As local governments experiment with artificial intelligence, they risk building on a weak foundation, according to new research from the Open Data Institute (ODI) and technology company Nortal. Their joint research found that while many councils are piloting AI to predict demand, reduce costs and improve services, most datasets remain unsuitable for the use of algorithms.
The report is Insights from the UK council on standards, readiness and reform to modernize public data for AIis based on interviews with council leaders, technical teams, and program partners, along with analysis of public and operational datasets. We analyze ten local authority cases from Dorset to Leeds. It found that councils that are making tangible progress have a common characteristic: they approach data standards and infrastructure as strategic rather than technical assets.
Professor Elena Simpal, research director at ODI, said: “Councils don’t need perfect data to make progress, but they do need the right kind of data to do their job.” “What works for search may not work for predictive modeling, and what works for predictive modeling is not suitable for generative AI. Our framework moves the discussion from asking whether Congress is ready for AI to asking what end it is ready for and what needs to change to get there.”
“AI-enabled data is becoming the true infrastructure of modern government,” said Priit Liivak, Chief Government Technology Officer at Nortal. “Identifiers, metadata, and versioned pipelines are not side issues; they are what make services auditable and securely extensible.”
Three paths to readiness
The study identifies three complementary forms of response: search, machine learning, and generative AI, each requiring different data conditions.
If you like this content…
- Search readiness Rely on structured, discoverable data with canonical identifiers such as UPRNs and clear metadata.
- Machine learning readiness Relies on reproducible datasets, transparent pedigrees, and documentation of bias.
- Compatibility with generative AI We need context-rich corpora, segmentation, and APIs that allow models to capture and infer across datasets.
The report notes that while many councils are improving discoverability through open data standards, far fewer have achieved the consistent, machine-readable infrastructure needed for predictive and generative systems.
Partial progress pattern
Dorset City Council’s use of structured social care data is highlighted as an example of predictive responsiveness, and the London fire incident dataset shows how consistent coding can enable advanced analysis. However, familiar barriers still exist for most authorities, such as inconsistent identifiers, missing metadata, and limited API access. These shortcomings not only slow innovation, but also make it difficult to audit AI output and retrain models as conditions change.
From pilots to policies
The report’s authors argue that councils should go beyond pilots and invest in data quality, interoperability and governance as a shared infrastructure.
“Councils that now standardize and document their data will find that search, prediction, and generation tools come naturally to them,” Riebak said. “AI success starts with the data architecture, not the application.”
