GDS says good data, not tools, is the key to AI adoption

TL;DR:

  • A Government Digital Service project with The National Archives found that AI readiness depends on data maturity and governance, not on deploying AI tools.
  • The pilot tested exposing legislation and case law to AI via Model Context Protocol (MCP), with early evidence it improved output quality over training data alone.
  • GDS will launch an updated data maturity service, warning that “good data managed poorly is not AI ready”.

The UK government’s digital arm has a blunt message for public bodies rushing to adopt AI: sort out your data first. A Government Digital Service project with The National Archives, which has moved from discovery into an alpha phase, concluded that successful AI adoption rests as much on organisational capability and governance as on the technology itself.

Readiness is organisational, not technical

Rather than starting with a tool, the project assessed the quality, governance and management of the underlying legal data. It found The National Archives’ datasets already close to AI-ready — not just because the data was high quality, but because strong governance, practices and leadership surrounded it. “Good data managed poorly is not AI ready,” GDS noted. The team also reached a deliberately humble conclusion about scope: government may add more value by developing ways to evaluate and validate AI outputs than by building chatbots that global tech firms, with far deeper resources, are already racing to ship.

As part of the work, GDS, The National Archives, the Department for Business and Trade and the Ministry of Justice tested exposing legal datasets through Model Context Protocol — an emerging open standard for connecting AI systems to external data. A hackathon of around 40 lawyers, engineers and academics found early evidence that MCP meaningfully improved AI output quality compared with relying on a model’s existing training data.

Looking forward

The findings give a practical spine to the government’s broader AI-readiness push, from the £200m drive to get UK business using AI to DSIT’s support for open-source models. GDS says the National Archives work could be a blueprint for other departments, and will relaunch its data maturity service — first introduced in 2023 — to help them assess governance and usability before committing to AI projects. The unglamorous takeaway is the durable one: the bottleneck to public-sector AI is rarely the model, and almost always the data plumbing beneath it.

Share this article