Artificial intelligence is no longer an experiment in government. Departments are already using AI and automation to triage information, process citizen data and support decision-making. But while adoption is widespread, scaling it safely and effectively remains a complex task.
New research from Government Transformation Magazine, produced in association with Oracle, shows that 85 per cent of central government departments are using AI in some form. Of these, 45 per cent embed AI through software-as-a-service (SaaS) platforms, while 41 per cent develop bespoke tools to meet specific needs. Only 14 per cent remain in an exploratory phase.
This pragmatic mix reflects a shift from pilots to production. SaaS-based platforms offer speed, security and built-in governance, while custom tools serve high-value, context-specific use cases. The challenge now lies not in experimentation but in confident, ethical implementation.
Departments identify three main barriers to scaling AI: data quality and availability (32 per cent), skills shortages (23 per cent) and ethical or privacy concerns (17 per cent). These are not new problems, but they become more pressing as automation extends into live service delivery.
Dr Kate Marks OBE, Deputy Director of Digital Services and Solutions at the Environment Agency, describes how her organisation is embedding AI in a focused, risk-aware way. “We use AI to automate lower-risk flood alerts and triage incoming citizen data. CoPilot integration helps us accelerate adoption safely.”
Adoption is the easy bit. Operationalising it is harder
This kind of operational deployment is increasingly typical across Whitehall. Departments are moving from proof-of-concept to embedding AI within core business processes. The emphasis has shifted from innovation to assurance: ensuring outputs are reliable, interpretable and transparent.
Alexis Castillo-Soto, Group Deputy Director for Digital Missions and Transformation across the Department for Energy Security and Net Zero and the Department for Science, Innovation and Technology, highlights the cultural dimension: “AI is a tool. But we also have to prepare for the backlash. Adoption is the easy bit. Operationalising it is harder.”
That balance between ambition and caution defines the current phase of government AI. The report finds that 68 per cent of respondents rate SaaS platforms equal to or better than bespoke solutions for AI readiness, largely because of their governance, integration and scalability.
Yet the move towards automation cannot succeed through technology alone. It demands cross-functional governance, high-quality data and clear accountability. Departments that treat AI as an integral part of their digital platform strategy, rather than an add-on, are seeing faster progress and fewer setbacks.
Paul Collman, Deputy Director for Enterprise Services at the Cabinet Office, summarises this cautious optimism: “We’re investing in pilots, but only where we trust the data and the risk is manageable. AI should assist, not decide.”
The lessons from early adopters are clear. The most successful departments are those aligning technology with delivery discipline. They treat AI as a capability to be managed, not a shortcut to transformation.
As Liam Walsh, Chief Technology Officer at Defra, points out, “AI is being embedded across multiple parts of our estate to enhance, not replace, human decision-making.”
The practical reality is that AI in government must evolve alongside broader transformation. Its value lies not in automation for its own sake, but in helping teams focus their effort where human judgement matters most.
The research concludes that operationalising AI will depend on three things: good data, capable teams and strong governance. Departments that can align all three will be the ones that move from isolated experiments to whole-organisation value.
Read the full report: Build, Buy or Both? Enabling Whole-Organisation Digital Transformation in Government →