How to achieve scalable success and avoid common pitfalls in large AI projects

Amidst the AI noise, there is genuine excitement about the transformational potentials of AI in governments and regulated industries, with new and exciting use cases that promise to streamline operations, improve decision-making and deliver operational efficiency.  

But the reality is that the vast majority of AI projects fall short of their potential. And it’s because in the excitement of AI, and environment is being created where overambitious goals, unclear objectives, and organisational resistance are becoming the norm, resulting in delays, cost overruns, or outright failures. 

The problem isn’t AI itself. It’s the approach to delivering it. Many organisations aim for large-scale implementations without first proving value in a measurable, sustainable way.  

The solution lies in a pragmatic approach to AI: start small, focus on measurable outcomes, and scale thoughtfully. It’s a concept we’ve articulated in our whitepaper. This playbook for a pragmatic approach to AI is based on our experience of enabling AI programmes of work to several public sector, highly regulated environments which are heavily unionised.  

The whitepaper offers a clear, actionable framework for achieving AI success. But, before we look at some of the best practises we list in the whitepaper, it’s probably worth looking at the common reasons why AI projects fail. How many of these resonate with you? 

1. Falling into the trap of ‘all or nothing’

Many organisations have overloaded ambitions, so they fall into the trap of trying to solve too many challenges with one single AI system. This often results in complex, unwieldy projects that stretch resources and fail to deliver meaningful outcomes. 

2. Lack of measurable outcomes

Without clearly defined goals, projects lose focus. AI isn’t any different. Metrics such as Time to Value (TTV), operational efficiencies, or user adoption rates are crucial for tracking success. But they are often overlooked because of the experimental nature of AI programmes of work. 

3. Cultural and organisational resistance

Many people are scared that AI will lead to them being handed a P45 and a wave farewell. Others are cynical about AI’s ability to deliver reliable results, so they sow seeds of mistrust. It means that adoption often faces pushback, particularly in unionised or highly regulated environments, creating a significant barrier to successful implementation. 

Pragmatic AI provides a structured approach to achieving AI success, focusing on real-world outcomes, incremental progress, and scalable solutions. Here are a few principles of our approach: 

Start small with targeted projects 

Rather than attempting to implement sweeping changes, instead begin with focused Proof of Concept (PoC) projects that address specific, high-impact challenges. Then grow as you prove value. We call this use-case-driven AI solutions. NHS Scotland is doing a great job around this. They’ve implemented PoC’s to improve cancer detection rates, initially focused on lung cancer. Having proved its value, the AI later iterating to include breast cancer screening too. Using AI-powered imaging analysis, the system flagged 12% more potential cases for review within six months.  

Prioritise measurable success 

Setting clear metrics is essential for ensuring AI projects deliver value. Metrics such as Time to Value (TTV), employee adoption rates, and process improvements provide tangible proof of impact, helping to build trust and secure stakeholder buy-in. 

Northumberland County Council worked with Amazon Web Services (AWS) to modernise its contact centre. By starting with a small-scale PoC, the council reduced response times and improved reliability within months. The success of this initiative – provable because it was measurable - led to further AI-driven upgrades across departments. 

Integrate AI into existing workflows 

AI should augment, not disrupt, existing systems. So Pragmatic AI ensures tools are seamlessly integrated into current workflows, minimising resistance and reducing complexity. TfL introduced AI to reduce congestion in pollution hotspots by optimising traffic signal timings in a single borough. This low-disruption project achieved a 10% reduction in congestion and improved air quality, creating a replicable model for citywide expansion. 

Scale thoughtfully 

Once a targeted project delivers measurable results, use it as a blueprint for broader AI adoption. This iterative step-by-step scaling reduces risk while ensuring alignment with organisational goals. 

For CIOs and technology leaders, the path to AI success lies in rethinking their approach.  

Big projects don’t have to fail, but they do need to start small. Proof of Concept projects aren’t just experiments; they’re the foundation for sustainable AI maturity. 

The fact is that AI success doesn’t require massive, all-encompassing initiatives. It requires a pragmatic approach that prioritises measurable impact, thoughtful integration, and scalable solutions. Pragmatic AI provides a clear path to achieving these goals, helping organisations turn ambition into reality. The whitepaper, which can be immediately downloaded, provides actionable strategies, real-world case studies, and a roadmap for sustainable AI success.