How the UK should take AI from sandboxes to scale

This is a guest blogpost by Rupal Karia, SVP & GM – UKI, Northern Europe & MEA at Celonis.

As we enter London Tech Week, many conversations will be focused on how to make the UK a key market for international AI investment. Yet outside the excitement of the exhibition halls, a practical conversation is taking place across companies and public bodies focused on one question – will AI start to produce meaningful returns?

Boardrooms are realigning their focus from the speculative wonder of large language models to the hard realities of return on AI investment. For the UK to capture an advantage from AI, we must move away from generic automation and toward grounding AI systems in how organisations actually run and how they could and should run.

Moving past theoretical sandboxes to true enterprise scale

The UK’s proposed Growth Bill and the launch of the AI Growth Lab reflect a strong ambition to establish the nation as a leader in applied AI and productivity growth. By moving toward a practical “test-and-build” environment, the UK is seeking to shift the focus away from theoretical frameworks and toward the practicalities of industrial scale. This represents a significant step for attracting international capital, but moving from regulatory testbeds to everyday business use will be the real test.  For AI to succeed, outcomes must be traceable in day-to-day operations, not only demonstrable in controlled pilots.

Often, systems sit on top of deeply fragmented data structures and legacy applications that simply do not communicate. Without end-to-end visibility, tracing an autonomous output back to its operational root cause is nearly impossible. For the UK to become a primary engine for productivity, businesses must connect AI to their underlying operational reality. Visibility is no longer a luxury for any company looking to scale ambitious projects on British soil.

The cost question behind agentic AI

While policy frameworks adapt, the corporate trend of “tokenmaxxing”; the freewheeling, unconstrained consumption of model processing power, has quickly become hard to justify. Goldman Sachs Research indicates that agentic AI is expected to drive a massive 24-fold increase in token consumption by 2030. Yet, the financial sustainability of this volume is already under intense scrutiny from corporate boardrooms.

Executives are now taking active steps to rein in unguided model use. This does not necessarily mean AI budgets will fall. FTSE CEOs and CIOs still face pressure to show progress on AI. But strategic businesses are steering employees away from generic, high-cost prompts and low-yield workflows, reserving heavy token usage exclusively for specialised engineering teams where value is easily quantified and therefore justified.

From personal productivity to business performance

The immediate driver of this fiscal discipline is the widening gap between localised worker efficiency and actual corporate value realisation. Global organisations are discovering that personal productivity hacks do not naturally translate into stronger business outcomes.

One issue is that when an automated agent operates in isolation, it lacks a shared understanding of how the broader business runs. It may summarise an email or generate code faster, but it remains process-blind. It cannot see how a friction point in procurement ripples down into a liquidity bottleneck within finance. This operational blind spot leads AI to make probabilistic guesses that result in wasted tokens and silent retries, ballooning processing costs without moving the needle on performance.

Making AI traceable in the real economy

To close these blind spots, organisations need a clearer view of how work actually moves through the business. In the UK, where sectors such as financial services, healthcare, manufacturing and critical infrastructure operate under close scrutiny, AI systems will need more than access to enterprise data. They will need context: how processes connect, which rules apply, how KPIs are defined, and where decisions have a compliance or customer impact.

Agents cannot be accountable if their actions cannot be traced back to the underlying business process. When real-time process data is combined with agreed organisational knowledge, AI can act with a clearer understanding of cause and effect. It reduces the risk of speculative reasoning and helps control costs – giving CIOs and boards a more reliable basis for assessing whether AI is delivering measurable value.

A positive future direction for British innovation

The same visibility and traceability required to satisfy the UK’s new regulatory testbeds is exactly what delivers financial return on investment. Visibility is no longer a luxury for any company looking to scale ambitious technology projects on British soil.

When AI understands the processes behind everyday work, it can move from isolated experimentation to continuous improvement. People, automations and agents can then be directed towards shared outcomes, rather than optimising separate tasks in separate systems. For the UK, that is where the bigger opportunity lies. By prioritising operational understanding, trustworthy deployment and measurable productivity over raw model consumption, the country can build a more sustainable basis for AI-led growth.

As London Tech Week puts AI ambition in the spotlight, the UK’s real test will come afterwards: whether it can turn pilots into productivity.