The next phase of AI adoption requires better questions, not bigger pilots

Three years after the arrival of generative AI, ambition across the public sector remains high. Organisations are exploring AI agents, investing in new capabilities and identifying opportunities to improve services, productivity and decision-making.

Yet there’s still a noticeable gap between aspiration and delivery. Recent Gartner research suggests that only 17 percent of organisations have deployed AI agents today, while more than 60 percent expect to do so within the next two years. At the same time, Gartner predicts that 40 percent of enterprise applications will include task-specific AI agents by the end of 2026. The direction of travel is clear. Operational readiness has simply not kept pace.

Across the market, conversations about AI increasingly centre on five interconnected themes: agentic AI, trusted data, frugal AI, build-versus-buy decisions and people and adoption. Together, they reflect a broader shift in focus from exploring what AI can do to understanding how it can create measurable outcomes.

Experimentation creates evidence

Targeted experimentation helps close that gap because it creates evidence. It provides a safe environment to test assumptions, understand risks and evaluate potential approaches before making significant investment decisions.

This becomes particularly valuable as interest in agentic AI accelerates.  While autonomous systems offer considerable potential, organisations must begin by understanding the outcomes they are trying to achieve before determining the level of AI capability required. Rapid focused experimentation helps determine whether agentic AI is suited to a particular solution or whether simpler approaches can deliver better results with lower cost, lower risk and stronger adoption.

The same principle applies to wider technology decisions. Frugal AI encourages organisations to select models based on outcomes, criticality, cost and environmental impact. Experimentation makes those trade-offs visible early. It also supports build-versus-buy decisions, helping organisations understand where existing platforms can accelerate progress and where tailored solutions could provide greater control, integration or reuse.

When experiments are driven by clear objectives and measurable success criteria, decisions become evidence-based rather than assumption-led.

Connecting innovation, adoption and scale

Experimentation is only the beginning of the journey. Many AI initiatives struggle to gain momentum because innovation, delivery and adoption are treated as separate activities. Ideas are generated in one part of the organisation, technology is developed elsewhere and user adoption becomes an afterthought. The result is often a collection of interesting pilots that never become embedded within day-to-day operations. 

A wider AI ecosystem creates that route. An AI Lab can identify opportunities, test ideas and prove value using real organisational data. Those insights then need to connect to the engineering and delivery capabilities required to industrialise successful solutions, integrate them into existing services and support their continued development.

Trusted data is part of this journey. Effective AI systems need more than accurate information; they need context, provenance and governed access. These foundations must be developed alongside the technology, not added after success has already been proven.

People and adoption are equally important. Frontline teams understand how services work in practice and where technology can create value or introduce challenges. Involving them early improves the quality of experimentation while building confidence, capability and trust. Benefits measurement should begin early so that organisations assess success through outcomes rather than the number of pilots completed.

When governance, delivery disciplines, operating models and skills evolve together, rapid, focused experimentation becomes part of a structured path from idea to scalable service.

Government has additional considerations

For government organisations, experimentation becomes even more important because innovation must be balanced with trust, resilience, accountability and sovereignty.

Public services cannot afford to discover governance, security or data requirements after solutions have been designed. These considerations need to be explored from the outset, alongside questions of value, feasibility and adoption.

Experimentation provides more than technical evidence. It helps organisations understand how AI will operate within their environment, using their data and reflecting their service priorities. Teams can better determine where additional oversight, security or governance is required and how to balance access to innovation with the need for strategic control.

Sovereign capability forms an important part of this discussion. It can provide greater control over data, architecture and deployment while still allowing organisations to benefit from wider innovation, expertise and platforms. Experimentation helps determine which capabilities should be owned internally, where external partners can accelerate progress and how these choices influence long-term flexibility, trust and resilience.

What success looks like

When organisations get this right, innovation becomes easier to govern, evidence guides investment decisions and AI services are shaped around real organisational priorities rather than technology-led assumptions.

For employees, success looks like having more time to focus on judgement, expertise and service delivery rather than repetitive administrative work. AI becomes a trusted tool that supports people rather than another system to manage.

For organisations, it means developing the confidence and capability to move promising ideas into operational services, making experimentation a core part of decision-making and investment prioritisation.

For citizens, it means more responsive, personalised and accessible services that continue to meet expectations around transparency, accountability and trust.

 There’s no single blueprint for AI adoption. Every organisation has different priorities, constraints and definitions of value.

That’s why rapid focused experimentation matters.

It allows organisations to learn within their own environment, using their own data and responding to their own priorities. It helps them understand which capabilities they need to build, which they can access through the wider ecosystem and where greater control will create long-term value.

Ultimately, success isn’t simply about adopting AI. It’s about developing the capability to evaluate new opportunities continuously, integrate them responsibly and evolve as technology changes.

Most importantly, it’s about giving organisations the confidence to own more of their AI: shaping it around their mission, their services and the outcomes they are trying to achieve.