Why AI Projects Fail and How to Avoid It

The AI Execution Gap 

Every year, organisations spend millions of dollars on AI projects, hoping to achieve results ranging from improved business workflows to better communication among team members. A July 2025 MIT NANDA report put the figures at $30–40 billion in enterprise investment with a 95% failure rate in GenAI alone, highlighting the stark difference between investment and value generated. This is mostly due to what I will call the Execution Gap; a lot of organisations lack the structure, processes, and feedback mechanisms to follow through, and this ultimately affects final delivery.

Continuous Improvement as a Tool for Successful AI Adoption

Continuous improvement is one way to ensure successful implementation of AI projects. This is because adapting to the people, process, technology, and even culture of the organisation is integral to effective adoption. While this seems like a fairly simple concept in theory, not a lot of companies get it right. To avoid this, leveraging the right framework could be a useful tool to close the execution gap.

One such framework is the Kaiban framework, my adaptation of the popular Agile framework, Kanban, widely popular for the execution of technology projects. While the Kanban framework typically captures project execution in four stages: To Do, Doing, Review, and Done, the Kaiban framework adds an extra stage: Adapt, which aims to continuously refine output, ensuring relevance, efficiency, and sustainability.

How Kaiban Can Be Leveraged for Successful AI Implementation

• To Do: At this stage, the necessary stakeholders are recognised and documented, requirements are painstakingly gathered, an execution strategy is decided, and envisioned results are determined. This is to ensure synergy between the people, process, and technology that will both be contributing to and benefiting from the AI implementation. For example, an organisation hoping to implement an AI tool to improve team communication will do well to recognise pain points plaguing stakeholders, map the root causes, and then document how the tool will help achieve the end goal. This helps lay a strong foundation for the rest of the stages and ensures that the solution actually matches the problem.
• Doing: In the doing stage, the AI tool is designed and developed based on the requirements gathered in the To Do stage. At this stage, the organisation will benefit from thoroughly matching the deliverables to the relevant stakeholder requirements using a traceability matrix or any other relevant mapping tool. This helps determine whether the solutions are actually meeting the expectations, limit scope creep, and ultimately avoid a launch before the efficacy has been determined.
• Review: At this stage, the AI tool is closely examined to ensure efficacy. The review stage will benefit strongly from User Acceptance Testing, ensuring that the stakeholders are actually deriving value from the tool. The stakeholders utilise the tool in the everyday conditions it was designed for, and based on the feedback provided, the tool is either declared to meet the next stage of Done or the project team reverts to the Doing stage.
• Done: At this stage, the project is declared user-ready, and the project team can kickstart the rollout strategy. An adoption plan comes in handy in this stage as it helps stakeholders successfully transition from their old systems to the newly delivered one without losing momentum through timeline communication, training, and support. A change management plan can also be leveraged alongside the adoption plan to communicate the why, reduce resistance, ensure stakeholders’ buy-in, and ultimately help achieve the end goal for which the tool was intended.
• Adapt: During the adapt stage, stakeholder feedback is leveraged to ensure that the tool continuously meets requirements gathered at the To Do stage and performance is being constantly optimised. It is the stage directly responsible for closing the execution gap, not only because it helps ensure stakeholder satisfaction but also because it makes sure evolving stakeholder needs are being met by the tool.

Conclusion

Closing the execution gap is not only a matter of a bigger budget, but it is a matter of thoroughness and follow-through. Although the Kaiban framework alone will not guarantee success, it provides a solid structure and discipline that organisations can utilise to ensure successful implementation that not only meets stakeholders’ needs but also helps avoid AI projects that stop contributing value post-announcement. It helps align people, processes, and technology, driving measurable and sustainable results.