CIOs: Are you ready to kill your AI darlings?

Applying a disciplined approach to opportunities 

Many AI pilots are valuable. Many are duplicative. Some are potentially risky. The CIO must find a way to channel this energy productively, creating the structures, platforms, and governance mechanisms that allow promising pilots to be identified, assessed, and scaled with repeatability, rather than left to wither in isolation or grow unchecked without oversight.  

How to stop testing and start governing  

Consolidate in the early phases 

• Rather than allowing every unit to run its own experiments in isolation, consolidating AI innovation in the early phases helps surface how much different parts of the organization have in common, and builds a shared foundation faster than a fully distributed model allows. 

Stop pilots that are not delivering 

• Have a clear process for stopping pilots that are not delivering. Establish explicit exit criteria for every pilot at the outset, defining what would need to be true at each stage for the initiative to continue, and what would trigger a decision to stop.  

Provide a portfolio view of initiatives  

• Establish a portfolio view of AI initiatives across the enterprise, applying consistent criteria around strategic fit, feasibility, data readiness, and expected business impact. This will ensure that energy goes into the initiatives most likely to generate lasting value.  

Build a robust value framework 

• Underpinning all of this is the need for a robust value framework. Connecting AI initiatives to tangible business outcomes – whether revenue growth, cost reduction, risk mitigation, or customer experience improvement – will enable you to benefit genuinely from AI, rather than simply accumulate proofs of concept. 

Temper expectations 

• Many AI pilots will fail. Many that succeed technically won’t demonstrate business value at scale. Temper unrealistic expectations, call out governance gaps, and keep the organization honest about what is working and what is not. Building this kind of credibility, over time, is worth considerably more than any individual AI initiative you could deliver.