The first wave of AI pilots is well underway, and the organizations seeing the strongest returns are those building foundations to last.

Over £78 billion has been invested in AI across the UK, with targeted pilots already delivering on the ground, among them £23 million for EdTech tools in schools and five dedicated AI Growth Zones.

These short automation experiments are delivering real gains, from faster processing times and measurable cost reductions to sharper decision-making, and they are exactly the right place to set out a credible path to production.

The real differentiator is what comes next – treating each win as a springboard to stress-test infrastructure, upskill employees and strengthen data foundations at every stage of that journey.

The difference between a pilot that stalls and one that delivers is ultimately a strategic approach.

The question is whether organizations are prepared to do the groundwork to make it matter. Get that right, and early enthusiasm evolves into long-term ROI.

Building AI into the foundations

Building a house on crumbling foundations doesn’t make the house stronger, it makes it dangerous. The same is true for AI. The organizations seeing the strongest returns are those treating AI as a structural priority, designing their IT infrastructure, people and data foundations to support it from the outset. That means designing not just for the pilot environment, but for the real-world demands of production from day one.

Generative and agentic AI operate on an entirely different logic to legacy business software. Legacy systems were built on a simple premise: structured inputs, structured outputs. Modern AI interprets intent, generates novel outputs and requires continuous refinement. In fact, research has warned that over 40% of agentic AI projects will be abandoned by 2027, because legacy systems cannot support them rather than the technology itself being flawed.

Getting those foundations right from the start is also the smarter commercial decision. Embedding the right architecture, governance and workflows from the outset avoids the expensive, time-consuming process of reworking systems and redeploying tools after the fact. The organizations that will see genuine returns are those willing to rethink their workflows from the ground up, building infrastructure that is AI-ready, not just AI-adjacent.

Small steps, big returns

One of the most common mistakes organizations make is running before they can walk with immediate, large-scale AI deployment.

The appetite is understandable; investment is soaring and the pressure to show results is growing. Research among global executives found that most organizations wait two to four years for satisfactory ROI on a typical AI use case, far beyond the seven-to-twelve-month time frame usually expected from technology investments. Speed without structure, however, is precisely what prevents long term ROI delivery.

Short, focused pilot phases measure whether a tool fits the workflow it is being deployed into, surfacing issues early and building the case for what comes next. Each phase should be treated as a step in a longer journey – generating the insight needed to move forward with confidence, not just proving the technology works.

Research points to workflow redesign as the single biggest driver of measurable impact from generative AI, meaning pilots need to be designed around process fit, not just feature capability.

The organizations that get the most from AI resist the urge to scale prematurely, using each stage to deepen their understanding of what full deployment will require – building confidence across teams as much as testing the technology itself.

The human factor isn’t optional

Even the strongest foundations cannot compensate for poor buy-in. At an executive level, the right questions about operational impact, productivity and real-world outcomes are too often overlooked in favor of how advanced or innovative deployment looks.

Research among global CEOs found that despite pledging to move beyond the piloting phase, 60% remained stuck in the experimenting stage a year later. The gap between intention and execution is rarely technical — it is human.

Below the boardroom, the picture is equally revealing. Almost three quarters (73%) of UK employees have had no AI training, yet two-thirds of UK workers use AI daily at work.

Understanding how to apply it meaningfully to specific business functions is an entirely different skill and one most organizations are not investing in. The result is uneven adoption and a workforce using AI on instinct rather than understanding.

Where training is specific and built around the tools and workflows that matter, adoption becomes a collective process. Where it isn't, AI becomes something people work around rather than with.

Data as a strategic asset

Data is where AI ambitions most commonly come unstuck. Many pilots appear to succeed in controlled environments, only to hit a wall when moved into production where the messiness of real enterprise data surfaces.

Agentic systems coordinate complex, multi-step workflows autonomously, while LLMs handle the heavier cognitive lifting, synthesizing information and interpreting unstructured data at scale. When the data beneath them is fragmented, inconsistent or poorly governed, both simply amplify every flaw.

Treating data as a strategic asset, with clear ownership, embedded governance and architecture designed for AI from the outset, is what separates organizations that scale successfully from those that are stuck in a cycle of relaunching pilots.

Building for AI at scale

The organizations that will define the next era of AI are those channeling the excitement into something built to last – moving deliberately, building sustainable capabilities on solid foundations rather than simply chasing the next wave of experimentation.

The organizations that realize AI’s full potential will be those who treat each pilot not as the destination, but as the first step in a longer, more deliberate journey – one where the real work begins after the experiment ends.

Getting those foundations right across infrastructure, people, data and governance is where the most significant returns are waiting to be unlocked – and the window to do that work is closing faster than many expect.

We tested 70+ best AI tools.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit