Global AI Projects: A Recurring Pattern of Disappointment

Introduction

The promises surrounding artificial intelligence are immense, but the reality is often disappointing. Various studies show that most AI projects never progress beyond the pilot phase or are even abandoned altogether before completion. What began as a wave of enthusiasm, for many organizations ends in costs, frustration, and missed opportunities.

Recent research by McKinsey, IBM, and BCG confirms this picture: “75% of organizations lack a comprehensive AI roadmap.” In practice, this means many companies never get beyond isolated experiments, without strategic anchoring. The root causes are strikingly consistent: a lack of long-term vision, weak data foundations and governance, and a mismatch in skills and culture to scale AI effectively.

On top of this, the Capgemini Research Institute (2023) reports that “88% of all AI pilots never make it to production”, while the IBM Institute for Business Value (2023) found that only “15% of companies are truly AI-reinvention ready.”

BARC reinforces this view: the core of the problem is not the technology itself, but the way organizations structure their data, processes, and governance. Only one in five companies has a mature, enterprise-wide AI approach; the rest remain stuck in fragmented experiments. And as Donald Farmer put it: “AI may deliver the first draft, but analytics still owns the last mile.”

Real-World Examples

Concrete real-world cases illustrate just how brittle many AI initiatives are. In June 2024, McDonald’s ended its three-year collaboration with IBM on AI for drive-thru ordering. The reason: countless viral videos on social media showed frustrated customers struggling to get the system to understand their orders, a clear example of a pilot that didn’t scale well to production.

Air Canada also came under fire: in February 2024, the airline was ordered to pay damages after its virtual assistant provided incorrect information to a passenger at a critical time. Perhaps the most notorious example is IBM Watson for Oncology, a collaboration with the MD Anderson Cancer Center in Texas. Watson frequently produced incorrect treatment recommendations, in part because it was trained on hypothetical patient data rather than real clinical data. The project cost MD Anderson more than $62 million without producing any meaningful results.

These issues are not limited to global giants; similar patterns can be observed in Belgium as well. Proximus, for instance, experimented with AI-driven chatbots to reduce the pressure on its customer service. While the technology functioned, the real-world impact was disappointing: customers complained that the bots failed to understand context, delivered generic responses, and ultimately redirected them back to human agents. The result was frustration on both sides and additional costs for the company. Proximus eventually scaled back the initiative and shifted toward hybrid models where AI plays only a supportive role — a clear example that without integration into processes and culture, AI cannot deliver lasting value.

These cases demonstrate that the issues with AI pilots are not confined to one industry or region. From fast food and aviation to healthcare, the same patterns emerge again.

Conclusion

These examples confirm the common thread found in the research from McKinsey, Capgemini, S&P Global, and BARC: the problems with implementing AI rarely lie in the technology itself, but almost always in how organizations approach AI. The main obstacles are consistent: absence of a clear strategy or roadmap, weak data foundations that undermine reliability, inadequate governance and regulation with unclear ownership and compliance, a shortage of the right skills and culture to bridge business and AI. And finally, the lack of defined success criteria or KPIs to prove ROI from these pilots.

The result is that AI initiatives remain susceptible to hype, generating frustration and costly failures. The real challenge is therefore not so much to build ever more powerful technology, but rather to establish the strategic, organizational, and cultural foundations that are essential for making AI sustainably valuable.