4 Min Read

Kittipong Jirasukhanont/Alamy

Success has many fathers, but failure is an orphan, the old saying goes. When it comes to a failed AI project, who should be the C-level leader responsible for pulling the kill switch?

Dovi Geretz, CTO at travel services firm SlickTrip, said he usually defines AI failure in terms of scalability, reliability, data quality and whether the AI tool operates securely across the enterprise. "On the other hand, CFOs often view failure through a financial lens -- they look at missed ROI targets, rising costs or unclear economic value." Then there is the CEO, who usually defines failure in strategic terms, such as whether the AI initiative advances business transformation or market differentiation. 

"These various definitions can cause tension, but they also provide a healthy system of checks if they are all aligned," he said.

The CFO usually holds the most influence over killing an AI initiative, since funding ultimately determines survival, said Steeve Lavoie, CTO at AI-driven photonic products Allied Scientific Pro. "A CIO may flag technical gaps and a CEO may question strategy, but when projected returns miss targets for two or three consecutive quarters, finance pulls the plug," he said.

Related:InformationWeek Podcast: Catching hidden errors in AI-powered code

Yet it's not always so clear cut. The decision to kill a failing AI initiative is rarely owned by a single company executive, Geretz said. "Instead, influence on the kill decision shifts, based on why the initiative is failing." For example, if the issue is related to the AI's technical feasibility, data readiness or ability to integrate with core systems, the CIO will typically have the strongest say in the decision, he said. Meanwhile, if costs rise without a clear ROI, the CFO's influence on the decision will increase. 

"Keep in mind, though, that the CEO always has the final authority, especially when the project is tied to a long-term strategy, brand impact or competitive positioning," Geretz said.

Defining failure via checkpoints

Over time, AI projects that began as useful initiatives can drift toward wastefulness, leading to the need for a thorough reassessment, said Greg Fletcher, CTO at analytics platform provider Ocula Technologies. "Before starting an AI initiative, define tangible checkpoints upfront, including internal adoption rates, accuracy thresholds and cost benchmarks, so that the decision to scale, pivot or stop becomes a structured process and not politically fraught."

Align on what success looks like before the project begins, Fletcher advised. "Mismatched expectations are the single biggest source of internal friction delaying AI projects," he said. Leadership should share a common understanding of the AI tool's capabilities and limitations, and agree on what a successful initiative should look like, he added. It's much simpler to determine whether an initiative should be killed when all stakeholders are comparing the same results against the same benchmarks. 

Related:Ask the Experts: The red flags that signal an AI project isn't worth pursuing

"To this end, try to ensure that all key decision makers have the opportunity to meet and pose questions to the AI team that's implementing the project," he recommended. If stakeholders start measuring the AI project against different criteria, it means there's an alignment gap. "Get agreement on shared KPIs early to ensure progress reviews stay focused on evidence, rather than becoming a philosophically-charged standoff."

For many leaders, success is defined by business value and direct ROI, said Ashish Verma, chief data and analytics officer at business advisory firm Deloitte. "Leaders should recognize that even AI failures can be valuable, offering useful data and experience to inform future strategies," he stated. Testing and learning are fundamental to adopting innovative technologies. "Organizations shouldn't let fear of failure prevent them from making ambitious bets on AI where they see opportunities."

Related:Shutterstock CTO's playbook for scaling AI without vendor sprawl

Decision time on AI termination

Geretz said he believes the decision to shut down an AI initiative should be a joint call. "As the CIO, I believe that each AI project should have predefined success metrics, stage gates and kill criteria that are discussed and agreed upon by IT, finance and the business," he said. 

Whenever those criteria can't be met, the CIO should lead the technical assessment, the CFO should assess the financial impact, and the CEO should weigh all of the strategic implications. "Having this shared accountability will help reduce decisions driven by emotions while keeping trust intact between the company leaders," he advised.

The shutdown decision should be shared, with clear success metrics agreed on before launch, Lavoie said. "Defining those metrics upfront prevents internal friction and keeps debates fact-based instead of political."

Fighting C-suite friction

What matters most isn't who makes the final decision on initiatives that aren't meeting expectations, but achieving collaboration, measurement, and alignment with business goals, Verma said. "The best organizations foster close partnerships across functions so that the CFO, CIO, CTO, CEO and CDAO, among other leaders, are communicating about AI projects and making informed decisions." 

About the Author