same facts three months later with more money spent.

The underlying point

The things that make projects succeed are mostly unglamorous: a written scope, working software early, decisions with owners, and honest conversations when something changes.

None of that is about technology, which is why choosing a supplier on technical capability alone is a poor filter. How they scope, how often they ship, and whether they will tell you something is a bad idea matter considerably more.

If you have a project that has stalled and want an outside read on it, describe what has happened — we will give you an honest view, including if the answer is that your current supplier is doing fine.