High hopes, low uptake: data and AI's untapped potential in government decision-making

GGF report: Data and AI hold more value for government than most teams are using

Most civil and public servants agree that better use of data would improve outcomes. Yet access, quality, and culture gaps still limit how often data and AI guide decisions across government.

A new Global Government Forum report, Empowering public servants to make confident, informed decisions with data and AI, surveyed 600+ officials worldwide. The findings point to clear benefits, stubborn barriers, and a near-term opportunity to advance beyond experiments to consistent, trusted use.

What officials are seeing

• 80-90% say better use of data would improve service design, team productivity, transparency, accountability, and innovation.
• 38% use data for all or most decisions; 39% use it often but not consistently; 17% use it occasionally or only for specific decisions.
• Only 24% always have the data they need. 41% usually do, 25% sometimes do, and 8% rarely do.
• 81% see some or great potential for AI to improve decision-making.

Top barriers to consistent, data-informed decisions

• Siloed data and poor integration across departments and systems (43%).
• Difficulty accessing relevant data (29%).
• Insufficient investment in tools, systems, or roles (29%).
• Skills gaps in analysis and interpretation (27%).
• Poor data quality (26%).
• Leaders who don't model data-informed decisions (25%).

What "good" looks like

"I feel" and "I think" aren't enough for high-stakes public outcomes. As one public sector advisor put it, officials should be able to say "I know we should do this" - because the right information is available at the right time.

That confidence comes from the basics done well: clean data, clear governance, practical skills, and leadership that sets expectations and removes friction.

Focus areas to unlock value this year

• Data quality at the source: standard definitions, validation rules, and shared reference data across agencies.
• Governance for sharing and trust: clear data ownership, lawful sharing agreements, and audit trails that protect the public.
• Access that works: searchable data catalogs, role-based access, and APIs that connect systems without extra manual effort.
• Skills that stick: practical training for analysts and non-analysts, paired with coaching on real use cases.
• Leaders who model the behavior: set decision standards (what data was used, what assumptions were tested) and ask for the same in briefs.
• Targeted investment: fund roles and tooling where bottlenecks are clearest (integration, quality, or insight).

AI today - and what's next

Current uses include reviewing consultation responses, sorting large volumes of policy and regulatory text, and accelerating access to guidance. These help teams move faster while maintaining accountability.

The next step is agentic AI - systems that can take successive actions across combined tasks with oversight. Awareness is low: 73% report little or no awareness of its potential; only 27% say their organisation uses it, 40% do not, and 33% are unsure.

Practical guardrails for AI adoption

• Start with defined workflows: triage, summarisation, document retrieval, drafting with human review, and case routing.
• Use approved data: separate sensitive data, log prompts and outputs, and retain human accountability for decisions.
• Measure value: track cycle time, error rates, and user satisfaction; scale only where benefits hold up.
• Prepare for agentic tasks: map multi-step processes and define checkpoints where humans must review or approve.

Where this is already working

The report highlights examples and use cases from the UK, Ireland, Australia, Canada, Singapore, the US, and Nigeria - from analysing public feedback at scale to managing complex datasets that blend geology, regulation, and community sentiment, and enabling quick retrieval of policy documents and rules.

Leadership signals that move the needle

• Make data a standing item at the top table, with a named senior owner for outcomes.
• Publish a plain-language data and AI operating model: who does what, how decisions are checked, and how quality is maintained.
• Fund a small, cross-functional "delivery cell" to fix data access and quality issues in the path of priority services.

Voices from government and partners

Contributors include senior civil servants and experts from partners such as SAS (Government). The message is consistent: build confidence through data quality, clear governance, and practical use cases that staff can rely on.

As one UK digital service leader put it, the future is data with a seat at the top table - with chief data and AI officers present in core decisions and accountable for outcomes.

Build the skills to execute

If your team is standing up pilots or scaling AI-assisted workflows, structured training helps shorten the learning curve and avoid common pitfalls.

• Explore role-based options: AI courses by job function for policy, operations, data, and delivery teams.

What to do next

• Identify three priority decisions that need better data - define the inputs, quality rules, and timelines.
• Stand up a shared data catalog and role-based access for those decisions.
• Pilot two AI-assisted workflows with human-in-the-loop review and clear success measures.
• Report progress monthly and scale what works.

About the report

The Global Government Forum report, Empowering public servants to make confident, informed decisions with data and AI, details the survey results, barriers, and examples across multiple countries. It also covers governance, data quality, skills, and how to prepare for agentic AI. Search for the title to read the report in full.