Skip to content
Rivl
14 November 2026Build speed10 min

Scope creep in software projects, and where it starts

Everyone blames the client who keeps asking for one more thing. The public audit record on large software programmes points somewhere less comfortable: at the business case nobody reopened, and at a scope that was never written precisely enough to creep away from.

The story is always told the same way. The project was scoped, the client kept adding, and the timeline went. Said like that, scope creep is something that happens to a team rather than something a team can structure against.

It is worth checking that against the one body of evidence on software delivery that is published rather than marketed. Government audit offices have to report on programmes that went wrong, in detail, with the reasons attributed. The picture there is not a stream of small client requests. It is a scope that was never pinned down, defended for years after it stopped being right.

The sections of this note in order: why the change request is the wrong suspect, what the audit record attributes scope movement to, the fail slow pattern and what it costs, the three places scope actually moves, the change request habit that works, what belongs in the contract instead of a scope freeze, and honest limits.
The second and fourth sections are the ones that change how a project is set up.

The change request is the wrong suspect

A client asking for something new is information. It usually means they have understood the thing you are building well enough to see a gap in it, which is the point of showing work early. A project where nobody ever asks for a change is normally a project nobody is reading.

The damage comes from how the request is absorbed. Taken into the current sprint without a decision, it displaces work that was already committed, and nothing in the record says why the date moved. Six of those and the schedule has drifted by a month with no single cause anyone can point at. The request was not the problem. The absence of a decision was.

What the audit record attributes it to

The US Government Accountability Office profiled 16 of the federal government's most important IT acquisitions, selected from 101 that agencies themselves nominated. In Information Technology: Key Attributes of Essential Federal Mission-Critical Acquisitions, 11 of those 16 had been rebaselined during development, meaning their cost, schedule or performance goals were formally changed.

The reasons agencies gave are the part worth reading. GAO lists delays in defining the cost, schedule and scope; budget cuts and hiring freezes; technical challenges; and changes in development approach. The first item is not scope changing. It is scope never having been defined in the first place, which is a different failure with a different fix.

That distinction matters because the two are treated identically in most project post mortems, and only one of them is preventable by a process. You cannot stop a budget cut. You can stop starting work against a scope that four people would describe differently.

The fail slow pattern, and what it costs

GAO's June 2026 report Weapon Systems Acquisition: Beyond Business as Usual puts a name to the pattern. It describes programmes structured to "fail slow", where time and money go into efforts that stagnate while the technology around them moves. On the Department of Defense's costliest programmes, it reports that the expected time frame to deliver an initial capability now exceeds 12 years.

The contrast GAO draws is the useful bit. It says leading commercial companies iteratively develop business cases in response to users' needs and finish fast, reassess those business cases regularly to catch problems sooner, and increase investment as a product demonstrates progress. Its conclusion is that the acquisition system has not adopted this because its policies do not treat iterative development as a founding principle.

Read that as a statement about scope and it says something specific. The expensive failure is not a scope that moved. It is a scope that was fixed once, early, by people with the least information they would ever have, and then protected from revision for a decade. Scope creep is the small, visible, survivable version of that. A frozen business case is the fatal one.

These are defence programmes at a scale no commercial project reaches, and the budget mechanics are their own world. The structural point travels anyway, because it is about when decisions get made rather than how much money is involved.

The three places scope actually moves

On ordinary commercial projects, scope moves in three distinct ways, and they need three different responses. Treating all of them as "the client keeps changing their mind" is why the problem feels unfixable.

How scope movesWhat it looks likeWhat it needs
Never definedFour people describe the feature differently and all think they agreeAcceptance criteria before estimating, not after
DiscoveredBuilding it reveals a case nobody had considered, often in the dataA named contingency in the estimate, spent visibly
AddedA genuinely new request arrives after work startedA written decision: in, out, or next phase

The middle row is the one that gets misfiled. Discovered scope is not a change of mind by anybody. It is work that was always required and was invisible at estimating time, and the honest way to carry it is a contingency that is declared up front and drawn down where everyone can see it. Charging it as a change request makes the client pay for your estimate being incomplete, and they can usually tell.

This is also where the proposal is doing more work than people realise. A proposal that describes outcomes without acceptance criteria has already created the first row. Structure is most of the defence, and what makes a B2B proposal that wins covers it from the writing side.

The change request habit that works

The mechanism does not need software to track it. It needs four things written down at the moment a request arrives, before anyone has started on it.

  • What is being asked, in the acceptance criteria form you would test against.
  • What it costs in days, estimated by whoever would build it, not by whoever is talking to the client.
  • What it displaces, named specifically. Not "the timeline", but which committed item moves and to when.
  • The decision and who made it, dated. In this phase, out of scope, or scheduled for the next one.

Three of those four are usually present somewhere. The third is almost always missing, and it is the one that does the work. A client told that a request costs four days will often take it. The same client told that it costs four days and pushes the reporting module past their board meeting will often not. You have not refused anything. You have handed them the trade they were going to be handed eventually, early enough to choose.

Keep the overhead honest. If every request needs a form, people stop raising them and start smuggling small things in, which is worse than the original problem. Anything under roughly half a day goes in without ceremony and gets noted; the process is for the ones that displace something.

What goes in the contract instead of a scope freeze

A fixed scope in a contract reads like protection and behaves like a trap. It holds right up to the first discovered case, and then both sides are arguing about whether something was implied rather than deciding what to do about it.

What survives contact is narrower and more boring. Name the acceptance criteria for the things that matter most, name a contingency as a percentage of the estimate with the rule for spending it, name a review point where the plan can legitimately change, and name who on the client side can approve a change request. That last one prevents more drift than any clause about scope, because most uncontrolled additions arrive from someone who was never authorised to ask.

Who carries the risk of all this is a separate decision and worth taking deliberately, which is the subject of fixed price or hourly. The review point is the commercial version of the thing GAO is describing: a scheduled permission to reassess rather than a plan defended to the end.

Honest limits

There is no percentage here. The widely quoted figure that about half of projects experience scope creep comes from a practitioner survey published in 2018, and the organisation that published it no longer reports a comparable number. We have not cited it, because a 2018 self-reported survey of unnamed projects is not evidence about yours. The GAO figures above are narrower and verifiable, and they are about federal programmes rather than commercial software.

Some of this is not scope creep at all. If every project you run ends over, the estimate is the problem rather than the change requests, and no amount of change control fixes an estimating habit. Why software projects take so long covers the other causes, most of which are waiting rather than building.

The three ways scope moves, from the table in this note: scope that was never defined and needs acceptance criteria before estimating, scope that is discovered during the build and needs a declared contingency, and scope that is genuinely added and needs a written decision.
Three different failures. Only the third is a change of mind.

And the cheapest fix is earlier than any of this. Most of the never-defined row disappears if somebody clicks through the thing before it is built, which is the whole argument for a clickable prototype before development. Requirements documents hide disagreement well. A screen somebody can click does not.

If you are scoping something now and want a second read on where it is loose, the places to look first are the features described by outcome rather than behaviour, and anything involving a data import. Those two produce most of the discovered work on most projects, and school management system requirements is a worked example of separating what matters from what merely sounds thorough.

Describe it. We build it.

Seven or twelve days, pay on delivery, a year of maintenance included. Bring the problem, not a spec.

Book a meeting

Read next