Two recent exciting finds, which I would like to share with you. Both discussing why so many enterprise AI pilots seem to fail so far.
Both articles definitely worth a read!
#1 – “95% of AI pilots are failing — and that might be good news“, a brilliant summary by Rob Verheul of a recent MIT study.
Rob provides a list of top 10 insights from MIT Media Lab’s Project NANDA which overall finds that 95% of corporate generative-AI pilots fail to deliver measurable ROI. Despite >80% experimentation, transformation lags—what the authors call the “GenAI Divide.” Disruption is concentrated in technology and media/telecom, while most sectors (e.g., healthcare) see little structural change. Large enterprises lead in pilots but struggle to scale; in-house builds underperform vendor partnerships; budgets over-index on sales/marketing instead of higher-ROI back-office automation. A core barrier is weak learning/adaptability in enterprise AI systems. Shadow, employee-driven AI use is widespread and often more productive. Real returns come from cost avoidance rather than mass layoffs; workforce impacts are selective. Framed as the “trough of disillusionment” in Gartner’s hype cycle, the moment invites focus on durable value: target process pain points, buy vs. build, demand systems that learn, and mine bottom-up use cases.
You can read the full article at Linkedin.
#2 – “Why 74% of Enterprise CX AI Programs Fail — And How to Make Them Work“, by Ricardo Saltz Gulko. A truly outstanding article providing a brilliant overview of the topic, including a series of real-life cases, pratical recommendations to avoid failure, and a list of valuable references.
Despite rapid adoption, most enterprise AI initiatives in customer experience (CX) are failing to deliver value. While nearly 80% of organizations use AI in CX, only 26% report measurable gains, leaving 74% struggling with stalled pilots, unmet expectations, or outright failures. The root causes are consistent: poor data quality, lack of strategy, cultural resistance, and premature over-automation.
Drawing on real-world cases across fintech, telecom, travel, financial services, publishing, and public-sector organizations, Ricardo contrasts costly failures – such as Klarna’s chatbot layoffs, Air Canada’s legal liability, and NYC’s faulty “MyCity” bot – with successes from NAB, Telstra, IKEA, Zoom, and Wiley, where AI improved personalization, augmented human agents, and streamlined service at scale. The difference lies not in technology but in disciplined execution: clean, unified data, clear CX outcomes, human-AI collaboration, governance, and compliance integration from the outset.
To help leaders avoid common pitfalls, the article presents 20 practical recommendations for AI-enabled CX. These include defining strategic intent early, treating data as infrastructure, embedding humans in the loop, aligning AI to lifecycle touchpoints, building legal and ethical guardrails, and scaling only after proven results.
The key lesson is that AI is neither hype nor a silver bullet. It is a precision tool that must be deployed deliberately, empathetically, and strategically. Organizations that balance innovation with governance, and augmentation with human oversight, will unlock measurable gains in loyalty, efficiency, and customer trust, becoming the future leaders in AI-powered customer experience.
You can read the full article at Linkedin or (if you might not be there) at the Eglobalis blog.