Stopping AI projects
The case for stopping AI projects
Everyone is measured on launching AI. Almost no one is rewarded for killing it. That asymmetry is why enterprises keep funding bets that will never pay off.
The short answer
Stopping an underperforming AI project is often the highest-return decision a board can make. Every dollar pulled from a bet that won’t pay off is a dollar freed for one that will. Disciplined stopping, based on evidence and pre-set thresholds, is not a failure of AI strategy; it is the strategy working as intended.
- MIT’s NANDA initiative found 95% of enterprise GenAI pilots produce no measurable P&L impact.
- Money left in a losing bet is money unavailable for a winning one, the real cost is opportunity cost.
- Sunk-cost bias and reputational risk keep dead projects alive far past their evidence.
- Pre-committed kill thresholds turn stopping from a political fight into a routine decision.
Why enterprises can’t stop
Stopping is hard for human reasons, not technical ones. Careers are attached to launches. Budgets already spent feel like they must be justified. Nobody wants to be the executive who killed the AI project that a competitor later made work. So dead bets keep breathing, absorbing capital and attention.
The cost is invisible because it’s an opportunity cost. A bet that limps along at break-even doesn’t show up as a loss on any report, but the high-value bet it starved never gets funded. Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027, and the sooner those calls are made, the more capital survives. Across a portfolio, that quiet misallocation is where most AI waste actually happens.
Stopping is a portfolio decision, not a verdict on people
The reframe that unlocks discipline is treating each initiative as a bet in a portfolio. In a portfolio, stopping a bet says nothing about the people who proposed it, it says the evidence didn’t clear the bar this round, and the capital is better deployed elsewhere.
That only works if the bar is set in advance. When you agree the fund / fix / kill thresholds before you start, a kill decision becomes the process doing its job, not an argument you have to win.
How to make stopping routine
Enterprises that stop well share a few habits:
- Set explicit success and kill criteria for each bet before funding it.
- Require behavioural evidence, not status updates, at each stage gate.
- Separate the decision to stop a bet from any judgement of the team.
- Celebrate cheap, fast kills as saved capital, the same way you celebrate launches.
Evidence makes the kill decision easy
The reason our clients can stop projects without drama is that the decision is grounded in data. Pretotyping produces a clear behavioural signal early, people used it or they didn’t, they paid or they didn’t, which removes the ambiguity that lets weak bets survive.
That is what an AI Bets Audit delivers: for each bet, a fund / fix / kill call the board can stand behind. It’s the same discipline that has helped enterprise teams save on the many bets that were never going to pay off, so they could back the few that would.
Sources
- 95% of enterprise generative-AI pilots produce no measurable P&L impact. MIT NANDA, State of AI in Business 2025
- Over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs and unclear value. Gartner, 2025
- More than 80% of AI projects fail, roughly twice the failure rate of non-AI IT projects. RAND Corporation, 2024