Skip to main content

When to stop an AI pilot: setting kill criteria.

Gartner found organisations with rules for stopping AI pilots were 2.3 times more likely to be top performers. How to set those rules, and how to apply them.

Topic
AI ROI
Product
Novie Control
Format
Guide
Reading time
2 min
Published
5 November 2026
In short

Stop an AI pilot when it misses the result agreed at approval by the review date: cost per task above the limit, value below the baseline, adoption below the target, or a risk that cannot be controlled at a reasonable cost. Set those criteria before the pilot starts. Gartner found organisations with stopping rules were 2.3 times more likely to be among its top performers.

Few organisations have stopping rules.

In Gartner's survey published in October 2026, 31 percent of CIOs had stopping rules for AI pilots and 17 percent had exit criteria for AI in production. Organisations with those rules were 2.3 and 2.6 times more likely to be in its top-performing group. KPMG found that 49 percent of organisations had rephased AI agent deployments when costs outweighed value, which shows stopping is happening, often without a rule written in advance.

TakeawayMost organisations stop AI work only when the bill forces it, and the ones with rules written in advance perform better.

The criteria to set.

CriterionStop if, at the review date
Cost per task.It is above the limit in the business case, and no fix is in sight.
Value against baseline.The measured gain is below the agreed minimum.
Adoption.Fewer people or tasks use it than the case assumed.
Quality.Error or rework rates are above what the work can bear.
Risk.A data, security or compliance risk cannot be controlled at a reasonable cost.

TakeawayFive criteria cover most decisions: cost per task, value, adoption, quality and risk.

How to apply them.

Write the criteria into the business case, with the review date. Measure monthly so the review has numbers. At the review, the owner proposes scale, fix or stop, and the sponsor decides. Record the decision and the reasons. If it stops, retire it properly: revoke access, set its budget to zero and move any work it was doing.

TakeawayCriteria written at approval, a monthly measure and a recorded decision make stopping routine.

How Novie Control does it

You can see what AI costs to build.Can you see what it costs to run?

See how Novie Control helps.

Sources

  1. Gartner, Survey Finds Only 13% of CIOs Say Their Enterprise Has Realized Significant or Transformational Value from AI (19 October 2026)
  2. KPMG International, Global AI Pulse Q2 2026 (June 2026)

Other articles

Frequently asked questions

When should you stop an AI pilot?

When, at the review date agreed at approval, it misses its criteria: cost per task above the limit, value below the agreed minimum, adoption or quality below target, or a risk that cannot be controlled at a reasonable cost.

What are kill criteria for AI projects?

Conditions set before a project starts that end it if they are met at review, usually covering cost per task, value against baseline, adoption, quality and risk.

How many companies have exit criteria for AI?

Gartner found in 2026 that 31 percent of CIOs have stopping rules for AI pilots and 17 percent have exit criteria for AI in production.

Build the intelligenceyour organisation compounds.