Stop, pause or redirect an AI project when it misses outcome or feasibility gates set before the pilot, when its updated costs and risks outweigh plausible future value, or when a better alternative can achieve the business goal. Give it one more bounded test only if the team can identify what failed, what it will change, what evidence would count as success, and the test’s budget and deadline.
Set the stop criteria before the pilot
Before funding a pilot, document the business problem, current baseline, target outcome, measurement period, maximum total cost, feasibility assumptions, risk limits and the person authorized to decide whether it continues. That makes the decision about evidence rather than enthusiasm or sunk cost. Gartner recommends realistic value measures, lifecycle cost models and explicit criteria to pursue, scale or stop; PwC likewise recommends setting benchmarks and timelines before moving from pilot to deployment.
As an Amazon Associate I earn from qualifying purchases.
Choose measures tied to the work the project is supposed to improve: for example, cost per completed task, error or rework rate, service quality, cycle time, revenue contribution, or capacity that is actually redeployed. Count of models, prompts, agents or pilot users is activity, not proof of business value. If the claimed benefit is time saved, specify how that time will be used or whether staffing or other costs will change; expected efficiency gains do not automatically become savings. Gartner’s AI investment framework recommends explicit value measures and trade-offs.
Choose: continue, repair, pivot or stop
| Decision | When it fits | What to do next |
|---|---|---|
| Continue or scale | The project meets outcome and safety criteria, its realistic lifecycle economics remain attractive, and an operating team can support it. | Confirm benefits in representative use, validate all-in costs and fund the next scale gate against written benchmarks. Gartner and PwC recommend explicit measures before scaling. |
| Repair in a bounded test | A material assumption failed, but there is a specific corrective action with a plausible route to the target. | Name the cause, change, owner, budget ceiling, deadline and pass/fail evidence. If the test misses its gate, do not extend it without a new, justified decision. |
| Pivot or replace | The business problem still matters, but this model, supplier, scope or AI approach is not the best solution. | Compare a non-AI method, commercial alternative, narrower use case or different implementation on value, feasibility, cost and risk. CSIRO’s project-selection guidance describes cases where representative testing or commercial alternatives changed the decision. |
| Pause or stop | The project misses agreed gates without a credible remedy; remaining costs or harms exceed plausible future value; there is no accountable owner or user pathway; or alternatives dominate. | Stop discretionary spending and plan dependency management, communications, required records and staged decommissioning. The Australian National AI Centre’s guidance covers intervention and termination planning. |
Recalculate the case using remaining costs and value
Do not keep funding a project simply because the company has already spent money on it. That expenditure is sunk; the relevant question is whether the next investment is better than the available alternatives. Re-estimate the costs still ahead: build and data work, integration, inference or vendor charges, human review, monitoring, security, retraining, change management, scaling and eventual retirement. Include likely vendor price changes and the cost of operating the system, not just the pilot budget. Gartner recommends lifecycle cost models that include build, operations, scaling and retirement, and making resource trade-offs explicit.
#1 Best Overall
Compare those forward costs with the value the project can still plausibly deliver, the time to realize it, and the value of the best realistic alternative. A simpler analytics or business-intelligence option may deliver the needed result faster or cheaper; a commercial product may make a custom build no longer worthwhile. Gartner’s framework calls for considering non-AI alternatives, while CSIRO’s guide announcement describes using ROI, NPV and real-options analysis alongside qualitative factors.
Test whether the project works where it will actually be used
A model can perform well in a demo and still fail the business case. Check it on representative data, users, workflows and operating environments. Confirm that data access, integration, security and legal constraints are workable, and that users can adopt the redesigned process with clear training and ownership. Look for workarounds: if staff routinely bypass the tool, a good model score may not translate into an operational result.
CSIRO’s project-selection guide announcement gives a concrete failure mode: a predictive-maintenance system had not been tested on the specific vehicles it was meant to monitor. It also describes a custom tool overtaken by commercial alternatives. Those examples illustrate why technical performance on a convenient test set is not enough; the relevant evidence is performance and cost in the intended setting.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Do not confuse delayed value with no value
Immediate ROI is not always a fair test, particularly when a GenAI benefit is indirect, company-specific or expected to emerge as workflows change. Gartner notes that returns can vary by company, use case, role and workforce, and may take time to materialize. PwC’s Lead-Lag-Exit approach also treats some strategic investments as longer-horizon choices.
If leaders choose a longer horizon, make the extension bounded: identify the specific strategic benefit, leading indicators that can be observed before the final payoff, a review date and the maximum additional exposure. “It may pay off later” is not a decision rule without evidence and an endpoint. Conversely, low near-term ROI by itself does not prove that an initiative should stop if an agreed strategic case remains credible and its risks and costs are controlled.
Signals that warrant intervention
- The business problem is no longer a priority, the accountable sponsor is gone, or the economic case relies on a benefit that cannot be measured or acted upon.
- Repeated gates are missed and the team cannot name a specific, testable correction with a finite cost and deadline.
- Updated lifecycle costs, vendor exposure, data remediation or operating burden exceed the plausible value of the remaining work.
- Results fail on representative data or in the intended environment, or a less expensive commercial or non-AI option now dominates.
- Residual risks remain above the organization’s tolerance, or incidents indicate the need to restrict, pause or decommission the system.
- No operational owner, user-adoption plan or credible path to safe, reliable service exists after the pilot.
These are decision prompts, not a universal formula. Apply the same evidence discipline to proposals to continue and to terminate; executive sponsorship and AI adoption targets are not substitutes for outcomes, feasibility, risk and cost.
Stopping requires an operational plan
Termination can affect users, connected systems and records as well as the project budget. Before shutdown, identify dependencies and service impacts, decide whether an alternative or continuity path is needed, notify affected parties, and determine what data and records must be extracted, returned, preserved or deleted. Assign an accountable decision owner and document intervention points. The Australian National AI Centre recommends assessing shutdown impacts, minimizing disruption, managing data and records, and planning communications and alternatives in its AI ethics guidance. It is guidance, not a legal ruling; applicable obligations depend on jurisdiction and use case.
Free tools Windows power users keep installed
One-click scans. No signup required.
What failure-rate claims can—and cannot—tell you
Gartner forecast in July 2024 that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs or unclear business value. That was a forecast, not an observed 2025 abandonment rate established by the cited release. Gartner also reported a range of $5 million to $20 million for different GenAI business-model-transformation deployment approaches; it is not a general cost estimate for every AI project.
Best Value
In a survey of 822 business leaders conducted from September to November 2023, Gartner reported respondent averages of 15.8% revenue increase, 15.2% cost savings and 22.6% productivity improvement among earlier adopters. Gartner cautioned that outcomes vary by company, use case, role and workforce; these figures are not forecasts for an individual project. CSIRO’s 2025 announcement attributes the statement that “up to 80 per cent” of AI projects fail to guide lead author Dr Stefan Hajkowicz, but the release does not provide enough methodological detail to treat it as a universal, independently verified rate.
Other evidence also needs the right scope. PwC reported that companies making a meaningful AI investment of 1–2% of revenue had a sector-median total shareholder return 21% higher from 2022 to 2025; that is a comparative association, not proof that spending caused the difference or that a particular project should continue. Isin Guler’s 2018 peer-reviewed study found an association between termination capability and performance among venture-capital firms dealing with unsuccessful investments, not corporate AI initiatives. Neither result supplies a numerical stop threshold for an individual company.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →
