HM Land Registry AI tool cuts land charges migration from months to weeks
TL;DR:
- HM Land Registry has built an AI tool that processed the London Borough of Newham’s local land charges records in four weeks with four staff, compared to an estimated three months with 20 staff manually.
- Every batch of AI-processed data passed quality checks first time, eliminating rework cycles that typically add weeks to migration projects.
- The tool, which took over five years to develop, combines OCR, large language models and custom algorithms, and has already been used by two further councils.
The Local Land Charges Programme has a straightforward goal: move millions of historical property records from individual council systems into a single central register. The execution has been anything but straightforward. Manual reviews require large teams, take months per authority, and frequently need rework when data quality checks fail.
HM Land Registry’s Data Science team has built an AI system to speed the process up. In a pilot with the London Borough of Newham, the tool extracted, structured and validated textual land charges data in four weeks using four staff. The manual approach would have required an estimated 20 people working for three months. Every batch passed quality assurance first time.
How it works
The system combines optical character recognition to read scanned documents, large language models to interpret and structure the text, and bespoke algorithms to catch errors such as misread characters. Controls are built in to prevent the models from generating data that does not exist in the source material — a necessary precaution given that the records have legal weight.
Mark Kelso, programme director for local land charges at HM Land Registry, said the tool frees staff to focus on “more complex, high-value work that benefits from human expertise.” Only one team member now handles data processing before the quality assurance team verifies the output.
Expanding beyond the pilot
Dacorum Borough Council and Cotswold District Council have both used the tool since the Newham pilot. The data science team is now working on handling handwritten documents and poorly scanned records, which remain difficult for automated systems.
The tool took over five years to develop, evolving from bespoke code written for individual authorities into a reusable cloud-based system. That timeline is a useful corrective to assumptions about rapid AI deployment — this was patient, incremental engineering.
Looking forward
For local authorities still sitting on paper-based land charges records, the tool offers a practical route to digitisation without the usual staffing burden. It is also one of the clearer examples of AI delivering measurable efficiency gains in UK public services, in a domain where the data quality bar is genuinely high.