Why AI still isn’t delivering and how to fix it

By Kirsty Biddiscombe, EMEA Business Lead for AI, ML & Data Analytics, NetApp.

Over the past few years, most organisations’ relationships with AI have moved beyond being curious about the technology to becoming an active area for priority investments. Strategies have been drawn up, tools deployed, and partners engaged. Yet, as generative AI matures, the mood is shifting. Many business leaders are now asking tougher questions: Why isn’t this scaling? Where is the ROI? Is AI heading the same way as blockchain or the metaverse with a big promise, low delivery?

What we’re seeing now is the evolution from hype to execution. The fundamentals of AI’s transformative potential are still sound. However, the engine is not the problem, it’s that many organisations are trying to run it without fuel, a roadmap, or a defined destination. That’s why, despite more innovations like the rise of agentic AI, we’re still seeing the same barriers stalling the AI engine. Namely, inadequate data foundations, governance challenges, issues around talent, and unclear use cases. The good news is that each is addressable, but they need practical attention. 

Rubbish in, rubbish out

Every AI project begins and ends with data. Many businesses still try to build models on top of siloed, inconsistent, or low-quality data, and then expect game-changing results. Imagine handing a chef random ingredients with no labels. You might end up with a meal, but the results will be unpredictable. And it’s certainly not likely to be consistent, or even particularly good. 

AI needs data that is clean, connected, and context-rich. That requires modernising data estates, eliminating duplication, and establishing data access policies that support effective model development. This approach also ensures models are better placed for performance, as well as supporting governance, integration, and lifecycle management. Without strong data lineage and observability, it becomes difficult to understand how outputs are generated and how to improve them over time.

Build in compliance at your core

The thrill of AI’s potential makes it tempting to get innovating and ignore the cookbook. Great chefs will know that the dishes start with a reliable recipe and following it balances both creativity and consistency.  In AI, governance plays the same role. It’s not there to limit businesses, but to provide the starting point that ensures the rest of the process is scalable, safe, and ultimately, successful.

Regulatory pressure around AI is increasing, from GDPR to the EU AI Act, to even the evolving copyright and IP frameworks. The most effective organisations are attempting to get ahead of this and are treating governance as a design principle to bake in from day one. That includes documenting data provenance, managing model updates, applying ethical guardrails, and building auditability into every stage of development and deployment. With these in place, governance is a powerful enabler for scale. It unlocks trust, accountability, and resilience, and drives more effective AI deployments.

Don’t focus on unicorns in the wild

The shortage of machine learning and data science skills remains a challenge. Recruiting experienced AI professionals is difficult and expensive. But in my experience, organisations that rely solely on hiring external expertise often struggle with sustainability and retention.

The solution? Look inward. The best AI teams are not built solely from “unicorn” hires, they combine upskilled internal talent with selectively sourced external expertise. Upskilling initiatives, hands-on labs, and cross-functional groups are creating a new breed of AI-literate professionals who understand both the tech and the business. And importantly, these professionals are more likely to remain with the organisation because they have grown with the mission, and understand the company better than most.

Focus is key

Many AI projects start with ambition and end with ambiguity. Vague aspirations like “enhance customer experience” or “unlock innovation” may sound great on paper but frequently lack clear KPIs and business alignment. The result is therefore predictable. Businesses will see stalled progress, unclear results, and stakeholder fatigue. Instead, the most successful AI programmes all have one thing in common, they start small and stay focused. 

So while the technology is exciting, let its first applications be boring. Use it to automate invoice processing, predict supply chain bottlenecks, or to improve asset utilisation. Narrow, high-impact use cases can create value quickly, and build internal momentum for broader adoption. With results in hand, it becomes much easier to secure further investment and scale responsibly.

From blueprints to bullet trains

The path to scalable, business-ready AI lies in stepping back and aligning ambition with architecture and infrastructure. The organisations that succeed are not necessarily those with the largest AI budgets or the flashiest demos. It will be the ones that build deliberately, with solid data foundations, embedded governance, targeted use cases and teams equipped to evolve alongside the technology.

And, just as a chef needs a kitchen that can handle the lunch and dinner rush, day in and day out, rather than a pop-up for a press night, successful AI initiatives are not built for a single demo or splashy launch. it’s about repeatable excellence, batch after batch. 

By François Haykal, Senior Project Consultant at BCS, the specialist services provider to the...
By David Higgins, Senior Director, Field Technology Office, CyberArk.
By Frank Jones, CEO, IMS Evolve.
Orange Business continues to pursue its digital transformation to deliver simpler, more flexible,...
As Official Partner of the Ferrari Hypercar team, ServiceNow’s AI Platform powers Ferrari’s...
By Asha Palmer, SVP of Compliance Solutions at Skillsoft.
Q&A with Mark Scrivens, FPT Software UK Chief Executive Officer, FPT Corporation.