Only 14% of public-sector bodies have AI-ready data, Aker Systems warns

TL;DR:

• A new paper by Aker Systems warns that UK government, defence and national-security AI programmes risk stalling unless underlying data infrastructure and governance are modernised first.
• The paper finds only 14% of organisations currently have a fully AI-ready data platform, while roughly half cite legacy systems and scalability as major barriers — a structural rather than tactical bottleneck.
• The argument: AI readiness is “an operational capability, not a destination”, and cloud migration alone is not enough to make legacy public-sector estates fit for real-time AI-driven decision-making.

Written by Stephen Dowdeswell, technical pre-sales solution architect at Aker Systems, the paper focuses on environments where data quality and operational resilience carry significantly higher stakes than commercial deployments — central government, defence and national security. “In high-stakes environments, the impact of poor data goes far beyond cost,” Dowdeswell writes. “It affects operational effectiveness and strategic outcomes.”

Why this lands now

The paper’s 14% AI-ready data finding pairs uncomfortably with last week’s Think Digital reporting that 65% of UK government staff are experimenting with AI but only around 30% have integrated it into delivery. The picture forming is one of widespread AI activity sitting on top of inadequate data foundations — exactly the conditions in which AI programmes scale poorly and fail visibly. HMRC’s parallel rollout of Microsoft 365 Copilot to 28,000 staff (with plans to extend to 50,000) is the kind of high-profile programme where data infrastructure will determine outcomes far more than the model choice.

The legacy-systems problem

Most public-sector systems were built for transactional processing, record management or static reporting — not for real-time AI inference. The paper argues organisations need integrated, interoperable data environments capable of supporting secure, scalable and continuously updated pipelines, including stronger governance, standardisation, metadata management and cross-organisational data sharing. Critically, Dowdeswell argues cloud migration alone does not solve this: “siloed data, inconsistent standards and infrastructure that cannot easily support modern AI workloads” persists in cloud environments unless the underlying data architecture changes.

National security implications

The paper flags defence and national-security environments as where data-governance weakness bites hardest. Sensitive information across multiple classifications, systems and operational teams demands “secure-by-design architectures, stronger data lineage and governance models capable of balancing operational agility with security and compliance requirements” — not just centralised storage. For the AI Security Institute and MoD AI deployments, the message is direct: getting Mythos-class models to perform useful work safely requires data governance maturity that most public bodies have not yet built.

Looking forward

Expect this finding to surface in the next round of Cabinet Office and DSIT capability assessments. With Sovereign AI now deploying capital and HMRC at the front of large-scale Copilot rollout, the structural risk Aker Systems describes will move from “interesting consultancy observation” to “named bottleneck in delivery reviews” within the year. For UK SMEs supplying AI tools to government, expect procurement to start asking harder questions about how the tooling handles data lineage, not just model accuracy.