Execution Intelligence 5 min readNovember 9, 2026

When to Stop a Failing Project: How Leaders Decide

Set kill criteria before you start, review at fixed gates and let someone other than the sponsor make the call. A practical framework for leaders.

NA

NeuroAgent Editorial

Editorial Team

The short answer

You should stop a project when the evidence that justified it no longer holds, not when the team runs out of energy or the budget runs dry. The most reliable way to get there is to write kill criteria before you start and review them on fixed dates, with someone other than the project's sponsor making the call. It matters more than it sounds: a 2012 McKinsey study of 1,616 US companies (1990–2005) found that the top third of resource reallocators, which shifted an average of 56% of capital across business units over 15 years, earned on average 30% higher total returns to shareholders annually than the bottom third (McKinsey study). You cannot move resources to better bets without stopping something.

When to Stop a Failing Project: How Leaders Decide — a NeuroAgent Execution Intelligence research illustration

Most companies find it far easier to start initiatives than to end them.

In the same McKinsey study, one-third of the businesses in the sample received almost exactly the same capital each year as the year before, with a mean correlation of 0.99. In a company of 50 to 500 people, the same inertia shows up less in capital lines and more in people and leadership attention: the initiative nobody quite believes in keeps its best engineer, its weekly meeting and a slot on the board agenda. The usual reasons are familiar: money already spent, a sponsor's reputation tied to the project, and no forum in which stopping counts as a legitimate result. Two caveats: the data covers listed US companies and is now old, and the study also found that over periods shorter than three years, heavy reallocators delivered lower returns. Stopping is not a goal in itself; it is how you free up resources for something better.

Write the kill criteria before you need them.

Criteria written after disappointing results arrive tend to bend around those results. At launch, record five things on one page: (1) the thesis in one sentence, such as "mid-sized customers will onboard themselves without sales support"; (2) two or three leading indicators with thresholds and dates, such as "20 paying pilot customers by the end of Q2"; (3) a budget and time ceiling; (4) a named stop owner who is not the sponsor; and (5) the four options available at every review: continue, change course, pause or stop. A pre-mortem at launch is a good source for the indicators, because it shows which early signals would tell you the plan is failing.

Review at fixed gates, not when someone gets nervous.

Put review dates in the calendar at launch, for example at 90 days, six months and twelve months, and ask the same four questions each time. If we were not already doing this, would we start it today with what we now know? Which original assumption has been disproven? What would these people and this budget do instead? What does waiting one more quarter cost, and is a pause easier to reverse than a stop? Keep the decision separate from the people: stopping a project is a judgment on a bet, not on the team that ran it. Process matters here more than extra analysis. In a study of 1,048 major decisions published in McKinsey Quarterly in 2010, Dan Lovallo and Olivier Sibony found that process mattered more than analysis by a factor of six (source: McKinsey Quarterly). The outcomes in that study were self-reported by managers, but the direction is clear: a structured gate review beats another spreadsheet.

More on Execution Intelligence — every week.

Join 2,400+ leaders getting the decision intelligence briefing.

Watch for signs that you are rationalising instead of deciding.

Five warning signs come up again and again: the success metric has changed since launch; the deadline has moved twice; every review ends with "just one more feature, hire or quarter"; the sponsor is the only person still arguing for it; and status updates describe effort rather than results. None of these alone proves the project should stop. Two or more mean you should bring the next gate review forward rather than wait for it.

Hypothetical example.

A 180-person German industrial software company launched a self-service product for smaller customers 14 months ago. Its kill criteria said 300 paying accounts and monthly churn below 4% by month 12. It has 110 accounts, churn of 7% and two senior engineers tied up in support. The sponsor proposes a price change and six more months. The stop owner, the CFO, runs the gate review. The core thesis, that small customers would onboard themselves, is contradicted by the support data. The alternative use of the two engineers is a backlog of enterprise integrations with signed customers waiting. The decision: stop new sales, keep existing accounts on a maintained version for 12 months, move the engineers, and record the reasoning and lessons in the decision log. Because the product is paused rather than shut down, the company can revisit the decision if the market changes.

Make stopping a normal outcome.

Announce stopped projects openly, credit the team for what it learned, and review the logged stop decisions once a quarter alongside the projects you continued. If a stop decision affects employees, customer contracts or write-downs, involve your legal and financial advisers; this article is general guidance, not legal, tax or financial advice. When a continue-or-stop call is contested inside the leadership team, you can try the NeuroAgent council at /try to have independent AI advisors challenge both positions, or use a Decision Sprint to turn the question into a decision-ready brief.

Research context

PMI's 2024 research reports an average project success rate of 73.8%, leaving substantial room for better execution systems.

Project Management Institute: Amanda · Social Connector of the Profession 2024 (2024)

This independently published statistic provides context for the topic; it does not substantiate every claim in this essay.

Kill criteriaResource allocationDecision reviews

One decision framework, every week.

The week's best articles on making and executing high-stakes decisions, backed by sourced research.

Send me the weekly Decision Intelligence newsletter (unsubscribe anytime). Privacy Policy

More from the blog

See it in action

Run your next important decision through a NeuroAgent council in under 5 minutes.

Launch a Council