National AI Office: Where should £998,822 actually go?
Ramsey MHK Lawrie Hooper has asked for a written breakdown of the £998,822 approved for a new national artificial intelligence office, directing the question to Enterprise Minister Tim Johnston. The request follows the government's AI strategy launch this year. Chief Minister Alf Cannan has said the office will lift competitiveness and improve public sector literacy with the technology.
That goal is achievable-if the spend is structured. Below is a practical framework departments can use to assess the reply and track delivery.
What a credible breakdown should cover (example allocation ranges)
- People and capability (30-40%) - Core team hiring, secondments from departments, pay for scarce skills (AI product leads, data engineers, policy leads), and expert advisory capacity.
- Governance, policy, and risk (8-12%) - Guidance, assurance processes, DPIAs, model risk reviews, and an approvals pathway for high-impact use cases.
- Infrastructure and tooling (15-25%) - Secure environments, model access, evaluation tools, data pipelines, and monitoring. Prefer shared services to reduce duplicate spend.
- Security and data foundations (8-12%) - Identity and access controls, audit logging, red-teaming, data quality improvements, and retention policies.
- Pilots in priority services (10-15%) - Time-boxed pilots with clear exit criteria in areas like contact centres, case triage, inspections, and policy drafting support.
- Training and change (5-8%) - Role-based training for policy, legal, procurement, analysts, and frontline staff; manager playbooks and usage guidelines.
- Procurement and legal (3-5%) - Supplier due diligence, licensing, IP terms, and audit rights.
- Communications and engagement (2-4%) - Public transparency notes, internal briefings, and user feedback loops.
- Evaluation and reporting (2-4%) - Independent reviews, benefits tracking, and publication of metrics.
- Contingency (2-4%) - Buffer for security findings or scaling successful pilots.
These are planning ranges, not a substitute for the official line-item detail. They give a sense of what "complete" looks like.
Clear KPIs that matter to government
- Number of departments with at least one AI pilot in production, with measured service improvements.
- Time saved in core processes (hours per case, queue time, or decision turnaround).
- Cost avoided through shared platforms vs. duplicate tools across departments.
- Staff capability: percentage of managers and frontline staff trained and using approved tools safely.
- Risk posture: percentage of AI use cases with DPIA, model evaluation, and human oversight in place.
- Transparency: publication of use-case summaries and impact assessments on a fixed cadence.
Governance that prevents rework
Appoint a named SRO and a cross-department board including CIO, DPO, Treasury, HR, and AG's Chambers. Publish a simple RACI for approvals, spend, and risk sign-off. Set a monthly decision cycle so projects don't stall in process.
Procurement and risk guardrails
- Use open standards and clear exit clauses to avoid lock-in. Require vendor audit access.
- Apply DPIAs and human-in-the-loop for high-impact decisions. Keep a record of model versions and prompts used for official decisions.
- Adopt the UK's Algorithmic Transparency Recording Standard for publishable use-case summaries.
- Ensure accessibility compliance and plain-language guidance for any citizen-facing tool.
Immediate actions for departments
- List top three use cases per directorate with cost/time targets and risk level.
- Confirm data owners and sharing agreements before pilot start dates.
- Nominate a business owner for each pilot and agree the "stop/scale" criteria up front.
- Upskill managers and staff on safe, productive use. For structured learning, see AI courses by job role and the latest AI courses.
Questions the written reply should answer
- Full line-item budget with FTE counts, grades, and contractor spend; which costs are one-off vs. ongoing.
- Delivery plan: first 3 pilots, go-live dates, and benefits targets per pilot.
- Tooling: which platforms will be procured centrally, procurement route, and total cost of ownership.
- Risk: model evaluation approach, DPIA process, and escalation path for issues.
- Standards: security baselines, records management, and transparency commitments.
- Reporting: KPI dashboard cadence, public disclosures, and audit arrangements.
- Financial controls: virement rules, carry-forward, and what happens if savings aren't met.
Suggested timeline
- 0-90 days: Stand up the core team, publish governance, select two shared tools, and start three low-risk pilots.
- 90-180 days: Scale successful pilots, publish transparency notes, and deliver the first quantified savings/time reductions.
- 6-12 months: Move from pilots to platforms, expand training to all managers, and embed KPI reporting into quarterly performance.
The funding can move the island from scattered experiments to dependable capability. The value will come from clear ownership, shared platforms, and measurable outcomes-in that order.
Your membership also unlocks:
