We have multiple solutions running, but they're fragmented
Consolidating through governance, standards, commercial alignment and shared ownership
Fragmentation is normal, and fixable
If your AI initiatives have grown organically and now feel disconnected, you are not alone. This is the natural consequence of the distributed experimentation the AI Playbook encouraged across 2025. The Action Plan: One Year On notes fragmentation across departments and ALBs as one of the biggest remaining barriers — different teams pursuing AI independently, each with their own data, security, assurance and commercial approach. The good news is consolidation creates significant value, and the playbook for doing it is now well understood.
AI Knowledge Hub live on ai.gov.uk. Central repository of practical "how-tos" and reusable patterns from real public sector AI deployments
£100m+ committed to the National Data Library to standardise how public sector data is curated and made AI-ready
(Spending Review 2025)
AI Commercial Strategy published January 2026, with AI Accelerator Tenders and the i.AI scan function speeding up AI procurement across government
Understanding the challenge
Fragmentation looks like a technology problem, but it almost always begins as a people, governance and commercial one.
Different teams have built things in different ways, with different standards, different data sources, different definitions of “responsible AI” and different supplier contracts. The technical debt is the symptom; the absence of shared ownership and standards is the cause. The fix is not a new platform. It is the connective tissue that makes consolidation easier than continued divergence.
Three things that will move this forward this quarter:
1. Establish a working coordination forum chaired at ExCo level. Cross-functional, with enough seniority to make decisions stick. This aligns directly to the Blueprint for Modern Digital Government's call to elevate digital leadership to the centre of public sector decision-making
2. Commission a 60-day fragmentation audit. Every active AI initiative, its data sources, its assurance status, its commercial vehicle, its overlap with others. Output: a consolidation plan, not another inventory.
3. Mandate the AI Playbook’s ten principles as your common standard. Align ATRS reporting, shared component reuse and procurement decisions to it. One standard beats ten local variants.
REACH
At this stage: Architect and Hardwire
Architect
Design the governance, standards and shared services that make integration the default rather than the exception.
Hardwire / Habits
Embed new behaviours into the operating model so they outlast the programme. The cross-functional governance forum becomes a permanent part of decisions; the shared component library becomes part of the SDLC.
What success looks like
A governed digital core
Consolidated, secure data foundations with modular architecture aligned to the National Data Library guidelines for AI-ready public sector datasets.
Strategic governance that actually meets
Cross-functional forums that review roadmaps, hold the line on the ten principles, and make real prioritisation calls. Not a steering board that rubber-stamps.
Reusable building blocks
Shared services, code, prompts, evaluation harnesses and patterns that accelerate every new deployment. Your internal version of the AI Knowledge Hub, and, where capabilities are mature enough, open-sourced and shared across government in line with the AI Opportunities Action Plan commitment to build common infrastructure that the whole public sector can benefit from.
Aligned commercial vehicles
Procurement routes that make consolidation easier than divergence. Use the AI Commercial Strategy, AI Accelerator Tenders and i.AI scan function deliberately.
Integrated risk and assurance
Consistent responsible AI standards across all deployments. ATRS, transparency and clear escalation paths — non-negotiable for central government and ALBs.
Practical actions for your teams
- Establish a working coordination forum. Cross-functional, with enough seniority to make decisions stick.
- Adopt interoperable standards. Modular infrastructure, common data contracts and shared evaluation approaches so solutions scale together. Joined-up data standards are what make it possible to design citizen journeys end to end, rather than stopping and starting at every organisational boundary.
- Build a shared component library. Code, prompts, evaluation harnesses, deployment templates and assurance patterns every team can reuse.
- Stand up a centre of excellence. A small, named team that codifies what works. Enablement, not gatekeeping.
- Align your commercial approach. AI Accelerator Tenders, the i.AI scan function and shared frameworks to avoid duplicate contracts and inconsistent terms.
- Embed the ten principles. Build them into your delivery rhythm. ATRS, transparency, fairness, safety and security as continuous activities.
- Prune what is not working. Distributed experimentation produces winners and a long tail of pilots that will not scale. Discard those honestly. It is learning, not failure.